Posts in Business (20 found)
Stratechery Yesterday

2026.11: Winners, Losers, and the Unknown

Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone . Additionally, you have complete control over what we send to you. If you don’t want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings . On that note, here were a few of our favorites this week. This week’s Stratechery video is on Anthropic and Alignment . Integration and AI . One of the most important and longest-running questions about AI has been whether or not models would be commodities; Microsoft once bet on their integration with OpenAI, but in recent years has made the best that the infrastructure they can build around models will matter more than models themselves. However, the most recent evidence — particularly Copilot Cowork — is that the companies who are best able to harness (pun intended) model capabilities are the model makers themselves. If none of that makes sense, Andrew and I do a much more extensive deep dive on these different layers of the evolving AI value chain on this week’s episode of Sharp Tech. — Ben Thompson The Team Test and a Basketball Disgrace. On Greatest of All Talk, we thought the news of the week would be the return of Jayson Tatum for the Boston Celtics, which provided a delightful excuse to take stock of the Celtics, Wemby’s gravity-defying Spurs, Shai’s Thunder, KD’s Rockets and the NBA’s field of title contenders using Ben’s very scientific Capital-T Team Test for contenders. That was a great episode. Unfortunately, on the follow-up Friday, we had to discuss the crime against basketball decency that took place in Miami Tuesday night. Come for the Team Test joy, then, and stay for Erik Spoelstra outrage (and also check out Ben Golliver’s column about the calamity on his new Substack ). — Andrew Sharp The US, China and Iran. The past two weeks in the China policy space have been full of debates over the implications of the war in Iran for China specifically, and the U.S.-China relationship generally. I wrote about all of it on Sharp Text this week , including thoughts on some takes from last year that haven’t aged well, and why, with respect to China, the war in Iran is best understood as the latest in a succession of U.S.-led body blows to Beijing’s global interests. At least over the past 12 months, countering China has been a consideration in almost everything the U.S. has done in the foreign policy space.  — AS MacBook Neo, The (Not-So) Thin MacBook, Apple and Memory — The MacBook Neo was built to be cheap; that it is still good is not only a testament to Apple Silicon, but also the fact that the most important software runs in the cloud. Copilot Cowork, Anthropic’s Integration, Microsoft’s New Bundle — Microsoft is seeking to commoditize its complements, but Anthropic has a point of integration of their own; it’s good enough that Microsoft is making a new bundle on top of it. Oracle Earnings, Oracle’s Cloud Growth, Oracle’s Software Defense — Oracle crushed earnings in a way that not only speaks to the secular AI wave they are riding but also to Oracle’s strong position An Interview with Robert Fishman About the Current State of Hollywood — An interview with MoffettNathanson’s Robert Fishman about the current state of Hollywood, including Netflix, Paramount, YouTube, Disney, and Amazon. Loud and Clear — The War in Iran is not entirely about China, but it’s definitely about China. MacBook Neo Review Designing for the Low End The Wildly Infectious Banana Plague The ‘Raising a Lobster’ Frenzy; Iran and US-China as Trump’s Visit Looms; Two Sessions Takeaways Tatum and the Team Test, The Spurs Continue to Defy Young Team Gravity, Russ Takes Aim at Kings Reporters Spo and Bam and a Basketball Betrayal, An SGA Early Warning System, Kawhi, Luka and The Other MVP Candidate Nerding Out with the Neo, Claude and the Integration Question, The End of Coding Language History

ai AI business Business
0 views
Ed Zitron's Where's Your Ed At Yesterday

Premium: The Hater's Guide To The SaaSpocalypse

Soundtrack: The Dillinger Escape Plan — Black Bubblegum To understand the AI bubble, you need to understand the context in which it sits, and that larger context is the end of the hyper-growth era in software that I call the Rot-Com Bubble .  Generative AI, at first, appeared to be the panacea — a way to create new products for software companies to sell (by connecting their software to model APIs), a way to sell the infrastructure to run it, and a way to create a new crop of startups that could be bought or sold or taken public.  Venture capital hit a wall in 2018 — vintages after that year are, for the most part, are stuck at a TVPI (total value paid in, basically the money you make for each dollar you invested) of 0.8x to 1.2x, meaning that you’re making somewhere between 80 cents to $1.20 for every dollar. Before 2018, Software As A Service (SaaS) companies had had an incredible run of growth, and it appeared basically any industry could have a massive hypergrowth SaaS company, at least in theory. As a result, venture capital and private equity has spent years piling into SaaS companies, because they all had very straightforward growth stories and replicable, reliable, and recurring revenue streams.  Between 2018 and 2022, 30% to 40% of private equity deals (as I’ll talk about later) were in software companies, with firms taking on debt to buy them and then lending them money in the hopes that they’d all become the next Salesforce, even if none of them will. Even VC remains SaaS-obsessed — for example, about 33% of venture funding went into SaaS in Q3 2025, per Carta . The Zero Interest Rate Policy (ZIRP) era drove private equity into fits of SaaS madness, with SaaS PE acquisitions hitting $250bn in 2021 . Too much easy access to debt and too many Business Idiots believing that every single software company would grow in perpetuity led to the accumulation of some of the most-overvalued software companies in history. As the years have gone by, things slowed down, and now private equity is stuck with tens of billions of dollars of zombie SaaS companies that it can’t take public or sell to anybody else, their values decaying far below what they had paid, which is a very big problem when most of these deals were paid in debt.  To make matters worse, 9fin estimates that IT and communications sector companies (mostly software) accounted for 20% to 25% of private credit deals tracked, with 20% of loans issued by public BDCs (like Blue Owl) going to software firms. Things look grim. Per Bain , the software industry’s growth has been on the decline for years, with declining growth and Net Revenue Retention, which is how much you're making from customers and expanding their spend minus what you're losing from customers leaving (or cutting spend): It’s easy to try and blame any of this on AI, because doing so is a far more comfortable story. If you can say “AI is causing the SaaSpocalypse,” you can keep pretending that the software industry’s growth isn’t slowing. That isn’t what’s happening. No, AI is not replacing all software. That is not what is happening. Anybody telling you this is either ignorant or actively incentivized to lie to you.  The lie starts simple: that the barrier to developing software is “lower” now, either “because anybody can write code” or “anybody can write code faster.” As I covered a few weeks ago … From what I can gather, the other idea is that AI can “simply automate” the functions of a traditional software company, and “agents” can replace the entire user experience, with users simply saying “go and do this” and something would happen. Neither of these things are true, of course — nobody bothers to check, and nobody writing about this stuff gives a fuck enough to talk to anybody other than venture capitalists or CEOs of software companies that are desperate to appeal to investors. To be more specific, the CEOs that you hear desperately saying that they’re “modernizing their software stack for AI” are doing so because investors, who also do not know what they are talking about, are freaking out that they’ll get “left behind” because, as I’ve discussed many times, we’re ruled by Business Idiots that don’t use software or do any real work . There are also no real signs that this is actually happening. While I’ll get to the decline of the SaaS industry’s growth cycle, if software were actually replacing anything we’d see direct proof — massive contracts being canceled, giant declines in revenue, and in the case of any public SaaS company, 8K filings that would say that major customers had shifted away business from traditional software.  Midwits with rebar chunks in their gray matter might say that “it’s too early to tell and that the contract cycle has yet to shift,” but, again, we’d already have signs, and you’d know this if you knew anything about software. Go back to drinking Sherwin Williams and leave the analysis to the people who actually know stuff!  We do have one sign though: nobody appears to be able to make much money selling AI, other than Anthropic ( which made $5 billion in its entire existence through March 2026 on $60 billion of funding ) and OpenAI (who I believe made far less than $13 billion based on my own reporting .) In fact, it’s time to round up the latest and greatest in AI revenues. Hold onto your hats folks! Riddle me this, Batman: if AI was so disruptive to all of these software companies, would it not be helping them disrupt themselves? If it were possible to simply magic up your own software replacement with a few prompts to Claude, why aren’t we seeing any of these companies do so? In fact, why do none of them seem to be able to do very much with generative AI at all?   The point I’m making is fairly simple: the whole “AI SaaSpocalypse” story is a cover-up for a much, much larger problem. Reporters and investors who do not seem to be able to read or use software are conflating the slowing growth of SaaS companies with the growth of AI tools, when what they’re actually seeing is the collapse of the tech industry’s favourite business model, one that’s become the favourite chew-toy of the Venture Capital, Private Equity and Private Credit Industries. You see, there are tens of thousands of SaaS companies in everything from car washes to vets to law firms to gyms to gardening companies to architectural firms. Per my Hater’s Guide To Private Equity : You’d eventually either take that company public or, in reality, sell it to a private equity firm . Per Jason Lemkin of SaaStr : The problem is that SaaS valuations were always made with the implicit belief that growth was eternal , just like the rest of the Rot Economy , except SaaS, at least for a while, had mechanisms to juice revenues, and easy access to debt. After all, annual recurring revenues are stable and reliable , and these companies were never gonna stop growing, leading to the creation of recurring revenue lending : To be clear, this isn’t just for leveraged buyout situations, but I’ll get into that later. The point I’m making is that the setup is simple: You see, nobody wants to talk about the actual SaaSpocalypse — the one that’s caused by the misplaced belief that any software company will grow forever.  Generative AI isn’t destroying SaaS. Hubris is.  Alright, let’s do this one more time. SaaS — Software As A Service — is both the driving force and seedy underbelly of the tech industry. It’s a business model that sells itself on a seemingly good deal. Instead of paying upfront for an expensive software license and then again when future updates happen, you pay a “low” monthly fee that allows you to get (in theory) the most up-to-date (in theory) and well-maintained ( in theory ) version of whatever it is you’re using. It also ( in theory ) means that companies need to stay competitive to keep your business, because you’re committing a much smaller amount of money than a company might make from a single license. Over here in the real world , we know the opposite is true. Per The Other Bubble, a piece I wrote in September 2024 : It’s hard to say exactly how large SaaS has become, because SaaS is in basically everything, from whatever repugnant productivity software your boss has insisted you need, to every consumer app now having some sort of “Plus” package that paywalls features that used to be free. Nevertheless, “SaaS” in most cases refers to business software , with the occasional conflation with the nebulous form of “the enterprise,” which really means “any company larger than 500 people.”  McKinsey says it was worth “$3 trillion” in 2022 “after a decade of rapid growth,” Jason Lemkin and IT planning software company Vena say it has revenues somewhere between $300 billion and $400 billion a year. Grand View Research has the global business software and services market at around $584 billion , and the reason I bring that up is that basically all business software is now SaaS, and these companies make an absolute shit ton on charging service fees. “Perpetual licenses” — as in something you pay for once, and use forever — are effectively dead, with a few exceptions such as Microsoft Windows, Microsoft Office, and some of its server and database systems. Adobe killed them in 2014 ( and a few more in 2022 ), Oracle killed them in 2020 , and Broadcom killed them in 2023 , the same year that Citrix stopped supporting those unfortunate to have bought them before they went the way of the dodo in 2019 . To quote myself again, in 2011, Marc Andreessen said that “ software is eating the world.” And he was right, but not in a good way. Andreesen’s argument was that software should eat every business model: Every single company you work with that has any kind of software now demands you subscribe to it, and the ramifications of them doing so are more significant than you’ve ever considered.  That’s because SaaS is — or, at least, was — a far-more-stable business model than selling people something once. Customers are so annoying . When they buy something, they tend to use it until it stops working, and if you made the product well , that might mean they only pay you once.   SaaS fixes this problem by giving them only one option — to pay you a nasty little toll every single month, or ideally once a year, on a contractual basis, in a way that’s difficult to cancel.  Sadly, the success of the business software industry turned everything into SaaS.  Recently, I tried to cancel my membership to Canva, a design platform that sort of works well when you want it to but sometimes makes your browser crash. Doing so required me to go through no less than four different screens, all of which required me to click “cancel” — offers to give me a discount, repeated requests to email support, then a final screen where the cancel button moved to a different place.  This is nakedly evil. If you are somebody high up at Canva, I cannot tell you to go fuck yourself hard enough! This is a scummy way to make business and I would rather carve a meme on my ass than pay you another dollar!  It’s also, sadly, one of the tech industry’s most common (and evil!) tricks .  Everybody got into SaaS because, for a while, SaaS was synonymous with growth. Venture capitalists invested in business with software subscriptions because it was an easy way to say “we’re gonna grow so much ,” with massive sales teams that existed to badger potential customers, or “customer success managers” that operate as internal sales teams to try and get you to start paying for extra features, some of which might also be useful rather than helping somebody hit their sales targets.  The other problem is how software is sold. As discussed in the excellent Brainwash An Executive Today , Nik Suresh broke down the truth behind a lot of SaaS sales — that the target customer is the purchaser at a company, who is often not the end user, meaning that software is often sold in a way that’s entirely divorced from its functionality. This means that growth, especially as things have gotten desperate, has come from a place of conning somebody with money out of it rather than studiously winning a customer’s heart.  And, as I’ve hinted at previously, the only thing that grows forever is cancer. In today’s newsletter I am going to walk you through the contraction — and in many cases collapse — of tech’s favourite business model, caused not by any threat from Large Language Models but the brutality of reality, gravity and entropy. Despite the world being anything but predictable or reliable, the entire SaaS industry has been built on the idea that the good times would never, ever stop rolling. I guess you’re probably wondering why that’s a problem! Well, it’s quite simple (emphasis mine): That’s right folks, 40% of PE deals between 2018 and 2022 were for software companies, the very same time venture capital fund returns got worse. Venture and private equity has piled into an industry it believed was taking off just as it started to slow down. The AI bubble is just part of the wider collapse of the software industry’s growth cycle. This is The Hater’s Guide To The SaaSpocalypse, or “Software As An Albatross.”  In its Q4 2025 earnings, IBM said its total “generative AI book of business since 2023” hit $12.5 billion — of which 80% came from its consultancy services, which consists mostly of selling AI other people’s AI models to other businesses. It then promptly said it would no longer report this as a separate metric going forward .  To be clear, this company made $67.5 billion in 2025, $62.8 billion in 2024, $61.9 billion in 2023 and $60.5 billion in 2022. Based on those numbers, it’s hard to argue that AI is having much of an impact at all, and if it were, it would remain broken out. Scummy consumer-abuser Adobe tries to scam investors and the media alike by referring to “AI-influenced” revenue — referring to literally any product with a kind of AI-plugin you can pay for (or have to pay for as part of a subscription) — and “AI-first” revenue, which refers to actual AI products like Adobe Firefly. It’s unclear how much these things actually make. According to Adobe’s Q3 FY2025 earnings , “AI-influenced” ARR was “surpassing” $5 billion (so $1.248 billion in a quarter, though Adobe does not actually break this out in its earnings report), and “AI-first” ARR was “already exceeding [its] $250 million year-end target,” which is a really nice way of saying “we maybe made about $60 million a quarter for a product that we won’t shut the fuck up about.”  For some context, Adobe made $5.99 billion in that quarter, which makes this (assuming AI-first revenue was consistent) roughly 1% of its revenue .   Adobe then didn’t report its AI-first revenue again until Q1 FY2026, when it revealed it had “more than tripled year over year” without disclosing the actual amount, likely because a year ago its AI-first revenue was $125 million ARR , but this number also included “add-on innovations.” In any case, $375 million ARR works out to $31.25 million a month, or (even though it wasn’t necessarily this high for the entire quarter) $93.75 million a quarter, or roughly 1.465% of its $6.40 billion in quarterly revenue in Q1 FY2026. Bulbous Software-As-An-Encumberance Juggernaut Salesforce revealed in its latest earnings that its Agentforce and Data 360 (which is not an AI product, just the data resources required to use its services) platforms “exceeded” $2.9 billion… but that $1.1 billion of that ARR came from its acquisition of Informatica Cloud , (which is not a fucking AI product by the way!). Agentforce ARR ended up being a measly $800 million, or $66 million a month for a company that makes $11.2 billion a year. It isn’t clear whether what period of time this ARR refers to.  Microsoft, Google and Amazon do not break out their AI revenues. Box — whose CEO Aaron Levie appears to spend most of his life tweeting vague things about AI agents — does not break out AI revenue .  Shopify, the company that mandates you prove that AI can’t do a job before asking for resources , does not break out AI revenue. ServiceNow, whose CEO said back in 2022 told his executives that “everything they do [was now] AI, AI, AI, AI, AI,”   said in its Q4 2025 earnings that its AI-powered “ Now Assist” had doubled its net new Annual Contract Value had doubled year-over-year ,” but declined to say how much that was after saying in mid-2025 it wanted a billion dollars in revenue from AI in 2026 .  Apparently it told analysts that it had hit $600 million in ACV in March ( per The Information )...in the fourth quarter of 2025, which suggests that this is not actually $600 million of revenues quite yet, nor do we know what that revenue costs.  What we do know is that ServiceNow had $3.46 billion in 2025, and its net income has been effectively flat for multiple quarters , and basically identical since 2023. Intuit, a company that vibrates with evil, had the temerity to show pride that it had generated " almost $90 million in AI efficiencies in the first half of 2025 ,” a weird thing to say considering this was a statement from March 2026. Anyway, back in November 2025 it agreed to pay over $100 million for model access to integrate ChatGPT . Great stuff everyone.  Workday, a company that makes about $2.5 billion a quarter in revenue, said it “generated over $100 million in new ACV from emerging AI products, [and that] overall ARR from these solutions was over $400 million.” $400 million ARR is $33 million.  Atlassian, which just laid off 10% of its workforce to “ self-fund further investment in AI ,” does not break out its AI revenues. Tens of thousands of SaaS companies were created in the last 20 years. These companies, for a while, had what seemed to be near-perpetual growth. This led to many, many private equity buyouts of SaaS companies, pumping them full of debt based on their existing recurring revenue and the assumption that they would never, ever stop growing. I will get into this later. It’s very bad. When growth slowed, the reaction was for these companies to raise venture debt — loans based on their revenue — and per Founderpath , 14 of the largest Business Development companies loaned $18 billion across 1000 companies in 2024 alone, with an average loan size of $13 million. This includes name brand companies like Cornerstone OnDemand and Dropbox, the latter of which took on a $34.4 million debt facility with an 11% interest rate. One has to wonder why a company that had $643 million in revenue in Q4 2024 needed that debt.

