Mark Zuckerberg announced Meta Compute, a bet that winning in AI means winning with infrastructure; this, however, means retreating from Reality Labs.
The deal to put Gemini at the heart of Siri is official, and it makes sense for both sides; then Google runs its classic playbook with Universal Commerce Protocol.
A page out of my Mphil project research. This is not a joke.
Dear Apple, I was, given my interest in virtual and augmented reality, already primed to have a high degree of interest in the Vision Pro, but even so, I appreciate how you have gone out of your way to make sure I’m intrigued. You let me try the Vision Pro the day it was announced , and while I purchased my own the day it shipped (and had it flown over to Taiwan), you recently sent me a demo version of the M5 Vision Pro (it’s definitely snappier, although I don’t like the Dual Knit Band at all; the Solo Knit Band continues to fit my head best). However, the reason I truly know you are trying to win my heart is that not only did you finally show a live sporting event in the Vision Pro, and not only was it an NBA basketball game, but the game actually featured my Milwaukee Bucks! Sure, I had to jump through VPN hoops to watch the broadcast, which was only available in the Lakers home market, but who am I to complain about watching Giannis Antetokounmpo seal the game with a block and a steal on LeBron James in my M5 Vision Pro? And yet, complain I shall: you have — like almost every video you have produced for the Vision Pro — once again shown that you fundamentally do not understand the device you are selling. I’m incredibly disappointed, and cannot in good faith recommend any model of the Vision Pro to basketball fans (or anyone else for that matter). Apple, you are one of the grandfather’s of the tech industry at this point; it’s hard to believe that you are turning 50 this year! Still, you are much younger than TV generally, and sports on TV specifically. The first U.S. television broadcast of a sporting event was a Columbia-Princeton baseball game on May 17, 1939 on NBC; there was one camera accompanying the radio announcer. Three months later NBC televised the first Major League Baseball game between the Brooklyn Dodgers and Cincinnati Reds; this time they used two cameras. All televised sports face a fundamental limitation when it comes to the fan experience: the viewer is experiencing something that is happening in real life 3D on a 2D screen; the solution NBC discovered from the very beginning was to not try and recreate the in-person experience, but to instead create something uniquely suited to this new medium. Two cameras became three, then four, then 147 — that’s how many cameras Fox used for last year’s Super Bowl broadcast . Of course many of those cameras were specialized: included in that number were 27 super slow motion cameras, 23 high resolution cameras, 16 robotic cameras, 10 wireless cameras, and two SkyCams. The job of stitching all of those cameras together into one coherent broadcast falls on the production team, housed in a specially equipped truck outside the stadium; that team coordinates with the broadcast booth to provide a seamless experience where every jump feels natural and pre-meditated, even though it’s happening in real time. It’s a great experience! And, of course, there is the pre-game, half-time, and post-game shows, which used an additional 64 cameras, including 12 wireless cameras, eight robotic cameras, seven augmented reality cameras, and a FlyCam. No broadcast is complete without something to fill the time when the game isn’t on. After all, as advanced as TV broadcasts may be, they still face the fundamental limitation that confronted NBC: how do you translate an in-person experience into something that is compelling for people on their couch looking at a 2D screen? When I first tried the Vision Pro the demo included a clip from an NBA game that was later cut from the demo that shipped with the device (which was the one available in Apple Stores); it jumped out at me at the time : What was much more compelling were a series of immersive video experiences that Apple did not show in the keynote. The most striking to me were, unsurprisingly, sports. There was one clip of an NBA basketball game that was incredibly realistic: the game clip was shot from the baseline, and as someone who has had the good fortune to sit courtside, it felt exactly the same, and, it must be said, much more immersive than similar experiences on the Quest. It turns out that one reason for the immersion is that Apple actually created its own cameras to capture the game using its new Apple Immersive Video Format. The company was fairly mum about how it planned to make those cameras and its format more widely available, but I am completely serious when I say that I would pay the NBA thousands of dollars to get a season pass to watch games captured in this way. Yes, that’s a crazy statement to make, but courtside seats cost that much or more, and that 10-second clip was shockingly close to the real thing. What is fascinating is that such a season pass should, in my estimation, look very different from a traditional TV broadcast, what with its multiple camera angles, announcers, scoreboard slug, etc. I wouldn’t want any of that: if I want to see the score, I can simply look up at the scoreboard as if I’m in the stadium; the sounds are provided by the crowd and PA announcer. To put it another way, the Apple Immersive Video Format, to a far greater extent than I thought possible, truly makes you feel like you are in