business Business
1 views
Justin Duke 2 days ago

Archiving the roadmap

Pour one out for Buttondown's transparent roadmap , which I formally archived yesterday evening after a year or so of informal archival. This felt like the journey that so many other companies have had who have tried to keep public roadmaps and then for one reason or another got rid of theirs. Mine had nothing to do with transparency. It was entirely due to the fact that Linear now makes a much better product than GitHub does — at least for the kind of project management I need — and if there was a way to easily make our Linear publicly visible, I would be happy to do so. The third-party services and integrations which purport to offer such functionality ( Productlane being the most notable) seem like more trouble and money than they're worth. More than anything, the reason I dithered about this for so long was a false sense of worry that there would be a backlash. Around 100 or so folks have commented, watched, or reacted to various issues over the years, which is not a huge amount but not a small one either, and it felt faintly bad to leave them all in the cold. But in reality, no one has minded or noticed that much. And whatever negative goodwill we generate from no longer having this public repository is offset by the negative goodwill we avoid from having that public repository look so obviously abandoned.

business Business open-source Open Source web-development Web Development
0 views
Stratechery 2 days ago

An Interview with Robert Fishman About the Current State of Hollywood

An interview with MoffettNathanson's Robert Fishman about the current state of Hollywood, including Netflix, Paramount, YouTube, Disney, and Amazon.

business Business media Media entertainment Entertainment
0 views
Krebs on Security 3 days ago

Iran-Backed Hackers Claim Wiper Attack on Medtech Firm Stryker

A hacktivist group with links to Iran’s intelligence agencies is claiming responsibility for a data-wiping attack against Stryker , a global medical technology company based in Michigan. News reports out of Ireland, Stryker’s largest hub outside of the United States, said the company sent home more than 5,000 workers there today. Meanwhile, a voicemail message at Stryker’s main U.S. headquarters says the company is currently experiencing a building emergency. Based in Kalamazoo, Michigan, Stryker [NYSE:SYK] is a medical and surgical equipment maker that reported $25 billion in global sales last year. In a lengthy statement posted to Telegram, an Iranian hacktivist group known as Handala (a.k.a. Handala Hack Team) claimed that Stryker’s offices in 79 countries have been forced to shut down after the group erased data from more than 200,000 systems, servers and mobile devices. A manifesto posted by the Iran-backed hacktivist group Handala, claiming a mass data-wiping attack against medical technology maker Stryker. “All the acquired data is now in the hands of the free people of the world, ready to be used for the true advancement of humanity and the exposure of injustice and corruption,” a portion of the Handala statement reads. The group said the wiper attack was in retaliation for a Feb. 28 missile strike that hit an Iranian school and killed at least 175 people, most of them children. The New York Times reports today that an ongoing military investigation has determined the United States is responsible for the deadly Tomahawk missile strike. Handala was one of several Iran-linked hacker groups recently profiled by Palo Alto Networks , which links it to Iran’s Ministry of Intelligence and Security (MOIS). Palo Alto says Handala surfaced in late 2023 and is assessed as one of several online personas maintained by Void Manticore , a MOIS-affiliated actor. Stryker’s website says the company has 56,000 employees in 61 countries. A phone call placed Wednesday morning to the media line at Stryker’s Michigan headquarters sent this author to a voicemail message that stated, “We are currently experiencing a building emergency. Please try your call again later.” A report Wednesday morning from the Irish Examiner said Stryker staff are now communicating via WhatsApp for any updates on when they can return to work. The story quoted an unnamed employee saying anything connected to the network is down, and that “anyone with Microsoft Outlook on their personal phones had their devices wiped.” “Multiple sources have said that systems in the Cork headquarters have been ‘shut down’ and that Stryker devices held by employees have been wiped out,” the Examiner reported. “The login pages coming up on these devices have been defaced with the Handala logo.” Wiper attacks usually involve malicious software designed to overwrite any existing data on infected devices. But a trusted source with knowledge of the attack who spoke on condition of anonymity told KrebsOnSecurity the perpetrators in this case appear to have used a Microsoft service called Microsoft Intune to issue a ‘remote wipe’ command against all connected devices. Intune is a cloud-based solution built for IT teams to enforce security and data compliance policies, and it provides a single, web-based administrative console to monitor and control devices regardless of location. The Intune connection is supported by this Reddit discussion on the Stryker outage, where several users who claimed to be Stryker employees said they were told to uninstall Intune urgently. Palo Alto says Handala’s hack-and-leak activity is primarily focused on Israel, with occasional targeting outside that scope when it serves a specific agenda. The security firm said Handala also has taken credit for recent attacks against fuel systems in Jordan and an Israeli energy exploration company. “Recent observed activities are opportunistic and ‘quick and dirty,’ with a noticeable focus on supply-chain footholds (e.g., IT/service providers) to reach downstream victims, followed by ‘proof’ posts to amplify credibility and intimidate targets,” Palo Alto researchers wrote. The Handala manifesto posted to Telegram referred to Stryker as a “Zionist-rooted corporation,” which may be a reference to the company’s 2019 acquisition of the Israeli company OrthoSpace. Stryker is a major supplier of medical devices, and the ongoing attack is already affecting healthcare providers. One healthcare professional at a major university medical system in the United States told KrebsOnSecurity they are currently unable to order surgical supplies that they normally source through Stryker. “This is a real-world supply chain attack,” the expert said, who asked to remain anonymous because they were not authorized to speak to the press. “Pretty much every hospital in the U.S. that performs surgeries uses their supplies.” John Riggi , national advisor for the American Hospital Association (AHA), said the AHA is not aware of any supply-chain disruptions as of yet. “We are aware of reports of the cyber attack against Stryker and are actively exchanging information with the hospital field and the federal government to understand the nature of the threat and assess any impact to hospital operations,” Riggi said in an email. “As of this time, we are not aware of any direct impacts or disruptions to U.S. hospitals as a result of this attack. That may change as hospitals evaluate services, technology and supply chain related to Stryker and if the duration of the attack extends.” This is a developing story. Updates will be noted with a timestamp. Update, 2:54 p.m. ET: Added comment from Riggi and perspectives on this attack’s potential to turn into a supply-chain problem for the healthcare system.

security Security business Business
0 views
ava's blog 3 days ago

'human oversight' is a meaningless buzzword

When talking about using AI for decision-making, you often hear that there will be " human oversight " or " human intervention ". One popular example that I have come across in conferences and webinars about data protection law is the hiring process and recruiting: Companies are already proudly using AI to select applicants. It summarizes CVs, compares qualifications with the job profile, and ranks candidates. At the end, HR decides who to invite for interviews based on this output. The fact that AI isn't just sending out the interviews itself immediately and instead, a human is required to write an email or press a button is the idolized "human oversight". The fact that someone could intervene and make a different decision is supposed to be enough. What bothers me is that despite being ranked as "high risk" under the AI Act (together with using AI for medical diagnosis, financial and legal advice, etc.), we aren't looking at how these systems are realistically used in practice. We shove a human in the loop ("HITL") somewhere to assuage fears and comply with legal requirements, but almost no one wants to talk about the fact that Think about it: You have an IT company that gets 400-600 applications on each open spot. Spending time on every single application weeding people out takes a lot of time. You want to save time using AI so the people whose CVs and motivational letters most closely match the job description are already pre-selected for you and ranked. You know the next few weeks will bring new application deadlines again and you're already behind. You just can't check all of the applications to see whether the AI messed up or not. You can do a random check here and there, but at what point will you just look at the top candidates, check their applications, see it was correctly summarized (or well enough), and assume the rest of applicants that weren't considered were assessed correctly as well? Why would you look at all or most of the applications again anyway when the AI system is advertised as saving you that time and step entirely? If anything, the human intervention here is for the companies - making sure that the AI didn't accidentally rank someone top that is completely unfitting for the task. It's not there for you . No one will notice if your perfectly fitting application has been disregarded by AI for no discernible reason, and no one will find it as part of the oversight process in the hundreds of other applications to make sure. If the AI makes the task quicker and the first top candidates sound fitting and plausible, that's it, nail in the coffin, why would HR put in more work? All you can realistically do is make them explain and check after each rejection where you were a good fit and know AI was used. If you don't do that, you can't know whether you've been unjustly treated by their AI hiring process or were rejected on a justifiable basis. As long as AI continues to hallucinate or leave things out inexplicably just to say sorry afterwards, this is a huge liability. Companies don't seem to really care for possible poor data quality, biases and systemic inequities that are subtle or deeply embedded, requiring more work and possibly an outside view to detect and mitigate. We are lacking nuanced oversight mechanisms, and I hope companies are prepared for the lawsuits this will generate. If a company wants to use AI in the hiring process, I'd at least expect them to do the following bare minimum: Unfortunately, companies have no incentive to do this! This is seen as more bureaucracy, more time and money wasted, restrictive to innovation. They're competing with companies who are grabbing talent even faster than them who don't give a shit about fairness in AI hiring. Each day they don't find a replacement or candidate for a new role is bad. And why hire more HR personnel to sift through hundreds of applicants if less HR personnel can handle it with AI? Organizational priorities and financial pressures don't allow enough checks and considerations to go into this delicate process. We need to question " human oversight " more closely and require more explanations on how they plan to combat opaque decision-making, automation bias and the pressure to optimize and make work as easy as possible. Until adequate systems are in place that combat this, it will always be ineffective and a buzzword to me. Reply via email Published 11 Mar, 2026 while HR does receive training on how to use AI and how it works, the reasoning behind AI selection and summaries is a black box for the users, AI recruiting is advertised as a huge time save, which stands in contrast to the checking you should technically be doing as a human to make sure the AI did a good job, most users will follow the AI recommendations blindly because they are presented in a way that sounds plausible and as time goes on, we get lazy and suffer from automation bias and oversight fatigue. having a clear documentation of AI capabilities and limitations for their employees incentivizing taking the time to question AI suggestions and do some 'manual' labor requiring detailed justification when accepting the AI suggestions/rankings the ability to explicitly name why the disregarded applications were denied by their AI system in each case (you're going to need this anyway when an applicant challenges the decision) testing the system and the employees by periodically entering a candidate application that should fit perfectly vs. one that is very unfitting, and see where they land and what HR does with them (similar to the existing practice of IT sending out fake phishing e-mails sometimes to test you) collecting decision patterns and errors to correct and adjust the AI system

career Career business Business ai AI
0 views
Ed Zitron's Where's Your Ed At 4 days ago

The Beginning Of History

Hi! If you like this piece and want to support my work, please subscribe to my premium newsletter. It’s $70 a year, or $7 a month, and in return you get a weekly newsletter that’s usually anywhere from 5000 to 185,000 words, including vast, extremely detailed analyses of NVIDIA , Anthropic and OpenAI’s finances , and the AI bubble writ large .  I just put out a massive Hater’s Guide To Private Equity and one about both Oracle and Microsoft in the last month. I am regularly several steps ahead in my coverage, and you get an absolute ton of value, several books’ worth of content a year in fact!. In the bottom right hand corner of your screen you’ll see a red circle — click that and select either monthly or annual.  Next year I expect to expand to other areas too. It’ll be great. You’re gonna love it.  Before we go any further: no, this is not going to turn into a geopolitics blog. That being said, it’s important to understand the effect of the war in Iran on everything I’ve been discussing. So, let’s start simple. Open Google Maps. Scroll to the Middle East. Look at the bit of water separating the Gulf Arab countries from Iran. That’s the Persian Gulf.  Scroll down a bit. Do you see the narrow channel between the United Arab Emirates and Iran? That’s the Strait of Hormuz. At its narrowest point, it measures 24 miles across. Around 20% of the world’s oil and a similar percentage of the world’s liquified natural gas (LNG) flows through it each year.  Yes, that natural gas, the natural gas being used to power data centers like OpenAI and Oracle’s “Stargate” Abilene (which I’ll get to in a bit) and Musk’s Colossus data center . But really, size is misleading. Oil and gas tankers are massive, and they’re full to the brim with incredibly toxic material. Spills are, obviously, bad . Also, because of their size, these tankers need to stick where to where the water is a specific depth, lest they find themselves stuck.  As a result, there are two lanes that tankers use when navigating through the Strait of Hormuz — one going on, one going out. This a sensible idea with the goal to reduce the risk of collisions, but it also means that the potential chokepoint is even smaller.   Anyway, at the end of last month, Iran’s Revolutionary Guard Corps unilaterally closed off the strait, warning merchant shipping that any attempt to travel through the strait was “not allowed .” This closure, for what it’s worth, is not legally binding. Iran can’t unilaterally close a stretch of international waters. And yes, while some of those shipping lanes cross through Iran’s territorial waters ( and Oman’s, for that matter ), they’re still governed by the UN Convention on the Law of the Sea (UNCLOSS) , which gives ships the right to cross through narrow geographical chokeholds where part of the waters belong to another state, and that says that nations “shall not hamper transit passage.” That requirement, I add, cannot be suspended.  Still, merchant captains don’t want to risk getting themselves and their crews blown up, or arrested and thrown in Evin Prison . Insurers don’t want to pay for any ship that gets blown up, or indeed, for the ensuing environmental catastrophe. And the UAE doesn’t want its pristine beaches covered in crude oil.  And so, the tankers are staying put . And they’ll stay there until one of four things happens: Of the first three, none feels particularly likely, at least in the short-to-medium term. Maybe I’m wrong. Maybe everything reverses and everyone suddenly works it out — Trump realizes that he’s touching the stove and pulls out after claiming a “successful operation.” The world is chaotic and predicting it is difficult. Nevertheless, before that happens, closing the Strait of Hormuz means that Iran can inflict pain on American consumers at the pump, and we’ve already seen a 30% overnight spike in oil prices , with the price of a barrel jumping over $100 for the first time since 2022 (though as of writing this sentence it’s around $95). With midterms on the horizon, Iran hopes that it can translate this consumer pain to political pain for Donald Trump at the ballot box.  This is all especially nasty when you consider that the price of oil is directly tied to inflation. It influences shipping costs, a lot of medicines, construction materials, and consumer objects have petrochemical inputs. In very simple terms, if oil is used to make your stuff (or get it to you), that stuff goes up in price. While this obviously hurts countries with which Iran has previously had cordial relations, (particularly Qatar which is a major exporter of LNG), I genuinely don’t think it cares any more.  I mean, Iran has launched drones and missiles at targets located within Qatar’s territory , resulting in (at the latest count) 16 civilian injuries. Qatar shot down a couple of Iranian jets last week . I’m not sure what pressure any of the Gulf countries could exert on Iran to make it back down.  I don’t see the security situation improving, either. Iran’s Shahed drones are cheap and fairly easy to manufacture, and developed under some of the most punishing sanctions, when the country was cut off from the global supply chain. It then licensed the design to Russia, another heavily-sanctioned country, which has employed them to devastating effect in Ukraine.  Iran can produce these in bulk, and then — for the fraction of a cost of an American tomahawk missile — send them out as a swarm to hit passing ships. Even without the ability to produce new ones, Iran is believed to have possessed a pre-war stockpile of tens of thousands of Shahed drones .  Shaheds aren’t complicated, or expensive, or flashy, or even remotely sophisticated, and that’s what makes them such a threat. It took Ukraine a long time to effectively figure out how to counter them, and it’s done so by using a whole bunch of different tactics — from l and-based defenses like the German-made Gepard anti-aircraft gun , to interceptor drones , to repurposed 1960’s agricultural planes , to (quite literally) people shooting them down with assault rifles from the passenger seat of a propeller-powered planes .  Ukraine has the experience in combating these drones, and even still some manage to slip through its defences, often hitting civilian infrastructure . Airstrikes can probably reduce the threat to shipping (though not without exacting an inevitable and horrible civilian cost), but they can’t eliminate it.  Hell, even the Houthis — despite only controlling a small portion of Yemen, and despite efforts by a coalition of nations to degrade its offensive capabilities — still pose a risk to maritime traffic heading towards the Suez Canal.  Given the cargo these ships carry, any risk is probably too much risk for the insurers, for the carriers, and for the neighbouring countries. While I could imagine the US, at some point, saying “great news! It’s fine to go through the Strait of Hormuz now,” and though it has started offering US government-backed reinsurance for vessels , I don’t know if any shippers will actually believe it or take advantage of it.  And so, we get to the last point on my list. Regime change.  Do I believe that the Iranian government is deeply unpopular with its own people? Yes. Do I believe that said government can be overthrown by airstrikes alone? No. Do I believe that Iran’s government will do anything within its power to remain in control, even if that means slaughtering tens of thousands of their own people? Yes.  Even if there was an uprising, who would lead it? Iran’s virtually cut off from the Internet , and movement within the country is restricted, making it hard for any opposition figures to organize. The two most high-profile outside opposition figures — Reza Pahlavi, the son of the former Shah, and Maryam Rajavi, leader of the MEK and NCRI — both have their own baggage, and they’re living in the US and France respectively.  As I said previously, this isn’t me wading into geopolitics, but more of a statement that there’s no way of knowing when things will eventually return to normal. This conflict might wrap up in a couple of weeks, or it might be months, or, even longer than that. All this amounts to a huge amount of global oil production being bottled up, which is made worse by the fact that there’s also the slight problem that Iran produces a lot of oil itself, sending most of it (over 80%) to China . With Iran unable to export crude, and its production facilities now under attack, China’s going to have to look elsewhere. Which will result in even higher oil prices.  Which, in turn, will make everything else more expensive.  That is what brings us back to the AI bubble.  Now, given that most of the high-profile data center projects you’ve heard about are based in the US, which is (as mentioned) largely self-sufficient when it comes to hydrocarbons, you’d assume that it would be business as usual.  And you would be wrong.  You see, this is a global market. Prices can (and will!) go up in the US, even if the US doesn’t import oil or natural gas from abroad, because that’s just how this shit works. Sure, there are variations in cost where geography or politics play a role, but everyone will be on the same price trajectory. While we won’t see the same kind of shortages that we witnessed during the last oil shock (the one which ended up taking down the Carter presidency ), it will still hurt . While the US managed to decouple itself from oil imports, it hasn’t (and probably can’t) decouple itself from global pricing dynamics.  The US has faced a few major oil shocks — the first in 1973 , after OPEC issued an embargo against the US following the Yom Kippur War, which ended the following year after Saudi Arabia broke ranks, and the second in 1979, following the Iranian Revolution — and both hurt…a lot. This won’t be much different.  First, inflation.  As the cost of living spikes, people will start demanding higher wages, which will, in turn, be passed down through higher prices.  At least, that’s what would normally happen. Paul Krugman, the Nobel-winning economist, wrote in his latest substack that US workers in the 1970s were often unionized, and they benefited from contractual cost-of-living increases in their work contracts.  Sadly, we live in 2026. Union membership hasn’t recovered from the dismal Reagan years, and with layoffs and offshoring, combined with an already tough jobs market, workers have little leverage to demand raises. We’re in an economy oriented around do-nothing bosses that loathe their workers , one where workers will get squeezed even further by the consequences of any economic panic, even if it’s one caused by multiple events completely out of their control. So, it’s unlikely that we’ll see a wage-based amplification of any inflation that comes from the current situation.  That said, depending on how bad things get, we will see inflation spike, and Increases in inflation are usually met with changes in monetary policy, with central banks raising the cost of borrowing in an attempt to “cool” the economy (IE: reduce consumer spending so that companies are forced to bring down prices).    And we’d just started to bring down interest rates, with the Fed announcing in December that it projected rates of 3.4% by the end of 2026 . Iran changes that in the most obvious way possible — if prices soar, interest rates may follow, and if rates go up, even by a point or two of a percentage, financing the tens and hundreds of billions of dollars in borrowing that the AI bubble demands will become significantly more expensive.  For some context, the International Monetary Fund’s Kristalina Georgieva recently said “...a 10% increase in energy prices that persists for a year would push up global inflation by 40 basis points and slow global economic growth by 0.1-0.2%,” per The Guardian, who also added… And remember : the AI bubble, along with the massive private equity and credit funds backing it, is fueled almost entirely by debt. All this chaos and potential for jumps in inflation will also affect the affordability calculations that lenders will make before loaning the likes of Oracle and Meta the money they need at a time when lenders are already turning their nose up at Blue Owl-backed data center debt deals . The alternative is, of course, not raising interest rates — which, if the Fed loses its independence, is a possibility — which would be equally catastrophic, as we saw in the case of Turkey, whose president, Recep Tayyip Erdogan, has a somewhat… ahem… “unorthodox approach to monetary policy .  Erdogan believes that high interest rates cause inflation — a theory which he tested to the detriment of his own people .  In simpler terms, Turkey has faced some of the worst hyperinflation in the developed world , and has a currency that lost nearly 90% of its value in five years.  It’s not just the data centers, either. As interest rates go up, VC funds tend to shrink, because the investors that back said funds can get better returns elsewhere , and with much less risk.  As I discussed in the Hater’s Guide to Private Equity , 14% of large banks’ total loan commitments go to private equity, private credit and other non-banking institutions , at a time when ( to quote Forbes ) PE firms are taking an average of 23 months fundraising (up from 16 months in 2021), after private credit’s corporate borrowers’ default rates (as in the loans written off as unpaid by the borrow) hit 9.2% in 2025 . Put really simply, private equity, private credit, venture capital and basically everything to do with technology currently depends on the near-perpetual availability of debt. The growth of private credit is so recent that we truly don’t know what happens if the debt spigot gets turned off, but I do not think it will be pretty . Things get a little worse when you remember that famed business dipshits SoftBank are currently trying to raise a $40 billion loan to fund its three $10 billion Klarna-esque payments as part of its $30 billion investment in OpenAI’s not-actually-$110-billion-yet funding round . How SoftBank — a company that raised a $15 billion bridge loan due to be paid off in around four months and has about $41.5 billion in existing debt that’s maturing that needs to be refinanced in the next nine months or so, per JustDario — intends to take on another $40 billion is beyond me. And that’s a sentence I would’ve written before the war in Iran began. There’s also evidence that links lower IPO numbers to rising inflation rates , which means that achieving the exit that investors want will become so much harder — and so, they might as well not bother. Need proof? SoftBank-owned mobile payments company PayPay delayed its IPO last week, and I quote Reuters , because “...markets were rattled by [the attack] on Iran, according to two people familiar with the matter.” Inflation also negatively affects company valuations — which, again, will influence whether investors open their purse strings.  This is all a long-winded way of saying that the AI industry is about to enter a world of hurt. Every AI startup is unprofitable, which means they need to raise money from venture capitalists, who raise money from investors that aren’t paying them, pension funds and insurers, and private equity and credit firms that raise money from banks, both of which will struggle should central bank rates spike.  The infrastructural layer — AI data centers — also requires endless debt ( due to the massive upfront costs for NVIDIA chips and construction ), and that debt was already becoming difficult to raise.  Then there's the practical opex and capex costs. Higher interest rates mean that any contractors building the facilities will insist on higher fees, because their costs — labor costs, the price of filling up a van or a truck with gas, or paying for building materials — has gone up. And they’ll probably pad the increase a bit to take into account for any future rises in inflation.  Those gas turbines you’re running to power your facility? Yeah, feeding those is going to get much more expensive. Natural gas is up as much as 50%, and a lot of US capacity is going to serve markets in Asia and Europe to take advantage of the spike in prices , which will mean an increase in prices for US consumers.  In fact, you don’t even need interest rates to spike for things to get nasty. As the price of oil continues to skyrocket, flying a Boeing 747 filled with GB200 racks from Taiwan to Texas or mobilizing the thousands of people that work ( to quote Bloomberg ) day and night to build Stargate Abilene will become extra-normally more expensive. And even in the very, very unlikely event that things somehow quickly return to whatever level of “normal” you’d call the world before the conflict started, even brief shocks to the financial plumbing are enough to destabilize an already-fractured hype cycle. Last week, Bloomberg reported something I’d already confirmed three weeks ago — that OpenAI was no longer part of the planned expansion (past the initial two (of eight) buildings) of Stargate Abilene, a project that’s already massively delayed from its supposed “full energization” by mid-2026 .  Oracle disputes the report (and if it’s wrong, I imagine investors will rightly sue) claiming that “Crusoe [the developer] and Oracle are “operating in lockstep,” which doesn’t make sense considering the delays or, well, reality. My sources in Abilene also tell me that the expansion fell apart due to Oracle’s dissatisfaction with the revenue it was making on buildings one and two, and that a bidding war was taking place between Meta and Google for the future capacity.  Bloomberg’s Ed Ludlow also reports that NVIDIA put down a $150 million deposit as Crusoe attempts to lock down Meta as a tenant — a very strange thing to do considering Meta is flush with cash, suggesting a desperation in the hearts of everybody involved. It’s also very, very strange to have a supplier get involved in a discussion between a vendor and a customer , almost as if there’s some sort of circular financing going on. As I reported back in October, Stargate currently only has around 200MW of power , and The Information reports that power won’t be available for a year or more, something I also said in October .  As self-serving as it sounds, I really do recommend you read my premium piece about the AI Bubble’s Impossible Promises , because I laid out there how stupid and impossible gigawatt data centers were before the war in Iran. We’ve already got a shortage in the electrical grade steel and transformers required to expand America’s (and the world’s) power grid, we’ve already got a shortage of skilled labor required to build that power (and data centers in general) , and we’re moving massive amounts of heavy shit around a large patch of land using thousands of people, which will cost a lot of gas. I don’t know why, but the media and the markets seem incapable of imagining a world where none of this stuff happens, clinging to previous epochs where “things worked out” and where “things were okay” without a second thought. In The Black Swan , Nassim Taleb makes the point that “…the process of having [journalists] report in lockstep [causes] the dimensionality of the opinion set to shrink considerably,” saying that they tend to “[converge] on opinions and [use] the same items as causes.”  In simpler terms, everybody reporting the same thing in the same way naturally makes everybody converge on the same kinds of ideas — that AI is going to be a success because previous eras have “worked out,” even if they can’t really express what “worked out” means.  The logic is almost childlike — in the past, lots of money was invested in stuff that didn’t work out, but because some things worked out after spending lots of money , spending lots of money will work out here.  The natural result is that reporters (and bloggers) seek endless positive confirmation, and build narratives to match. They report that Anthropic hit $19 billion in annualized revenue and OpenAI hit $25 billion in annualized revenue — which has been confirmed to refer to a 4-week-long period of revenue multiplied by 12 — as proof that the AI bubble is real, ignoring the fact that both companies lose billions of dollars and that my own reporting says that OpenAI made billions less and spent billions more in 2025. They assume that a company would not tell everybody something untrue or impossible, because accepting that companies do this undermines the structure of how reporting takes place, and means that reporters have to accept that they, in some cases, are used by companies to peddle information with the intent of deception. And thanks to an affidavit from Anthropic Chief Financial Officer Krishna Rao filed as part of Anthropic’s suit against the Department of Defense’s supply chain risk designation , it’s clear that the deception was intentional, as the affidavit confirmed that Anthropic’s lifetime revenue “to date” (referring to March 9th 2026) is $5 billion , and it has spent $10 billion on inference and training.  To be abundantly clear , this means that Anthropic’s previous statement that it made $14 billion in annualized revenue ( stated by Anthropic on February 12 2026, and referring, I’ve confirmed, to a month-long period multiplied by 12 ) — referring to a period of 30 days where it made $1.16 billion — accounts for more than 23% of its lifetime revenue.  This comes down to which Anthropic you believe, because these two statements do not match up. I am not stating that it is lying , but I do believe annualized revenue is a deliberate attempt to obfuscate things and give the vibe that the business is healthier than it is. I also do not think it’s likely that Anthropic made 23% of its lifetime revenue in the space of a month. What this almost certainly means is that the sources that told media outlets that Anthropic made $4.5 billion in 2025 were misleading them . The exact quote from the affidavit is that “...[Anthropic] has generated substantial revenue since entering the commercial market—exceeding $5 billion to date,” and while boosters will say “uhm, it says “exceeding,” if it were anything higher than $5.5 billion Anthropic would’ve absolutely said so.  We can also do some very simple maths that suggests that Anthropic’s “annualized” figures are…questionable. On February 12 2026, annualized revenue hit $14 billion. Five days before the lawsuit was filed, it was $19 billion, “ with $6 billion added in February ” (per Dario Amodei at a Morgan Stanley conference), suggesting that annualized revenue was $13 billion, or $1.083 billion.  Even if we assume a flat billion, that means that Anthropic made $2.16 billion between January and the end of February 2026. And that’s not including the revenue made in March so far.  But I’m a curious little critter and went ahead and added up all of the times that Anthropic had talked about its annualized revenue from 2025 onward, and the results — which you can find with links here! — and based on my calculations, just using published annualized revenues gets us to $4.837 billion.  We are, however, missing several periods of time, which I’ve used “safe” (as in lower, so that I am trying to give Anthropic the benefit of the doubt) numbers to calculate based on the periods themselves. With these estimates, we get a grand total of $6.66 billion (ominous!), which is a great deal higher than $5 billion. When you remove the estimates and annualized revenues for 2026, you get $3.642 billion, which heavily suggests that Anthropic did not, in fact, make $4.5 billion in 2025. There isn’t a chance in Hell this company made $4.5 billion in 2025 based on its own CFO’s affidavit. I also think it’s reasonable to doubt the veracity of these annualized revenues, or, in my kindest estimation, that Anthropic is using any kind of standard “annualized” formula.  Here are the ways in which people will try and claim I’m wrong: I think it’s reasonable to doubt whether Anthropic made anywhere near $4.5 billion in 2025, whether Anthropic has annualized revenues even approaching those reported, and whether anything it says can be trusted going forward. It appears one of the most prominent startups in the valley has misled everybody about how much it makes, or if it has not, that somebody else is perpetuating a misinformation campaign. Add together the annualized revenues. Look at the links. Do the maths. I got the links for annualized revenues from Epoch AI , though I have seen all of these before in my own research.  People are going to try and justify why this isn’t a problem in all manner of ways. They’ll say that actually Anthropic made less money in 2025 but that’s fine because they all could see what annualized revenues really meant. So far, nobody has a cogent response, likely because there isn’t one. I haven’t even addressed the $10 billion in training and inference costs, because good lord, those costs are stinky , and based on my own reporting — which did not come from Anthropic, which is why I trust it! — Anthropic spent $2.66 billion on Amazon Web Services from January through September 2025, or around 26% of its lifetime compute spend. That’s remarkable, and suggests this company’s compute spend is absolutely out of control. This leads me to one more quote from Anthropic’s CFO: Without attempting to influence their decision making, if I were a counterparty to a company like this, my biggest concern would now be that this filing appears to suggest that Anthropic’s revenues are materially smaller than I believed. Though it might seem dangerous to be like me, pointing at stuff and saying “that doesn’t make sense!” Or questioning a narrative held by the entire stock market and most of modern journalism, but I’d argue the danger is that narrow, narrative-led, establishment-driven thinking makes it impossible for reporters to report.  While you might be able to say “a source told me that something went wrong,” the natural drive to report on what everybody else is saying means that this information is often reported with careful weasel words like “still going as planned” or “still growing incredibly fast.” It’s a kind of post-factual decorum — a need to keep the peace that frames bad signs as bumps in the road and good signs as cast-iron affirmations of future success. This is a catastrophic failure of journalism that deprives retail investors and the general public of useful information. It also — though it feels as if reporters are “getting scoops” or “breaking news” — naturally magnetizes journalists toward information that confirms the narrative, or “leaks” that are actually the company intentionally getting something in front of a reporter so that they (the reporter) can appear as if this was “investigative news” versus “marketing in a different hat.” It also means that modern journalism is ill-equipped, and no, this is not a “new” phenomena. It is the same thing that led to the dot com bubble, the NFT bubble, the crypto bubble, the Clubhouse bubble, the AR and VR bubble, and many more bubbles to come.  To avoid being “wrong,” reporters are pursuing stories that prove somebody else right, which almost invariably ends with the reporter being wrong. “Pursuing stories to prove somebody else right” means that a great many reporters (and newsletter writers) that claim to be objective and fact-focused end up writing the narrative that companies use to raise money using evidence manufactured by the company in question.  In some cases, this is an act of cowardice. Following the narrative because it’s easy and because everybody’s doing it adds a layer of reputation laundering. If everybody failed, everybody was conned and thus nobody has to be held accountable, and because there really has never been any accountability for the media being wrong about any previous bubbles, the assumption is that it’ll never happen.  However you may feel about my work or what I’m saying, I need you to understand something: journalism, both historically and currently, is unprepared for the consequences of being wrong.  The current media consensus around the AI bubble is that even if it pops it will be fine , with some even saying that “even if OpenAI folds, everything will work out, because of the dot com bubble.” This is a natural attempt to rationalize and normalize the chaotic and destructive — an attempt to map how this bubble would burst onto previous bubbles because new things are difficult and scary to imagine.  There has never been a time when the entire market crystallised around a few specific companies — not even the dot com bubble! — and then built an entire infrastructural layer mostly in service of two of them, with a price tag now leering close to the $1tn mark .   Let’s get specific. The scoffing and jeering I get from people when I say that AI demand doesn’t exist or that AI companies don’t have revenues or that OpenAI or Anthropic are unsustainable is never met with a good faith response , just quotes about how “Amazon Web Services lost lots of money” or “Uber lost lots of money” or that “these are the fastest growing companies of all time” or something about “all code being written by AI,” a subject I discussed at length two weeks ago .  The Large Language Model era is uniquely built to exploit human beings’ belief that we can infer the future based on the past, both in how it processes data and in how people report on its abilities. It exploits media outlets that do not have people that are given the time (or held to a standard where they have) to actually learn the subjects in question, and sells itself based on the statement that “this is the worst it’ll ever be” and “previous eras of investment worked out.”  LLMs also naturally cater to those who are willing to accept substandard explanations and puddle-deep domain expertise. The slightest sign that Claude Code can build an app — whether it’s capable of actually doing so or not — is enough for people that are on television every day to say that it will build all software, because it confirms the biases that the cycle of innovation and incumbent disruption still exists, even if it hasn’t for quite some time. A glossy report about job displacement — even one that literally says that Anthropic found “no systematic increase in job displacement in unemployment” from AI — gets reported as proof that jobs are being displaced by AI because it says “AI is far from reaching its theoretical capability: actual coverage remains a fraction of what’s feasible.”  This is an aggressive exploitation in how willing people with the responsibility to tell the truth are willing to accept half-assed expectations, and how willing people are to operate based on principles garnered from the lightest intellectual lifts in the world. The assumption is always the same: that what has happened before will happen again, even if the actuality of history doesn’t really reflect that at all. Society — the media, politicians, chief executives, shit, everyone on some level — is incapable of thinking of new stuff that would happen, especially if that new stuff would be economically destructive, such as a massive scar across all private credit, private equity and venture capital, one so severe that it may potentially destroy the way that businesses (and startups, for that matter) raise capital for the foreseeable future. People are more willing to come up with societally-destructive theories — such as all software engineering and all journalism and all content being created by LLMs, even if it doesn’t actually make sense — because it fits their biases. Perhaps they’re beaten down by decades of muting the power of labor or the destruction of our environment. Perhaps they’re beaten down by the rise of the right and the destruction of the rights of minorities and people of colour.  Or more noxiously, perhaps they’re excited to be the one that called it first, so that the new overlords that they perceive will own this (fictional) future, so much so that they’ll ignore the underlying ridiculousness of the economics, refuse to do any further reading that might invalidate their beliefs, or simply say whatever they’re told because it gets clicks and makes their advertisers, bosses or friends happy. People are willing to fall in line behind mythology because conceiving an entirely-different future is an intellectually challenging and emotionally draining act. It requires learning about a multitude of systems and interconnecting disciplines and being willing to admit, again and again, that you do not understand something and must learn more. There are plenty of people that are willing to do this, and plenty more that are not, and those are the people with TV shows and writing in the newspaper. I believe we’re in a new era. It’s entirely different. Stop trying to say “but in the past,” because the past isn’t that useful, and it’s only useful if you’re capable of evaluating it critically, skeptically, and making sure that it’s actually the same rather than it feeling like it is.  I keep calling this era “The Beginning of History,” not because it directly reflects Francis Fukuyama’s theory (which relates to democracies), but because I believe that those who succeed in this world are not those who are desperate to neatly fit it into the historical failures or successes of the past, but are willing to stare at it with the cold, hard fury of the present.  There are many signs that the past no longer makes sense. The collapse of SaaS (which I’ll cover in this week’s premium), the collapse of the business models of both venture capital and private equity, the collapse of democracies under the weight of fascism because the opposition parties never seem to give enough of a fuck about the experiences of regular people.  That’s because using the past to dictate what will happen in the future is masturbatory. It allows you to feel smart and say “I know the most about anything, which means I know what’s going on.” It is, much like an LLM, assuming that simply reading enough is what makes somebody smart, that shoving a bunch of text in your head — whether or not you understand it is immaterial — is what makes somebody know something or good at something.  It’s an intellectually bankrupt position that I believe will lead those unable to adapt to the reality of the future to destruction. It leads to lazy thinking that grasps at confirmations rather than any fundamental understanding, depriving the general public of good information in the favor of that which confirms the biases and wants and needs of the malignant and ignorant.  It takes courage to be willing to be wrong with deliberacy, but only if you admit that you were wrong. This hasn’t happened in previous bubbles, and it has to again for us to stop bubbles forming. I have made a great deal of effort to learn more as time goes on. I do not see boosters doing the same to prove their points. I will be pointing to this sentence in the future, one way or another.  So much more effort is put into humouring the ideas of the bubbles, of proving the marketing spiel of the bubbles, framed as a noxious “both-sides” that deprives the reader, listener or viewer of their connection with reality. It might be tempting to say this happens with cynicism too, except the majority of attention paid to bubbles is positive , and saying otherwise is a fucking lie. Need to justify unprofitable, unsustainable AI companies? Uber lost money before. Need to explain why AI data centers being built for demand isn’t a problem? Well, the internet exists, and people eventually used that fiber.  You can ignore actual proof while pretending to provide your own, all just by pointing vaguely to things in the past. It takes actual courage to form an opinion, something boosters fundamentally lack.  I’m not saying it’s impossible to make predictions, but that the majority of people make them with flimsy information, such as “this thing happened before” or “everyone’s saying this will happen.” I’m not saying you can’t try and understand what will happen next, but doing so requires you to use information that is not, on its face, generated by wishcasting or events that took place decades ago.  In the end, the greatest lesson we can learn from is that, historically speaking, people tend to fuck around and then find out.  The assumption boosters make is that one can fuck around forever. History tends to disagree. Iran rescinds its ban on travel through the strait. The security situation improves (either because Iran’s ability to attack shipping becomes sufficiently degraded, or because the Gulf countries, or perhaps their Western allies, feel sufficiently confident that they can safely escort ships through the strait).  The current Iranian government is overthrown and the conflict ends.  Both sides reach an agreement and we return to the status quo.  April 1 to 30, 2025, which I estimate as $166 million based on reports of Anthropic’s annualized revenue being $2 billion at the end of March 2025. August 1 to August 20, 2025, which I estimate as $271 million based on July 2025’s revenues ($4 billion). November 1 to November 29, 2025, which I estimate as $556 million, based on October’s $7 billion in annualized revenues.  January 1 to January 11, 2026, which I estimate as $219.1 million, assuming $9 billion in annualized revenue (based on reported December revenues). “Ed, it’s commercial revenue!” — this is all revenue. Anthropic doesn’t have “non-commercial revenue,” unless you are going to use a very, very broad version of what “non-commercial” means, at which point you have to tell me why you trust Anthropic. “This doesn’t include all the revenue up until March 2026! Maybe this suit was written weeks ago!” — even if it doesn’t, based on Anthropic’s own numbers, things don’t line up. Also, this was written specifically as part of the lawsuit with the DoD. It’s recent.  “It says “exceeding”! — it also says “over $10 billion in inference and training costs.” Can I just say whatever number I want here? Because if this is your argument that’s what you’re doing. “That $5 billion number is accurate!” — the only way this makes sense is if some or all of these annualized revenues are incorrect.

business Business Finance
0 views
Stratechery 4 days ago

Copilot Cowork, Anthropic’s Integration, Microsoft’s New Bundle

Microsoft is seeking to commoditize its complements, but Anthropic has a point of integration of their own; it's good enough that Microsoft is making a new bundle on top of it.

cloud Cloud business Business ai AI
0 views
Stratechery 1 weeks ago

2026.10: Higher Powers and Lower Macs

Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone . Additionally, you have complete control over what we send to you. If you don’t want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings . On that note, here were a few of our favorites this week. This week’s Sharp Tech video is on why Amazon is ramping AI spending. Anthropic and the Military.  This week’s Stratechery Interview with Gregory Allen of the Center for Strategic and International Studies and was one of my favorite conversations of the year so far. After a week of overheated rhetoric in every direction, Ben and Greg talk through the parallels and differences between AI and nuclear weapons, how the military uses autonomous weapons and the state of the art in 2026, and Allen provides some great insight into the process of contracting with the U.S. military and Anthropic’s process, specifically. I’d recommend this to anyone who’s been reading about the Anthropic standoff all week, as it was the best treatment of the issues that I’ve seen anywhere.  — Andrew Sharp U.S. History and Our Political Present. On Sharp Text this week, I offered my own thoughts on the Anthropic mess , including a tour of American history that makes clear the government leaning on private businesses is not new, legal challenges have been common, and particularly given the security implications of AI, the tension here is not particularly surprising. More importantly, I find myself exhausted by the way everyone processes political controversies these days, including warnings about a dire American future that are now a daily occurrence online. Come for Anthropic, then, and stay for my one great hope for the future.  — AS Apple Goes Downmarket. Apple released an entirely new Mac, and, for the first time in a long time (maybe ever?), the overriding motivation was to be cheap. We discuss John’s hands-on experience with the MacBook Neo on Dithering; it is both a Tim Cook special — no iPhone chip will go to waste! — and also the exact opposite of the super thin MacBook that I wanted a sequel to. — Ben Thompson Anthropic and Alignment — Anthropic is in a standoff with the Department of War; while the company’s concerns are legitimate, it position is intolerable and misaligned with reality. Technological Scale and Government Control, Paramount Outbids Netflix for Warner Bros. — Why government is not the primary customer for tech companies, and is Netflix relieved that they were outbid for Warner Bros.? Anthropic’s Skyrocketing Revenue, A Contract Compromise?, Nvidia Earnings — Anthropic’s enterprise business is reaching escape velocity, which increases the importance of finding a compromise with the government. Then, agents dramatically increase demand for Nvidia chips, even if they threaten software. An Interview with Gregory Allen About Anthropic and the U.S. Government — An interview with Gregory Allen about Anthropic’s dispute with the U.S. government. The End of the World As We Know It — On Anthropic’s standoff with the U.S. government and the exhausting nature of modern news commentary. Anthropic and the U.S. Government MacBook Neo Thyristors Did to Power What Transistors Did to Logic Vancomycin: The Iconic Antibiotic of Last Resort All Eyes on Iran; Two Sessions Questions; Alibaba, DeepSeek and Distillation; Another UK Spying Scandal An Emergency Bullseye Designation, Reviewing a Surprisingly Eventful Week, Remembering the 2011 Lockout League The Anthropic Mess Continues, Frontier AI and the Uncertain Future of Law, Q&A on Netflix, Dating Apps, F1

business Business ai AI
0 views
Stone Tools 1 weeks ago

Lotus 1-2-3 on the PC w/DOS

What would a piece of software have to do today to make you cheer and applaud upon seeing a demo? I don't mean the "I'm attending a keynote and this is expected, please don't glower at me Mr. Pichai," polite-company type of applause. I mean the "Everything's different now." kind. For that, the bar is pretty high these days. "Photorealistic" fight scenes between Brad Pitt and Tom Cruise against an apocalyptic cityscape are generated out of nothing but a wish, and social media, smelling the cynical desperation, can offer no more than a clenched-teeth grimace. Within 48 hours the cold light of the epic battle has faded, leaving no residual heat. A sense of awe was easier to elicit back in the golden era. Bill Atkinson scrubbed out some pixels with an eraser in MacPaint to thunderous applause. Andy Warhol did a flood fill on an image capture of Debbie Harry, leaving an audience enraptured. Perhaps miracles work best when they're minor. Mitch Kapor has been on the receiving end of the adulation. As CEO of newly-formed Lotus Corporation, demos of their flagship product 1-2-3 generated significant light and heat with the crowds. In a 2004 interview with the Computer History Museum, Kapor said, "You could with one-click see the graph from your spreadsheet. You could not do that before. That was the killer feature when we demo’d it. I mean, literally, people used to applaud – as hard as it is to believe." He knew all too well the struggles of the VisiCalc crowd, having previously built VisiPlot and VisiTrend for VisiCorp. Those programs worked with VisiCalc data to draw graphs, but required a lot of disk swapping to move in and out of the various programs when fine-tuning charts and graphs. 48K on the Apple 2 made it essentially impossible to fit all of the software into memory at once, but they could at least put everything onto the same diskette, Kapor reasoned. Eliminating that song and dance would be useful to the customers. Depicted as a literal song-and-dance in their advertising. In an interview in Founders at Work, Kapor said, "At various times I raised a number of ideas with the publisher about combining ( VisiCalc and VisiPlot onto one disk) and they weren't interested at all. I don't think they really saw me as an equal. They saw me, when I was there as a product manager, as an annoyance—as a marginal person without experience or credentials who was kind of a pest. And I suppose I was kind of a pest." He said the feeling was mutual, and that was basically it for his employment with Personal Software and the VisiCalc team. He let them buy him out (i.e. the juicy royalties he was receiving for VisiPlot and VisiTrend ) for $1.2M, then took that money and went off to build the better mousetrap he had tried to pitch. Lotus 1-2-3 would quickly become the "killer app" for the nascent IBM-PC, doing for that system what VisiCalc had done earlier for Apple. 1-2-3 's success (and corporate in-fighting between Personal Software and VisiCorp) drove VisiCalc sales into the ground almost immediately. Two years later, Lotus would buy out Personal Software. One year later, Lotus would kill VisiCalc . Today, Microsoft Excel documentation still references Lotus 1-2-3 , not VisiCalc . I have no 1-2-3 experience going into this. I always thought "1-2-3" referred to its relationship to numbers. "1, 2, 3. Row numbers. Numbers in a spreadsheet. Mathy number stuff. I get it." I honestly had no idea "1-2-3" indicated something more. I'm learning that VisiCalc walked so 1-2-3 could run (over VisiCalc's ashes in a Sherman tank) . I have one goal in learning Lotus 1-2-3 . I want to understand what it did that was so superior to my beloved VisiCalc that it practically wiped them out in the first year of launch. Kapor had projected first year 1-2-3 sales of US$1M, but did US$53M instead. That's not just a little better than VisiCalc, that's " VisiWho ?" dominance. VisiCalc is a spreadsheet and 1-2-3 is a spreadsheet, so what's the big fuss? First, the platform of choice, the IBM-PC running PC-DOS (MS-DOS, to those buying it separately), affords two big wins right off the bat. 80-column text mode makes the Apple 2's 40-columns feel claustrophobic (and perhaps a bit un-business-like?). The greatly expanded memory of the 16-bit PC, max 640K vs. the 8-bit Apple 2's 48K, lets far more complex worksheets fill out those roomy 80-columns. As Lotus Corporation and magazines and Wikipedia pages and other blogs love to point out, the true game-changer is contained in the program's very name. "1-2-3" refers to the three components of this "integrated software" package. "1" is the spreadsheet capability, which surpassed most contemporaries handily in speed, being written in x86 assembly (until Release 3). "2" is for those graphing tools which had Kapor's audiences applauding. "3" was intended to be a word processor, but according to programmer Jonathan Sachs, "I was a few weeks into working on the word processing part, and I was getting bogged down. That's about when Context MBA came out, and I got a look at what they had done." "What they had done" was integrate a word processor, communications, and database, along with the spreadsheet and graphics components. Context 1-2-3-4-5 , as it were. When Sachs saw the database, that felt to him like a more natural fit and "3" was re-implemented as a database. "It would be a heck of a lot easier to implement," he noted. Woz bless our lazy programmers. The upshot is 1-2-3 plays nicely with last post's focus, dBase , which feels like a particularly powerful combination. I feel a tingle when skills picked up on a previous exploration pay dividends later. Deluxe Paint + Scala paid off similarly. Is this what it feels like to "level up?" Obtaining literature on Lotus 1-2-3 is only difficult in the " overchoice " sense. I expected to find a lot of books, but perhaps not the "What have I gotten myself into?" existential dread of 1,000 hits on archive.org. It wasn't just books, that period had an interesting side phenomenon of "software vendor published enthusiast magazines." Companies like Aldus, Corel and Oracle all had self-titled publications on newsstands. Lotus Corporation did as well with LOTUS Magazine . Published monthly by Lotus Corporation, it debuted with the May 1985 issue (probably on newsstands late March, early April). The tagline, "Computing for Managers and Professionals," oriented itself toward the decision makers, the ones with purchasing power. A poll of Lotus software users revealed, "Most of you see the computer primarily as a tool and are not interested in computing, per se." Toward that end, the magazine took a different tack than the BYTE s and PC Magazine s of the time. It was to be no-nonsense, non-techno-babble, short, easy-to-digest articles about computing from the manager's perspective. "What's all this I keep hearing about 'floopy disks' and 'rams' and 'memories' and such and so on? It's enough to drive a reasonable business computerist straight to distraction!" says the frazzled corporate executive trope. There there, fret not! LOTUS Magazine feels your pain and addresses it with the cover story of issue 1. "The world of computer memory has enough complexity and high-tech jargon to drive the most reasonable business computerist straight to distraction," leads in to "An Inside Look at Computer Memory" by T.R. Reid. The article explains the differences between RAM and ROM, floppies and hard disks, and so on, unfurrowing the knitted brows of befuddled mid-80's business executives. When it got into the 1-2-3 of it all, LOTUS Magazine didn't pull its punches. Articles were short, around four pages, and assumed a higher level of analytical aptitude than IT aptitude. Lots of charts of formulas, macro definitions with explanations, tips and tricks for faster data entry, and so on fill out the pages. That ran for about seven years, until the December 1992 issue, when publishing duties transferred to PC Magazine as PC Magazine: LOTUS Edition . It was PC Magazine with a mini-magazine's worth of Lotus-specific content appended each month, as a special imprint. That ran until August 1995 , marking a 10-year publication run which would have exceeded my prediction by about eight years. After judging books entirely by their covers, I've chosen the official Lotus manuals for 1.0A, 2.2, and 3.4, and two compilations of tips and tricks previously published in LOTUS Magazine . I flip through other stuff as well, but honestly nothing is holding my attention this time around; they all read the same, "dry and boring." 1,000 pages or more for some of those books and they didn't have room for even one joke? I promise at least seven in this post alone. See if you can spot them all! Launching into the program proper brings me to the expected "I'm a spreadsheet!" grid layout, with column and row labels, arrow-key controllable cell cursor, and a blank area at the top for VisiCalc -y stuff. Let's go. As an intermediate level VisiCalc user, I am delighted my menu muscle memory pays immediate dividends. Clearly Lotus welcomes defectors and even makes life easier on everyone by taking advantage of the 80-column display. VisiCalc 's single-letter menu mnemonics are enhanced in 1-2-3 by simply spelling it all out on-screen. Full menu item names are always visible, yet still accessible by single-letter commands. From the jump, 1-2-3 makes a strong case for itself, providing improved usability and discoverable tools. Before digging in too deeply, I should note that 1-2-3 does all of the VisiCalc things. A1-style cell references, slash menu, fixed and relative cell references, @ functions including transcendentals, range specifier, prefix for values, and on and on. It adds, it subtracts, it calculates interest. 1-2-3 "Yes, and..."s VisiCalc from there. We gain a lot, but there is a notable absence: the upper-right status check. VisiCalc shows calculation order, arrow-key toggle, and free memory in that spot. Those are all gone in 1-2-3 and good riddance, frankly. On the PC I have full arrow keys and more RAM than Woz; 1-2-3 sees my full 16MB of DOS Extended memory. There is no stopping me. 1-2-3 also says nuts to VisiCalc 's "calculation order" (by row or by column) hoo-hah and introduces "minimal recalculation." From the almost comically-straightforward named book Lotus 1-2-3, Release 2.3 , "When 1-2-3 recalculates a worksheet, only those formulas directly affected by a change in the data are recalculated." I am living large here in 1989, or 1991, or whatever year I'm pretending it is this week. Even VisiCalc 's gets a glow up. You know it today as and , both of which were present in 1-2-3 Release 1 back in 1983. At this rate, 1-2-3 is flirting dangerously close to "expected spreadsheet behavior in 2026." Don't get my hopes up, Lotus. There's only down from there. The more I encounter this, the more I wonder if we gave up on it too soon. This could be "blogger overly immersed in their subject matter" brain, but I'm growing to oftentimes prefer two-line horizontal menus over modern GUI menus. I find the left-right, up-down, left-right, up-down, scanning through GUI menus kind of tiring. With the two-line menu, I can step through top-level options with the left/right arrow keys, eyes focused on line two as I scan sub-menu items. It also provides something GUI menus don't: an immediate explanation of a menu item before committing its action to the document. If a menu item is not a sub-menu, line two describes it. It's easy to audit features in an unknown program. Also, every menu item has a keyboard shortcut; just type the first letter. This requires creativity by the developer when naming menu items such that each has a unique first letter, but it also creates a de-facto mnemonic for the user. Don't discount muscle memory! There's one "drawback," but I'll try to make a case for it. Specifically, it is probably impossible to fit everything in a modern GUI menu into a two-line scheme. There's just too much! I suggest the horizontal menu-bar solves this precisely because of that design constraint. If there's too much, the menu needs to be simplified. "Problem solved," the author asserted. This has to be one of 1-2-3 's greatest contributions to modern spreadsheets. It still exists, just open up your modern spreadsheet of choice and try it. Enter 1 through 5 down the A column. Starting with B2, enter the formula and copy it down a few rows. Old hands know that a symbol in a cell reference fixes that row or column of the reference, otherwise references are relative. That's a huge step up from VisiCalc 's "all or nothing" approach to cell references. Put in a formula and copy it through to other cells. For every cell reference, in every copy of the formula, VisiCalc prompts the user for "relative or fixed?" It is a complete drag, and Woz help you the day that formula needs updating. The approach is superior, allowing us to embed relativity into the formula itself. Then, copying a formula across cells copies our intent as a natural course. It's simple to understand and hard to mess up: my favorite combination. While it can't load non- 1-2-3 documents natively, Lotus does provide a nice translation tool for helping us get data out of the heavy hitters of the day. From a Stone Tools perspective, this handles everything I need so far, as VisiCalc and dBase are both accounted for and work as advertised. Translation works both ways, so bringing in dBase data, messing around with it in 1-2-3 , and going back out to dBase is possible, though there are cautions in doing so. One notable thing to watch out for is "deleted" records. dBase only "marks for deletion" (until a .PACK command), and that flag won't survive transit. A small inconvenience, all things considered. In the top-level menu is the shiny new option, the "2" in "1-2-3." I know exactly what I want: a pie chart of game software genres imported from dBase II . The options for are straightforward, and the limitations are self-evident. Notably, look at the "Ranges" settings. Range sets value labels which will appear along the X-axis. Ranges through define six, and only six, ranges of data to plot on the graph. That's it. Everything else you see is "make it pretty." Within the confines of my self-imposed time capsule, my only point of reference thus far is VisiCalc and its clones. Through that lens, I'm blown away by Lotus 1-2-3 . I mean, come on, 3-D bar charts ?! Am I living in the world of TRON right now?! The applause is well-earned, Mitch. Bravo! Encore, even! Now, Mr. Kapor, if you'll excuse me a moment, I need to have a quick, private chat with my readers. Yes, sorry, I'll only be a moment. Hello dear readers. Mitch can't hear us, yeah? We're safe? OK, between you and me, that graphing tool is a little underwhelming, huh? There's a lot we can do to make a graph look as pretty as possible for screens and printers of the time, but the core graphing options themselves are kind of anemic. Here's Google Sheets making the pie chat I'd hoped 1-2-3 could generate. However, 1-2-3 cannot do this because it can only graph strict numeric values; strings, like "genre" types, return blank charts. 1-2-3 also can't coalesce data, like we see Sheets doing above. To achieve my goal, I'll need to figure out a different approach. (Plus, maybe I've discovered a DOSBox-X bug ?) It's not fair to judge past tools as being "inferior" just because they don't live up to 2026 standards. Still, what I'm trying to do must have been one of the first things many business owners wanted to do, right? Am I storing my data in a style that hadn't been popularized yet? Is my 2026 brain making life more difficult for my 1991 doppelgänger unnecessarily? How does one graph out the count of each unique genre? Alright, this is going to get complicated, so I think a diagram is in order. This actually explains a lot about the Lotus 1-2-3 approach to data in general, how to manipulate it, how to query it, and generally how to interface with the more complex functions of the program. Having imported the dBase list of CP/M games from the dBase article, let's extract a list of all titles that are of genre "Simulation." I'll use a subset of the total data so everything fits on screen for demonstration purposes and perform (aka , aka The Notorious DQU, aka Query's L'il Helper) A worksheet is not just rows and columns of data. It also serves as a control mechanism for defining interactions with the data. A worksheet has columns up to IV (256) and rows up to 8192. What do we do with 2,000,000+ cells? In true Dwarf Fortress fashion, we section off areas ("ranges" in 1-2-3 speak) and designate functions to those areas. First, I have my data as the main table, field names at top. Then, I need to set up my query criteria. This is a separate portion of the worksheet, with the fields I want to query against and room below to accept the criteria definition. Think of it like building a little query request form. Then, Lotus needs a place to spit out the results. Again, I set up a little "form" to receive the data. Put in whichever field names are of interest in the final data capture. Now, what if there are multiple queries I want to re-use from time to time? Painful as it sounds, I must set up multiple query forms, one for each query I expect to re-use. So, re-copy all of the field headers of interest into a new portion of the worksheet. Re-copy the field headers for the output range. Put in the new query criteria. Do another extraction. Keep dividing the worksheet up into all of the various queries one might need to reuse. Each lives in its own little area of the worksheet, so maybe now's a good time to start labeling things? Maybe mentally divide the worksheet into "my queries live over here, in Q-Town" and "my results live over there, in Resultsville" and so on. For my stated goal, I need the unique list of genres for my game list and the count of each genre within the data set. From the previous section, I know how to extract a list of unique genres. To count them, can count all non-empty records which match my criteria. Lemme draw up another diagram here. After extracting the list of unique values for "Genre", I get a column of results as seen at in the image above. Notice the criteria at is empty? By not specifying anything, that equates to matching any "Genre". Next, I need to reformat that column into countable criteria for . Just like in a query, criteria consists of two vertically contiguous cells, the top of which is the field name and the bottom holds the parameter. The field name must be physically, immediately above each and every genre I want to count. will transpose a range of vertical or horizontal cells into their mirror universe opposite. That's how I generated the horizontal list at . A of the field name across row 15 generated nice pairings, perfect for use with . The cell formula outlined in yellow is essentially the same across , each lightly modified to point to a different criteria range. That calculates the count for each genre in column , and column holds my titles. Now I have what I need to generate the chart I wanted (aforementioned pie chart drawing bug notwithstanding). Here it is in glorious 3-D from the future (of the past)! Frustratingly, figuring all of that out took the better part of a day. But now I know! If only there were some way to make it easier. There are issues with my solution thus far, many of which boil down to the physical spaces assigned to hold queries and results and transformations and data. If I bring in new data with new genres, new result lists could physically lengthen and overlap one another. Planning a physical map for the worksheet is a priority. Building out the sheet, especially keeping cell references flexible to changes in data, is a drag. I'd also like to generate a graph from the new sheet arrangement, with just a simple hot-key. Like all great developers, I want to be lazy. The first step toward the promised land of laziness is "hard work," unfortunately. Hard work can be captured and reused, luckily, as Lotus 1-2-3 features "Friend of the Blog": macros. VisiCalc didn't have it, and 1-2-3 's implementation is robust enough that many books were devoted to understanding and taming it. Here's a simple macro, which hints at its latent power. 0:00 / 0:07 1× Custom menus are easy to build. Selecting an option could trigger a longer automation task, simplifying a multi-step process, or something as simple as a help menu. Macros are stored... ( say it with me now ) ...in the worksheet. Yep, whatever map you had in mind for dividing up the worksheet into query-related fiefdoms, redistrict once more to hold macro definitions. Custom menus are an easy way to illustrate macro structure. Here's a dumb example. The text in column A is mostly comments to organize our worksheet and thoughts. represents the keyboard shortcut assigned to the macro, accessed by . is a reference to a named cell range. Named ranges are an important improvement over VisiCalc . Once defined, a range can be invoked by name anywhere a range is expected. Assuming a cell range as has been assigned a name like , is totally valid. is a range defined as . is a range defined as . Notice a range only needs to define the first start of a macro definition. Macro execution will read each cell in order down a given column until the first empty cell. range names are interpreted by 1-2-3 as macro keyboard shortcuts automatically. The convention shown, of a human-readable label to the immediate left of a range by the same name is so common it has its own menu shortcut. applied to column A will auto-assign column B cells to the names in A. To a certain extent, a named range can function like a programming "goto". In the macro case, its saying "Goto the range named and continue executing the macro from there." Programmers in the readership are salivating at the deviously complex ways this "goto labeling" could be abused. Combine it with decision making through and iteration through and the possibility space opens wide. After doing dBase work last post, I noted that I had accidentally become a dBase developer without even trying; the dBase scripting language was precisely equivalent to the commands issued at the dot prompt. I'm not so lucky with 1-2-3 . Setting up a macro which issues a simple string of commands is easy enough, and reads (mostly) like how I'd type it at the menu, akin to Bank Street Writer 's approach to macros. For example, will issue to bring up the slash menu, access the ( W )orksheet menu, then the ( C )olumn sub-menu, and finally ( H )ide a column. ~ issues "enter", which at this point in the menu navigation will commit the prompt default, i.e. the current position of the cursor. Just like that, hiding the current column just became a single keystroke. There is also a menu tool which is "record every keystroke I do from now." That recording will be output into the worksheet. Apply a range name to that and it transforms into a macro. Very nice! That said, 1-2-3 macros go from zero to 100 pretty quickly and are visually difficult to parse and reason out. One must be super-duper intimately familiar with every command in the slash menu, plus the macro-specific vocabulary. Lotus understood things could get hairy pretty quickly and added a debugging tool to help make sense of things. enters mode, which executes macros one line at a time. The status bar at the bottom of the screen explains what is being run, so when something goes wrong I know who to blame. OK , are you ready to dig in and implement macros which simplify the queries and procedure discussed earlier? < cracking knuckles> Well, I'm not. < uncracks knuckles back to stiffness > The macro system has proven too complicated to feel any sense of control or mastery beyond Baby's First Macro™. With a couple of more weeks' study I think I could achieve my goal. Unfortunately, for this post, I am defeated. The "3" in "1-2-3", 1-2-3 can function as a database. A very simple, limited, one-row-equals-one-record, 8192 record max, 256 field max, flat database. Let's be honest, oftentimes that's more than enough. I showed examples of querying earlier, and that's as fancy as it gets for this. We can sort records ascending/descending by up to two keys, find and replace values, find records which match a search query, and extract those records into another area of the spreadsheet. And nothing else (at least for Releases 2.x). 0:00 / 0:52 1× Sorting dBase II data by genre. It may seem I'm giving this aspect of the program short-shrift, but so did Lotus. In their own manual for Release 2.2, macros have 300 pages devoted to them. Database functionality has 50, and the first 20 of those are instructions for typing in dummy data. Sorting, querying, finding, and extracting, the meat and potatoes of database-ing, warrant a mere 20 pages total. It's a useful feature and I'm glad it's here. It's enough to handle most of my meager needs. Beyond that, there's not much to say, except to note its legacy. It was an obvious idea to anyone who touched VisiCalc for more than five minutes, so its development feels inevitable. Do some database work in Excel tonight and light a candle for 1-2-3 . A very nice feature of 1-2-3 that fits right in with its "integrated" approach, is what we would call today "plug-ins" or "extensions," but which Lotus calls "add-ins." 1-2-3 shipped with a few. For example, one expanded macros by letting them live in-memory, for use across worksheets. Normally the only macros accessible to a worksheet are those defined within itself. Man, VisiCalc is just getting lapped by 1-2-3 's ingenuity, huh? According to a PC Magazine article about the state of add-ins, many business-people lived inside 1-2-3 all day long and wanted to do everything from within its confines . The 3rd party add-in after-market happily commodified those desires. In addition to obvious ideas, like automated save/backup utilities, or industry-specific analysis tools, add-ins could mold 1-2-3 into almost anything. Complete word processors, entire graphic subsystem replacements for complicated graphing needs, expert system logic, and non-linear function solvers were injected into the program. Oracle offered a way to connect to their external SQL databases from within the snugly confines of 1-2-3 's security blanket. The Lotus approach, being a product of lower-memory days, is both annoying and useful. Add-ins can be, though are not by default, loaded at app startup. Add-ins must be "activated" one-by-one to gain access to their extended powers, or "deactivated" to make room for other add-ins or a larger worksheet. I have enough memory, so I'm not in trouble here, though I'm sure it's easy to imagine on a 512K system that manual memory management was a real thing. Between macros and add-ins, 1-2-3 becomes an ecosystem unto itself, like dBase or HyperCard . One thing I don't like about Lotus's approach is how it can bifurcate the user experience. That's seen clearly with their own WYSIWYG add-in. With Release 2.3, Lotus included this add-in to help a world transitioning from textual interfaces into the flash and sizzle of OS/2, Windows, and Mac GUI interfaces. It's DOS for the GUI envious and frankly, I'm cold on it. It's not integrated elegantly, feels sluggish, and makes the program more difficult to use. Activating WYSIWYG switches the application from terminal mode to graphics mode, so already as a DOSBox-X user I'm annoyed at losing my lovely TrueType text. That's not Lotus's fault, but a blogger's gotta have his standards. The big usability problem is how the functionality of the program now splits in two. The menu works as before, but we also have a new menu for all things WYSIWYG. So, when you want to use a menu command, you must remember which menu holds that command. Many options appear at first blush to be the same as their counterparts, but they control WYSIWYG-specific parameters of those functions. Usually. That's not to say the add-in isn't useful for cell styling, or placing graphs into a worksheet directly. Making documents look nice is important after all. The boss needs to be impressed with those Q3 projection charts, even when they forecast doom. Especially then, probably! Release 3 embraced WYSIWYG as its main and only interface, no add-in required, which is probably why I keep gravitating to the 2.x releases. I'd chalk it up to being a stubborn old man, but the recent embrace of TUI interfaces by the Hacker News crowd seems to have me in good company. I'm writing this part on February 22. Two days prior, a project called "Pi for Excel: AI sidebar add-in for Excel" released and got good traction on Hacker News. As I noted in the XPER column , our current "AI" boom is the biggest, but not the first. English language interactions, first by keyboard and fingers-crossed-one-day-by-voice-if-AI-technology-continues-along-our-projected-path-of-wishes-and-dreams, were available as add-ins to various programs. Databases in particular were a notable target for those experiments. Consider how English-like dBase 's user interface is, and it doesn't take a huge leap to understand why developers felt something closer to true English was within reach. Symantec's Q&A had its natural language "Intelligent Assistant" built right in. R:BASE tried it with their CLOUT add-in, promising a user could query, "Which warehouses shipped more red and green argyle socks than planned?" The spreadsheet Silk promised built-in English language control over its tools. Like those self-published magazines at the start of this article, Lotus didn't want to miss out on this English parser party either. (For this exploration I must drop down into R2.01) Released for US$150 in late 1986, HAL is a memory-resident wrapper to 1-2-3 . We launch HAL directly, which in turn launches 1-2-3 . Its advertising explains the gimmick well enough. "Lotus HAL gives you the ability to perform 1-2-3 tasks using simple English phrases." What I've seen in my early time with it can honestly feel kind of magical. Look at how easily it generates monthly column headers. 0:00 / 0:22 1× That's pretty slick, I can't deny it. Similarly tedious actions are promised to be eased greatly by "requesting" HAL to do the heavy lifting. Here, I'm stepping through a quick tutorial to have HAL build an entire spreadsheet. I never touch the formula; I only describe it by intent. 0:00 / 1:14 1× HAL only recognizes the first three letters of anything. "Name" and "Names" and "Namaste" are all the same to well-meaning, but a bit dimwitted, HAL. As is the case for all such English-like languages for the time, it's English only within a generous definition of the word. Ultimately, we're learning to speak 1-2-3 's specific dialect and vocabulary. PC Magazine , February 1987, their HAL review was the cover story, " HAL comes with a 250-page manual. It is as important to read this manual as it is to read the 1-2-3 manual. All the commands are described as rigidly as the syntax of any command-line interface." That it takes a 250 page manual to explain how to speak "English" with HAL perhaps makes an argument against its own existence? The base 640K of DOS must hold both programs in memory at the same time, so this is a nice piece of corroborating history for those who think software today is too bloated. An industry-defining spreadsheet with graphing and database capabilities close to modern expectations, an online help system, plus a natural language interface, all run together in less than 1MB of RAM . There's the retro-computing dopamine hit I've been hoping for! HAL doesn't just provide an English-language interface to 1-2-3 's native tools, it brings its own unique toys to the Release 2.01 sandbox. I do need to emphasize the release version here, because some of these tools were later worked into the product proper over time. That said, HAL worked hard to be your friend. Even though HAL controls 1-2-3 , interfacing with it still feels bolted on. brings up the HAL dialog box, which isn't hard to remember, but never feels natural. Even after setting the HAL request dialog to remain on screen, it feels tenuous. Sometimes it toggles off after navigating a menu option, or the request box will intercept commands I wanted to do through the normal slash menu. It's in the way more than I expected, and I couldn't find a balance between "when I want it" and "when I don't." PC Magazine also felt that HAL is a bit of a kludge. Charles Petzold wrote in his review, "Is HAL really a natural-language interface for 1-2-3 ? Is it useful? Will it revolutionize the computer industry? Are menus dead? My answers are: Not really. Often. Give me a break. No way." This is all academic, because Lotus killed HAL . It has been difficult to find sales figures, though in a Raymond Chen post we catch a glimpse of the Softsel Hot List for December 1986. HAL hit the top 10 (along with other, future blog subjects), moving up the charts over the previous three weeks. On the other hand, it was only available for Releases 1A through 2.01, the pre-WYSIWYG releases, and never returned. Earlier I poked at macros, hoping to make charting "count by genre" easier, and failed. Then I got to ponderin' if HAL might be able to do it for me. Shockingly, HAL can, through its special vocabulary word "tabulate." It makes those previously complex actions, the ones I diagrammed earlier, so simple to perform I don't really need a macro (though I could make one). Check out this 80's magic . 0:00 / 0:22 1× We are supposed to be able to execute HAL requests via to have the system output the 1-2-3 commands HAL puts together to get the job done. It's a peek inside HAL 's brain, basically. If I watch HAL think, maybe it can teach me a better way to do all of the busywork I slogged through earlier? In 1962's Diffusion of Innovations , author Everett Rogers described five characteristics individuals consider when adopting new solutions to existing problems. If VisiCalc was the "existing problem," how well did Lotus 1-2-3 make its case as the "new solution?" In the VisiCalc post I talked about how much of its DNA is seen in modern spreadsheets. I see now that an equal case can be made for Lotus 1-2-3 . I'd phrase it as VisiCalc contributed the "look," and 1-2-3 contributed the "feel" we've come to expect. Where VisiCalc was life-changing for number crunchers, 1-2-3 positioned itself as an engine for business and executed that vision almost perfectly. Having gotten to know 1-2-3 over the past weeks, I can now say, "I get it." I see what the fuss was about and, truth be told, I'm a convert. Sorry, VisiCalc , you know I love you! But the next time I reach for a spreadsheet, I'm reaching for 1-2-3 . Ways to improve the experience, notable deficiencies, workarounds, and notes about incorporating the software into modern workflows (if possible). Obviously, it depends on what you're trying to do. For business work, it doesn't play well in groups unless you're the CEO and can dictate, "OK people, we're all switching to DOS now." For personal projects, it meets many common needs and doesn't feel too much like compromise, aside from the graphing. Heck, the DOS version supports mouse control, and you can always turn on WYSIWYG mode to approximate modernity. We're also in luck with Y2K compatibility. Even Release 1.0 supports dates up to the year 2099. Let's take a moment of silent appreciation for yet another 1-2-3 foresight which keeps its spirit alive and kicking here in the 21st century. DOSBox-X 2026.01.02, Windows x64 build. I updated from the 2025.12 build mid-investigation. CPU set to 286 DOS reports as v6.22 Windows folder mounted as drive C:\ holds multiple Lotus installations 2x (forced) scaling; 80 columns x 25 lines I flipped back and forth with TrueType text mode (this is moot for 1-2-3 's WYSIWYG mode) Lotus 1-2-3 Releases 2.01, 2.2, 2.3, 2.4, and 3.4 all get exercised to some extent; you'll see that reflected in the screenshots. I mostly gravitate toward R2.3; it does what I need without bogging me down in feature creep. "Sharpening the Stone" explains getting DOSBox-X to work with R3.x. dBase III Plus for compatibility testing with 1-2-3 . Undoing your last action. It's almost worth installing HAL just for this, though it is a little dangerous that is the keyboard shortcut. Entering a sequential list of days, months, letters, or numbers automatically, though I wonder if macros could duplicate this to a certain degree. Linking a cell in one worksheet to data in another. Release 2.3 has this. Referring to columns and rows by name is a very neat trick. In fact, it's so neat I'm going to ask you to remember this fact for a later article. Just keep it tucked away in the part of your mind devoted to spreadsheet history, as we all have. The cell-row-bellum, I think its called? (I refuse to apologize.) Worksheet "auditing" can identify cell relationships/dependencies, or list out all formulas in use by a table in natural English. Auditing would become an add-in in later 2.x releases. Find and replace; change all instances of a product name, for example. Macros can mix HAL English with native 1-2-3 macro commands. "Relative advantage  is the degree to which an innovation is perceived as better than the idea it supersedes." 1-2-3 received applause for one-button graphing. Check. "Compatibility  is the degree to which an innovation is perceived as being consistent with...past experiences, and needs of potential adopters." 1-2-3 shipped with a VisiCalc translation tool and its interface is clearly built to make VisiCalc users comfortable. Check. " Complexity  is the degree to which an innovation is perceived as difficult to understand and use." 1-2-3 was initially praised for the simplicity with which a user could get up to speed. Its adoption of high-level VisiCalc concepts, like the slash menu, @ functions, and A1 cell references, helped. Check. "Trialability  is the degree to which an innovation may be experimented with on a limited basis." Trial disks for software during the 80's and 90's wasn't so prevalent; there was a lot of "blind faith" in software purchasing. I can't find any widespread cases of 1-2-3 demo disks circulating. No check. " Observability  is the degree to which the results of an innovation are visible to others." If the live demos, prevalent advertising, and magazine write-ups didn't convince you, 1-2-3 made it clear in the product name itself that you're getting 3x what VisiCalc delivers. Check. As with ThinkTank , DOSBox-X provided a simple, pain-free experience to get Lotus running. Multi-disk installs are handled well, but could be improved. Specifically, the "Swap Disk" option when loading up a stack of disks into the A: drive could use a selector and/or indicator of which disk is currently loaded. in autoexec.bat to auto-mount at launch. Revision 3.4 would not run until I explicitly set in DOSBox-X. I noted the pie graph bug in Release 2.x. I suspect, but cannot prove, that some x86 assembly call is being mangled by DOSBox-X. 86Box, which strives to be as pedantically accurate a simulation of real-world hardware as possible, does not exhibit this issue. However, setting up 86Box comes with a whole day of learning about the parts and pieces of assembling one's own raw DOS system from virtual components, installing from diskettes, and all of the old-school troubleshooting that entails. It's a commitment, is what I'm saying. I found that DOSBox-X would run the for Release 2.2, but failed to run it for Releases 2.3 and 2.4. can launch and run without issue. is a front-end utility to launch auxiliary programs like GraphPrint . If you're mounting a system folder as a "hard drive" in DOSBox-X, it is trivial to extract your data files. The Lotus utility "Translate" is handy for moving data between formats. I found that native .wk1 files open in LibreOffice , as-is. From there, you have any number of modern exporting options, though you might find some quirks from time to time. Check your formulas, just in case! I'd recommend checking out Travis Ormandy 's site. He's smarter than me and performs magic I didn't think possible, like pulling live stock data as JSON into 1-2-3 . He also got the Unix build to work natively in Linux.

data-visualization Data Visualization business Business scala Scala culture Culture json JSON
0 views
Martin Fowler 1 weeks ago

Ideological Resistance to Patents, Followed by Reluctant Pragmatism

Naresh Jain has long been uncomfortable with software patents. But a direct experience of patent aggression, together with the practical constraints faced by startups, led him to resort to defensive patenting as as a shield in this asymmetric legal environment.

business Business
0 views
Stratechery 1 weeks ago

Anthropic’s Skyrocketing Revenue, A Contract Compromise?, Nvidia Earnings

Anthropic's enterprise business is reaching escape velocity, which increases the importance of finding a compromise with the government. Then, agents dramatically increase demand for Nvidia chips, even if they threaten software.

ai AI business Business
0 views
Jim Nielsen 1 weeks ago

w0rdz aRe 1mpoRtAnt

The other day I was looking at the team billing section of an AI product. They had a widget labeled “Usage leaderboard”. For whatever reason, that phrase at that moment made me pause and reflect — and led me here to this post. It’s an interesting label. You could argue the widget doesn’t even need a label. You can look at it and understood at a glance: “This is a list of people sorted by their AI usage, greatest to least.” But it has that label. It could have a different label. Imagine, for a moment, different names for this widget — each one conjuring different meanings for its purpose and use: Usage leaderboard implies more usage is better. Who doesn’t want to be at or near the top of a leaderboard at work? If you’re not on the leaderboard, what’s that mean for your standing in the company? You better get to work! Calling it a leaderboard imbues the idea of usage with meaning — more is better! All of that accomplished solely via a name. Usage dashboard seems more neutral. It’s not implying that usage is good or bad. It just is , and this is where you can track it. Usage wall of shame sounds terrible! Who wants to be on the wall of shame? That would incentivize people to not have lots of usage. Again, all through the name of the thing! It’s worth noting that individuals and companies are incentivized to choose words designed to shape our thinking and behavior in their interest. The company who makes the widget from my example is incentivized to call this a “Usage leaderboard” because more usage by us means more $$$ for them. I’m not saying that is why they chose that name. There may not be any malicious or greedy intent behind the naming. Jim’s law is a variation on Hanlon’s razor : Don’t attribute to intent that which can be explained by thoughtlessness. I do find it fascinating how little thought we often give to the words we use when they can have a such a profound impact on shaping our own psychology, perception, and behavior. I mean, how many “word experts” are on your internal teams? Personally, I know I could do better at choosing my words more thoughtfully. Reply via: Email · Mastodon · Bluesky “Usage leaderboard” “Usage dashboard” “Usage wall of shame”

business Business design Design
0 views
Stratechery 1 weeks ago

Technological Scale and Government Control, Paramount Outbids Netflix for Warner Bros.

Why government is not the primary customer for tech companies, and is Netflix relieved that they were outbid for Warner Bros.?

entertainment Entertainment media Media business Business
0 views
André Arko 1 weeks ago

Four months of Ruby Central moving Ruby backward

From the moment RubyGems was first created in 2004, Ruby Central provided governance without claiming ownership , to support the Ruby community. Providing governance meant creating processes to provide stability and predictability. Avoiding ownership meant allowing the community to contribute, to the point where unpaid volunteers created and controlled the entirety of RubyGems.org for many years. Last year, Ruby Central flipped that successful formula on its head . They now claim ownership of both Bundler and RubyGems, but refuse to provide governance . Ruby Central now claims sole control over all code and decisions, despite paying for only a few percent of the work required to create and sustain the projects across 22 years. Instead of providing stable and predictable processes, Ruby Central suddenly hijacked the Bundler and RubyGems codebases away from the existing maintainers, shut out the community, and started issuing the threats to sue. When confronted by the former maintainers after the hijacking, Marty Haught of Ruby Central stated (in a recorded video call) on September 17 that “yeah, we shouldn’t have changed that”. On September 18, Marty went on to write: In the past, we’ve made the mistake of conflating ownership of the code with ownership of the infra, and vice versa, and we’d like to straighten this out so that we aren’t put in a legal bind that requires us to take control of the entire codebase when, we all agree, that is not proper or correct given the existing model. In the words of Ruby Central itself, “we all agree, [taking control of the entire codebase] is not proper or correct.” Since the beginning of this conflict, Ruby Central has privately admitted it was wrong to hijack the GitHub organization and steal the repos, but has refused to acknowledge this in public. Unfortunately, despite privately admitting their actions were wrong, Ruby Central has publicly continued to dig their hole deeper. Instead of owning up to their mistake, they secretly negotiated a deal with Matz for ruby-core to take over the stolen RubyGems and Bundler repository, further violating the project governance policies. If this situation were just about me personally, I could believe it sprang from from individual disagreements. Ruby Central claims they had good reasons to unilaterally kick me out of the project, even though I don’t think their claims hold water . With that said, regardless of what you think about me personally, the other five long-term maintainers have never gotten any explanation of why they were suddenly kicked out or bypassed entirely, all in violation of existing project governance. In her only public interview about the situation, Ruby Central Executive Director Shan Cureton defended stealing Bundler from its team of fifteen years by saying the removed team “didn’t need to have the story, and it wasn’t their story to have”. Ruby Central has made their position clear: if they steal your project, you are not entitled to know their reasons , and neither is anyone else. There is nothing “community-oriented” about stealing the most-used gem in Ruby and refusing to share your reasons with the community. Despite Ruby Central’s unacceptable treatment of both projects and maintainers, the former RubyGems and Bundler team said we want to move Ruby forward . We offered Ruby Central a path to move past their illegitimate GitHub takeover, past their vicious personal attacks, and past their threats to sue us. It has been four months since we made that offer, and Ruby Central has not accepted . While declining to accept our offer, Ruby Central has nonetheless found the time to propose new governance documents for RubyGems . In those documents, they explicitly require existing maintainers approve adding or removing team members. That rule was already present in the previous governance, and is the exact rule that Ruby Central violated to execute their takeover . When asked why they violated the previous governance, and why the new governance would be any more trustworthy, Ruby Central refused to respond substantively, and then the question itself was hidden by marking it “off topic” . Instead of working to resolve the situation, Ruby Central has spent 4 months rejecting requests for an explanation, while repeatedly threatening to sue me personally. After Ruby Central suddenly took over the Bundler repo, I sent them a standard trademark notice. They replied with a threat to sue me. When I later informed Ruby Central I had learned they violated state employment law, they simply replied with the same threat to sue me again. They are threatening to sue me for “hacking” them, despite their own analysis publicly concluding “no evidence that user data or production operations were harmed” . Without seeking common ground, or even looking for some sort of resolution we can just live with and move on from, Ruby Central has offered all of us — nothing . Ruby Central has made no offer in reply to outreach from the other five maintainers. To me, after four grueling months of private “negotiation”, their entire offer is nothing more than to refrain from suing. But only if I agree to everything that they want. They say I must agree that I have no claim on the name Bundler, despite helping create it and leading the Bundler team for the last 15 years. They say I must agree I was paid legally and fairly, when California law clearly states I was not. They say I must agree that Ruby Central can take over open source projects they host, any time they feel like it, with no explanation, and no consequences. I don’t agree. Letting this situation stay unaddressed sets a dangerous precedent for all open source projects written in Ruby. Ruby Central has resolved nothing. Don’t let their delaying tactics convince you otherwise. The Ruby community cannot trust Ruby Central with control over our gems until there is accountability for destroying the very governance they were supposed to be providing . Until accountability arrives, take action . Tell Ruby Central they owe everyone an explanation for violating the project governance around six long-term maintainers, not just me. Don’t sponsor, attend, or speak at RubyConf. Contribute to projects that aren’t controlled by Ruby Central. The exiled maintainers are working on new projects, with a focus on clear governance, long-term financial sustainability, and community input: Join the gem.coop beta, and stop using RubyGems.org. Use jwl instead of RubyGems. Use or Ruby Butler instead of Bundler. A better world is possible! Ruby Central might want to keep Ruby in the past, but we can work together to build Ruby a future .

open-source Open Source business Business ruby Ruby
0 views
neilzone 1 weeks ago

I'm struggling to think of any online services for which I'd be willing to verify my identity or age

Identity verification and age verification is an increasinly common policy conversation at the moment, in numerous countries. Often, this is in combination with proposals to ban children from varying concepts of “social media”, which generally means that everyone would have to prove that they were not a child. I have yet to see a well-considered proposal. Worse, the question that they are trying answer is rarely stated clearly and concisely. And it is unusual to see any consideration of broader sociological issues, let alone an emphasis on this, with a focus instead on perceived “quick win” technosolutionism. But anyway… I was pondering last night for which services I, personally, would actually be willing to verify my age or identity. And… the answer is “none”. At least, none that I can think of at the moment. I appreciate that I compute in an unusual way (when compared with most computer users), and that much of what I do online is about accessing my own services . Some of those - my fedi server, my RSS server, my messaging services - are build around enjoying stuff from other people’s services. Would I be willing to verify my identity or age to read someone’s RSS feed? No. While I enjoy the myriad blogs that I follow, none are crucial to me. I occasionally watch videos (which started on YouTube, but which I download into my Jellyfin instance), and perhaps YouTube will be forced to do age verification. It would be a shame, but again, I’ll just not watch YouTube videos. Not a big loss. Mostly, I buy secondhand DVDs, rip them, and watch them from my Jellyfin instance. I haven’t been asked to verify my age for a DVD purchase (online or offline) in a very long time. Friends have had to attempt to block access to their sites from the UK. While I can still access their sites via Tor, that’s what I tend to do. I feel sorry for them for the likely significant drop in visitors, likely affecting their enjoyment and in some cases their revenue, and, probably their incentive to continue to write / post / record stuff. I don’t use any individual forums any more (their demise is a shame; I’d prefer this over centralised discussion sites), nor do I use Reddit. I occasionally look at the comments on HN if one of my posts is surfaced there, but if HN forced identify or age verification, I’d just stop doing it. No big deal for me. Websites with comments sections? I don’t want to see the comments anyway, so I block those, which makes for a very pleasant browsing experience. I don’t comment myself. Code forges / places to contribute to FOSS? Most of my FOSS contributions are non-code, but even so, I use some organisation’s GitLab repos, and occasionally I contribute to projects on other forges. I doubt that my contributions are meaningful in themselves, and it may not be an option to switch infrastructure in any case (that might ont make the requirement go away), but since I am not a massive, or particularly valuable contributor, I’d feel less bad about simply stepping away. For Wikipedia, I’d probably rebuild my Kiwix instance and use that instead. Yes, articles would not be quite so up to date, but I rarely access Wikipedia for rapidly-changing information. In any case, there are tradeoffs, and personally I would prefer my privacy, the security of my personal data, and, well, just not being part of this kind of censorship. Signal? That would be a pain. I don’t have a workaround for that. I’m happily using XMPP, but as a complement to Signal, not an alternative. Teams/Zoom? I don’t have accounts on those services, but I do join, via my browser, when a client sends me a link. If I was faced with a choice of having to verify my identity/age for these services, then I’d have to consider the position carefully. Realistically, I am not in a position to say “no, I will not use Teams”, as some long-term clients are not going to change their corporate approach just because Neil doesn’t like something, and I’d rather not lose them as clients. So that could be a pain, if those services were within scope. I’ll still object to these measures - “I’m okay, Jack” would be a selfish stance - but, in practice, yes, I’d be surprised if they impacted me. Self-imposed (or, at least, self-controlled) digital isolationism, perhaps. Or perhaps, in the future, some service will pop up that I will really, really want to use, despite it requiring identity / age verification.

culture Culture security Security business Business
0 views
Stratechery 1 weeks ago

Anthropic and Alignment

Just because you do not take an interest in politics doesn’t mean politics won’t take an interest in you. ― Pericles This is not an Article about the campaign being waged by the U.S. against Iran, but it’s a useful — and timely — analogy. There is a never-ending debate that can be had about the concept of International Law and who might be violating it. Some will argue that the U.S. is in violation for the attacks; others will note that Iran has been serially violating International Law with both its overt actions and its support of terror networks for my entire life. What is important to note is that the entire debate is ultimately pointless: the very concept of “international law” is fake, not because pertinent statutes and agreements don’t exist, but because their effectiveness is ultimately rooted in their enforceability. That, by extension, means there must be an entity to enact such enforcement, with the capability to match, and such an entity does not exist. Yes, there is the United Nations, but said body only operates by the agreement of its members, and their willingness to subjugate themselves to not only its edicts, but to also put forward the capabilities to enforce its mandates. In other words, the only agents that matter are nation states themselves, and the relative power of those nation states is not a function of lawyers and judges but rather their ability to project force and coerce others. To put it another way, if, after this weekend, you want to hold onto the concept of International Law, then realize the debate has been resolved: Iran was in violation, because their military just had their clock cleaned by the U.S., which means the U.S. decides who is right and who is wrong. While most of the U.S. and certainly the rest of the world were preoccupied with the happenings in Iran, another fervent debate has been ongoing in tech. Once again one of the parties is the United States itself, but the other entity in question is a private company, Anthropic. From the Wall Street Journal : The federal government will stop working with Anthropic and designate the artificial intelligence company a supply-chain risk, a dramatic escalation of the government’s clash with the company over how its technology can be used by the Pentagon. While Anthropic’s relationship with the administration hit a new low, rival OpenAI said late Friday that it reached an agreement with the Defense Department to have its models used in classified settings, until recently a status only held by Anthropic. Friday’s quick-fire developments between the Pentagon and two Silicon Valley darlings are poised to shape the future of how the federal government and, particularly the Pentagon, uses cutting-edge AI tools. Anthropic staked out its position earlier in the week in a Statement from Dario Amadei on [its] discussions with the Department of War : In a narrow set of cases, we believe AI can undermine, rather than defend, democratic values. Some uses are also simply outside the bounds of what today’s technology can safely and reliably do. Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now: To our knowledge, these two exceptions have not been a barrier to accelerating the adoption and use of our models within our armed forces to date. The Department of War has stated they will only contract with AI companies who accede to “any lawful use” and remove safeguards in the cases mentioned above. They have threatened to remove us from their systems if we maintain these safeguards; they have also threatened to designate us a “supply chain risk” — a label reserved for US adversaries, never before applied to an American company — and to invoke the Defense Production Act to force the safeguards’ removal. These latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security. Regardless, these threats do not change our position: we cannot in good conscience accede to their request. I actually didn’t realize before this episode that the National Security Agency (NSA) is a part of the Department of War; that certainly provides useful context around the surveillance point. And, as we saw a decade ago with the Snowden revelations, the NSA can be both aggressive and creative in its interpretations of what is legal in terms of surveillance. One might have hoped that telecom companies in particular might have taken a stand like Anthropic did. At the same time, what is the standard by which it should be decided what is allowed and not allowed if not laws, which are passed by an elected Congress? Anthropic’s position is that Amodei — who I am using as a stand-in for Anthropic’s management and its board — ought to decide what its models are used for, despite the fact that Amodei is not elected and not accountable to the public. And, on the second point, who decides when and in what way American military capabilities are used? That is the responsibility of the Department of War, which ultimately answers to the President, who also is elected. Once again, however, Anthropic’s position is that an unaccountable Amodei can unilaterally restrict what its models are used for. It’s worth noting that there are reports that Anthropic’s concerns may be broader than just fully autonomous weapons; from Semafor : Anthropic is one of the few “frontier” large language models available for classified use by the US government because it is available through Amazon’s Top Secret Cloud and through Palantir’s Artificial Intelligence Platform, which is how its Claude chatbot ended up appearing on the screens of officials who were monitoring the seizure of then-Venezuelan President Nicolás Maduro… Soon after the Maduro raid, during a regular check-in that Palantir holds with Anthropic, an Anthropic official discussed the operation with a Palantir senior executive, who gathered from the exchange that the AI startup disapproved of its technology being used for that purpose. The Palantir executive was alarmed by the implication of Anthropic’s inquiry that the company might resist the use of its technology in a US military operation, and reported the conversation back to the Pentagon, a senior Defense Department official said. Anthropic denied it objected to whatever involvement Claude may have had in the Maduro raid, but the Semafor story resonates given the trend in some tech circles to resist any involvement in military operations. And, to that end, one could argue that this stand-off is ending as it should: Anthropic and its models will be removed from the Department of War tech stack, and an alternative will take their place. Amodei has been outspoken about other aspects of AI and national security; from Bloomberg in January : Anthropic Chief Executive Officer Dario Amodei said selling advanced artificial intelligence chips to China is a blunder with “incredible national security implications” as the US moves to allow Nvidia Corp. to sell its H200 processors to Beijing. “It would be a big mistake to ship these chips,” Amodei said in an interview with Bloomberg Editor-in-Chief John Micklethwait at the World Economic Forum in Davos, Switzerland. “I think this is crazy. It’s a bit like selling nuclear weapons to North Korea.” This rather raises the stakes of a messy procurement decision: consider the implications if we take Amodei’s analogy literally. Start with Iran: beyond the fact that Iran has been responsible for the deaths of thousands of Americans throughout the Middle East and beyond, one of the arguments for the U.S. intervention is that Iran continues to pursue nuclear weapons capabilities. It’s North Korea that shows why: North Korea doesn’t need to buy nuclear weapons, because they already have them, and it certainly makes any sort of potential military action against them considerably more complicated. Nuclear weapons make you an effective lawyer in the (nonexistent 1 ) court of international law! In short, nuclear weapons meaningfully tilt the balance of power; to the extent that AI is of equivalent importance is the extent to which the United States has far more interest in not only what Anthropic lets it do with its models, but also what Anthropic is allowed to do period. This, I think, gives important context to the designation of Anthropic as a supply chain risk. Secretary of War Pete Hegseth said on X : In conjunction with the President’s directive for the Federal Government to cease all use of Anthropic’s technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security. Effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic. This would decimate Anthropic: at a bare minimum the company relies on cloud hosting from AWS, Microsoft, and Google, all of which have contracts with the Department of War; I imagine the same applies to Nvidia. Fortunately for the company, Hegseth’s declaration does seem out of step with the law , which limits Hegseth’s authority to work covered by U.S. government contracts; in other words, AWS could still serve Anthropic models, as long as it doesn’t use Anthropic models for any of its services offered to the U.S. government. Regardless, this is an extreme measure that has been met with near universal dismay, even amongst people who are sympathetic to the idea that a private company should not have veto power over the U.S. military. Why would the U.S. government want to kneecap one of its AI champions? In fact, Amodei already answered the question: if nuclear weapons were developed by a private company, and that private company sought to dictate terms to the U.S. military, the U.S. would absolutely be incentivized to destroy that company. The reason goes back to the question of international law, North Korea, and the rest: Anthropic talks a lot about alignment; this insistence on controlling the U.S. military, however, is fundamentally misaligned with reality. Current AI models are obviously not yet so powerful that they rival the U.S. military; if that is the trajectory, however — and no one has been more vocal in arguing for that trajectory than Amodei — then it seems to me the choice facing the U.S. is actually quite binary: Note that I’m not making the (very good) argument put forward by Anduril founder Palmer Luckey about the importance of democratic oversight; Luckey wrote on X : This gets to the core of the issue more than any debate about specific terms. Do you believe in democracy? Should our military be regulated by our elected leaders, or corporate executives?… The fact that this is a debate over AI does not change the underlying calculus. The same problems apply to definitions and use of ethically fraught but important capabilities like surveillance systems or autonomous weapons. It is easy to say “But they will have cutouts to operate with autonomous systems for defensive use!”, but you immediately get into the same issues and more — what is autonomous? What is defensive? What about defending an asset during an offensive action, or parking a carrier group off the coast of a nation that considers us to be offensive? At the end of the day, you have to believe that the American experiment is still ongoing, that people have the right to elect and unelect the authorities making these decisions, that our imperfect constitutional republic is still good enough to run a country without outsourcing the real levers of power to billionaires and corpos and their shadow advisors. I still believe. And that is why “bro just agree the AI won’t be involved in autonomous weapons or mass surveillance why can’t you agree it is so simple please bro” is an untenable position that the United States cannot possibly accept. Again, I think this is a good argument; the one I am putting forward, however, is much more basic and brutal, and doesn’t have anything to do with belief or not in the American experiment (although I’m with Luckey in that regard): it simply isn’t tolerable for the U.S. to allow for the development of an independent power structure — which is exactly what AI has the potential to undergird — that is expressly seeking to assert independence from U.S. control. I don’t, for the record, want Anthropic to be destroyed, and I want them to be a U.S. AI champion. I also, for the record, don’t trust Amodei’s judgment in terms of either national security or AI security. In terms of national security, I already commented on Amodei’s Davos comments on X : Last year I laid out in AI Promise and Chip Precariousness why I believed a systemic view of the U.S.-China rivalry entailed some painful tradeoffs when it came to chips and China: The important takeaway that is relevant to this Article is that Taiwan is the flashpoint in both scenarios. A pivot to Asia is about gearing up to defend Taiwan from a potential Chinese invasion or embargo; a retrenchment to the Americas is about potentially granting — or acknowledging — China as the hegemon of Asia, which would inevitably lead to Taiwan’s envelopment by China. This is, needless to say, a discussion where I tread gingerly, not least because I have lived in Taipei off and on for over two decades. And, of course, there is the moral component entailed in Taiwan being a vibrant democracy with a population that has no interest in reunification with China. To that end, the status quo has been simultaneously absurd and yet surprisingly sustainable: Taiwan is an independent country in nearly every respect, with its own border, military, currency, passports, and — pertinent to tech — economy, increasingly dominated by TSMC; at the same time, Taiwan has not declared independence, and the official position of the United States is to acknowledge that China believes Taiwan is theirs, without endorsing either that position or Taiwanese independence. Chinese and Taiwanese do, in my experience, handle this sort of ambiguity much more easily than do Americans; still, gray zones only go so far. What has been just as important are realist factors like military strength (once in favor of Taiwan, now decidedly in favor of China), economic ties (extremely deep between Taiwan and China, and China and the U.S.), and war-waging credibility. Here the Ukraine conflict and the resultant China-Russia relationship looms large, thanks to the sharing of military technology and overland supply chains for oil and food that have resulted, even as the U.S. has depleted itself. That, by extension, gets at another changing factor: the hollowing out of American manufacturing under Pax Americana has been directly correlated with China’s dominance of the business of making things, the most essential war-fighting capability. Still, there is — or rather was — a critical factor that might give China pause: the importance of TSMC. Chips undergird every aspect of the modern economy; the rise of AI, and the promise of the massive gains that might result, only make this need even more pressing. And, as long as China needs TSMC chips, they have a powerful incentive to leave Taiwan alone. The key thing to consider is the opposite scenario: cutting China off from advanced chips doesn’t just reduce the likelihood that Chinese companies are dependent on a U.S.-based ecosystem, it also reduces the cost of destroying TSMC. More than that, if AI becomes as capable as Amodei says it will — the equivalent, or more, of nuclear weapons — then it actually becomes game theory optimal for China to do exactly that: if China can’t have AI, then it is, at least under current circumstances, relatively easy to make sure that nobody does. Amodei is, as the quote above notes, cognizant of China as a threat generally; it concerns me that he consistently fails to acknowledge that the implication of his recommended course of action in terms of chip controls is to risk destroying AI for everybody. Then again, Amodei isn’t really a fan of AI for everybody: he and Anthropic have been vocal opponents of open source models, and were major drivers of what I considered a very misguided Biden executive order about AI . Like the Taiwan situation, I think these positions evince a failure to think systematically: There is certainly room for disagreement on these points; what concerns me about Amodei and Anthropic in particular is the consistent pattern of being singularly focused on being the one winner with all of the power, with limited consideration of how everyone else may react to that situation. Or, to be more blunt, the reality that other people exist and they have guns and missiles and yes, nuclear weapons. Might still makes right, and I personally would rather not hand over the future of humanity to a person and a company that seems to consistently forget that fact. I do think this post on X from Ramez Naam is the most optimistic way to frame the debate this weekend: I do have tremendous discomfort about AI’s surveillance capabilities in particular; there are a lot of safeguards we thought we had that were actually mostly due to the friction entailed in overcoming them. AI, even more than computers and the Internet, is a friction solvent, and I completely understand why Anthropic’s pushback on this specific point resonates broadly. The way to address this new reality, however, is with new laws and through strengthening accountable oversight; cheering or even demanding that an unelected executive decide how and where such powerful capabilities can be used is the road an even more despotic future. Our adversaries, meanwhile, will certainly be developing autonomous fighting capabilities (and yes, I admit my chip prescriptions make this more likely much sooner — tradeoffs are hard!); the U.S. will need to move in this direction if we are to remain the ultimate source of international law. And, by the U.S., I mean a democratically elected President and Congress, not a San Francisco executive. I don’t want that, and, more pertinently, the ones with guns aren’t going to tolerate it. Anthropic needs to align itself with that reality. Yes, The Hague exists; its subject to all of the same limitations as the United Nations  ↩ Mass domestic surveillance. We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values. AI-driven mass surveillance presents serious, novel risks to our fundamental liberties. To the extent that such surveillance is currently legal, this is only because the law has not yet caught up with the rapidly growing capabilities of AI. For example, under current law, the government can purchase detailed records of Americans’ movements, web browsing, and associations from public sources without obtaining a warrant, a practice the Intelligence Community has acknowledged raises privacy concerns and that has generated bipartisan opposition in Congress. Powerful AI makes it possible to assemble this scattered, individually innocuous data into a comprehensive picture of any person’s life—automatically and at massive scale. Fully autonomous weapons. Partially autonomous weapons, like those used today in Ukraine, are vital to the defense of democracy. Even fully autonomous weapons (those that take humans out of the loop entirely and automate selecting and engaging targets) may prove critical for our national defense. But today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America’s warfighters and civilians at risk. We have offered to work directly with the Department of War on R&D to improve the reliability of these systems, but they have not accepted this offer. In addition, without proper oversight, fully autonomous weapons cannot be relied upon to exercise the critical judgment that our highly trained, professional troops exhibit every day. They need to be deployed with proper guardrails, which don’t exist today. International law is ultimately a function of power; might makes right. There are some categories of capabilities — like nuclear weapons — that are sufficiently powerful to fundamentally affect the U.S.’s freedom of action; we can bomb Iran, but we can’t North Korea. To the extent that AI is on the level of nuclear weapons — or beyond — is the extent that Amodei and Anthropic are building a power base that potentially rivals the U.S. military. Option 1 is that Anthropic accepts a subservient position relative to the U.S. government, and does not seek to retain ultimate decision-making power about how its models are used, instead leaving that to Congress and the President. Option 2 is that the U.S. government either destroys Anthropic or removes Amodei. First, were there only closed AI systems, then unimaginable power would be vested in the owners of those systems; it seems that Amodei thinks that power should be wielded by him (at a minimum, I would prefer that be wielded by the U.S. government). Second, the idea that AI safety can only be guaranteed by a limited number of responsible stewards ignores the massive incentives that exist to build other models. This was clear years ago when only a few companies were working on AI models, and has been proven out by what has happened in reality so far. Third, in a world of AI proliferation, the best defense against AI will be AI; this means that more AI is actually safer than limited AI, which means open source is ultimately safer. Yes, The Hague exists; its subject to all of the same limitations as the United Nations  ↩

business Business
0 views
Jim Nielsen 1 weeks ago

Book Notes: “Blood In The Machine” by Brian Merchant

For my future self, these are a few of my notes from this book . A take from one historian on the Luddite movement: If workmen disliked certain machines, it was because of the use that they were being put, not because they were machines or because they were new Can’t help but think of AI. I don’t worry about AI becoming AGI and subjugating humanity. I worry that it’s put to use consolidating power and wealth into the hands of a few at the expense of many. The Luddites smashed things: to destroy, specifically, ‘machinery hurtful to commonality’ — machinery that tore at the social fabric, unduly benefitting a singly party at the expense of the rest of the community. Those who deploy automation can use it to erode the leverage and earning power of others, to capture for themselves the former earnings of a worker. It’s no wonder CEOs are all about their employees using AI: it gives them the leverage. Respect for the natural rights of humans has been displaced in favor of the unnatural rights of property. Richard Arkwright was an entrepreneur in England. His “innovation” wasn’t the technology for spinning yarn he invented (“pieced together from the inventions of others” would be a better wording), but rather the system of modern factory work he created for putting his machines to work. Arkwright’s “main difficulty”, according to early business theorist Andrew Ure, did not “lie so much in the invention of a proper mechanism for drawing out and twisting cotton into a continuous thread, as in […] training human beings to renounce their desultory habits of work and to identify themselves with the unvarying regularity of the complex automaton.” This was his legacy […] for all his innovation, the secret sauce in his groundbreaking success was labor exploitation. Not much has changed (which is kind of the point of the book). The model for success is: As the author says: [Impose discipline and rigidity on workers, and adapt] them to the rhythms of the machine and the dictates of capital — not the other way around. Reply via: Email · Mastodon · Bluesky Look at the technologies of the day Recognize what works and could turn a profit Steal the ideas and put them into action with unmatched aggression and shamelessness

ai AI books Books business Business
0 views
Allen Pike 2 weeks ago

Launch Now

Inside us are two wolves. One wolf wants to craft, polish and refine – make things of exceptional quality. The other wolf wants to move fast and get feedback now. The two wolves don’t always get along. For years, I’ve balanced this by working toward exceptional products but constantly collecting private feedback along the way. Then, once we’ve built something excellent, something worthy of attention, we launch it to the world with appropriate fanfare. Videos, marketing campaigns, polished onboarding, and so on. “Here’s something worth trying, we think you’ll really like it.” This totally works. At least, it works as a path to eventually ship high-quality software. Polished, usable, even delightful software. But when it comes to building something people will pay for, it’s neither reliable nor fast. Our first product at Forestwalk was a developer tool – a platform for building and running evaluations of LLM-powered apps . We learned a ton building it, but after a few months – as we approached our first pilot projects – feedback from demos and potential first customers convinced us that this was the wrong path. It was more likely to lead us into a lifestyle business than something big. So we pivoted. We spent a few weeks building a prototype a week, showing demos, doing customer research, and found a second promising product path. Our second product was a productivity tool – a work assistant that could capture, organize, and rationalize teams’ tasks . We learned a ton building it, but after a few months – as we approached a public beta – feedback from private testers and our investors convinced us that this was the wrong path. It was more likely to lead us into a lifestyle business than something big. So we pivoted. The third time purports to be the charm. But at the same time, doing the same thing over and over typically gets the same results. We need to build something profoundly useful, something people really want. We can’t keep hiding away, sending out private demos and prototypes, not fully shipping anything! So, we decided to push harder into the discomfort of showing our work early. Just before Christmas, we decided to commit to something and work towards getting it shipped. This third product is codenamed Cedarloop 1 . It’s a realtime meeting agent. Unlike AIs that passively listen in to meetings and just write up notes after the fact, Cedar joins calls and uses “voice in, visuals out” to screen-share useful observations and perform routine tasks live during a Google Meet or Zoom meeting. The vision is to build a kind of agentic PM assistant. It can respond within a second of you talking 2 , which – when it works – feels like magic. We’ve been learning a lot building it. Recently, we started working with an excellent designer here in Vancouver who was keen to get going. I’d like to do some user testing. What do people say when you let them try it? Well, obviously it’s so early right now. They won’t like it. The inference and onboarding need more work. But we’ve been doing research about problems, needs, willingness to pay, and things like that. Sure… but we should also let people try it. What if we launched now? Well, obviously we can’t launch now . I mean… obviously. Launching now would be embarrassing. It’s not my brand to launch something publicly that’s not ready. On the other hand… I keep a printed copy of Y Combinator’s list of essential startup advice on my desk. And if you know YC, you’ll know that the first point of advice is “Launch now”. Only last month I was interviewing Brett Huneycutt, Wealthsimple’s co-founder . He had a lot of great stories, but one that sticks out is that even as a $10B company, they prioritize launching “now”, for as close as they can get to that definition. It’s not just about speed: a rapid feedback loop is a core ingredient in getting to quality. So we launched now. As of today, people can check out our research-preview realtime meeting agent at Cedarloop.ai . With luck, they’ll report issues, inform what we should prioritize next, and tell us what problems they’d love to have automated away. We’re only a few hours in, and yep – people are reporting issues. Linear integration had an OAuth issue. Login didn’t work in social-media webviews. We’ve been so focused on the desktop experience that we’ve let the mobile layout get janky. This is embarrassing! But also, there’s signal. People are trying the Linear integration. Our desktop-focused app is being discovered on mobile. Folks care enough to click at all. And in a week or so, we’ll have a smoother onboarding flow than we would have gotten to with weeks of private user tests. So it’s worth the pain. We’re going to take the feedback, follow the signal, learn and re-learn, and do better. We’ll use it to forge the best damn live agent ever – or, if the feedback peters out, we’ll know we’re on the wrong path, and find the right one. In the meantime, there’s a lot to do. 3 Back to work! This is not a good name yet. For example, sometimes iOS mishears “Hey Cedar” as “Hey Siri”. But part of our move-fast strategy is to worry more about names once we’ve proven something has traction. At that point, we’ll put in the work to give it the right name – and eventually rename the company after it. ↩ It’s fascinating how much you can do to get LLM response times down. Our first prototype often took over 8000ms to respond, which doesn’t feel live at all. Once we got it under ~1200ms, voice-in-vision-out suddenly felt alive – a step change. We have a lot of work planned to get Cedarloop even faster and much more reliable, which I’m keen to write about when I can. ↩ Speaking of having a lot to do: if you’re an experienced product-minded developer in Vancouver who would be excited to iterate and build out realtime agents using LLMs and TypeScript, we’re hiring a Founding Engineer . Just sayin’. ↩ This is not a good name yet. For example, sometimes iOS mishears “Hey Cedar” as “Hey Siri”. But part of our move-fast strategy is to worry more about names once we’ve proven something has traction. At that point, we’ll put in the work to give it the right name – and eventually rename the company after it. ↩ It’s fascinating how much you can do to get LLM response times down. Our first prototype often took over 8000ms to respond, which doesn’t feel live at all. Once we got it under ~1200ms, voice-in-vision-out suddenly felt alive – a step change. We have a lot of work planned to get Cedarloop even faster and much more reliable, which I’m keen to write about when I can. ↩ Speaking of having a lot to do: if you’re an experienced product-minded developer in Vancouver who would be excited to iterate and build out realtime agents using LLMs and TypeScript, we’re hiring a Founding Engineer . Just sayin’. ↩

ai AI typescript TypeScript business Business agile Agile
0 views
Ruslan Osipov 2 weeks ago

Are AI productivity gains fueled by delivery pressure?

A multitudes study which followed 500 developers found an interesting soundbyte: “Engineers merged 27% more PRs with AI - but did 20% more out-of-hours commits”. While I won’t comment on the situation at Google, there are many anecdotes online about folks online who raise concerns about increased work pressure. When a response to “I’m overloaded” becomes “use AI” - we’re heading for unsustainable workloads. The problem is compounded by the fact that AI tools excel at prototyping - the type of work which makes other work happen. Now, your product manager can prototype an idea in a couple of hours, fill it with real (but often incorrect) data, sell the idea to stakeholders, and set goals to productionize it a week later. “Look - the prototype works, and it even uses real data. If I could do this in a couple of hours, how hard could this be for an experienced engineer?” - while I haven’t heard these exact words, the sentiment is widespread (again, online). In a world where AI provides a surface-level ability to contribute across almost any role, the path to avoiding global burnout is to focus on building empathy. Just because an LLM can churn out a document doesn’t mean it’s actually good writing, and we’re certainly not at the point where a handful of agents can replace a seasoned PM. However, because the output looks polished - especially to those without deep domain knowledge - it’s easy to fall into the trap of thinking you’ve done someone else’s job for them. That gap between “looking done” and “being right” is exactly where the extra professional pressure begins to mount. This is really caused by the way we still measure knowledge worker productivity - by the sheer number of artifacts they produce, rather than the outcomes of the work. The right way to leverage AI in workspace is as a license to work better and focus on the right things, not as a mandate to produce more things faster.

business Business ai AI career Career
0 views