a different place. The first thing that has been frustrating about the Vision Pro has been the overall absence of content; Apple, you produced a number of shows for launch, and then added nothing for months. The pace has picked up a bit, but that has revealed a second frustration: I think that your production stinks! One of the first pieces of sports content that you released was an MLS Season in Review immersive video in March 2024; I wrote in an Update : I have a lot to say about this video and, by extension, the Vision Pro specifically, and Apple generally. Let me work my way up, starting with the video: it’s terrible. The problem — one that was immediately apparent before I got into all of the pedantry below — is that while the format is immersive, the video is not immersive at all. This is the big problem: This is a screenshot of a stopwatch Mac app I downloaded because it supported keyboard shortcuts (and could thus use it while watching the immersive video). There are, in a five minute video, 54 distinct shots; that’s an average of one cut every six seconds! Moreover, there wasn’t that much gameplay: only 2 minutes and 32 seconds. Worse, some of the cuts happen in the same highlight — there was one play where there was a sideline view of the ball being passed up the field, and then it switched to a behind-the-goal view for the goal. I actually missed the goal the first time because I was so discombobulated that it took me a few seconds to even figure out where the ball was. In short, this video was created by a team that had zero understanding of the Vision Pro or why sports fans might be so excited about it. I never got the opportunity to feel like I was at one of these games, because the moment I started to feel the atmosphere or some amount of immersion there was another cut (and frankly, the cuts were so fast that I rarely if ever felt anything). This edit may have been perfect for a traditional 2D-video posted on YouTube; the entire point of immersive video on the Vision Pro, though, is that it is an entirely new kind of experience that requires an entirely new approach. I had the exact same response when you released a video of a Metallica concert last March : As for the concert itself, the video was indeed very cool. The opening shot following James Hetfield walking into the stadium was very compelling, and, well, it was immersive. And then you cut to another camera angle, and while that camera angle was also immersive, the video as a whole no longer was. What followed was a very enjoyable 30 minutes or so — I’ll probably watch it again — but it felt like a particularly neat documentary, not like I was at a concert. You had a monologue from each member of the band, you had shots of the crowd, you had three songs, all, as Apple proudly noted in their press release, shot with “14 Apple Immersive Video cameras using a mix of stabilized cameras, cable-suspended cameras, and remote-controlled camera dolly systems that moved around the stage.” That means the final product was edited together from those 14 cameras and the four interviews, which is to say it was a produced artifact of a live experience; at no point did I feel like I was at the concert. News flash: I didn’t watch the video again. I’m just not that interested in a TV-style documentary of Metallica. I added: We are nearly two years on from that introduction, and over a year beyond the actual launch of the Vision Pro, and there has yet to be an experience like I envisioned and thought was coming. What is frustrating is that the limiting factor is Apple itself: the company had 14 Apple Immersive Video cameras at this concert, but what I want is only one. I want an Apple Immersive Video camera planted in the audience, and the opportunity to experience the concert as if I were there, without an Apple editor deciding what I get to see and when. Needless to say, you probably already know why I thought Friday’s telecast was a big disappointment. I understand, Apple, why it’s not easy to record or even take a screenshot of a copyrighted game; please bear with me while I describe the experience using text. When I started the broadcast I had, surprise surprise, a studio show, specially tailored for the Apple Vision Pro. In other words, there was a dedicated camera, a dedicated presenter, a dedicated graphics team, etc. There was even a dedicated announcing team! This all sounds expensive and special, and I think it was a total waste. Here’s the thing that you don’t seem to get, Apple: the entire reason why the Vision Pro is compelling is because it is not a 2D screen in my living room; it’s an immersive experience I wear on my head. That means that all of the lessons of TV sports production are immaterial. In fact, it’s worse than that: insisting on all of the trappings of a traditional sports broadcast has two big problems: first, because it is costly, it means that less content is available than might be otherwise. And second, it makes the experience significantly worse . Jump ahead to game action. The best camera was this one on the scorer’s table: I have, as I noted, had the good fortune of sitting courtside at an NBA game, and this very much captured the experience. The biggest sensation you get by being close to the players is just how tall and fast and powerful they are, and you got that sensation with the Vision Pro; it was amazing. The problem, however, is that you would be sitting there watching Giannis or LeBron or Luka glide down the court, and suddenly you would be ripped out of the experience because the entirely unnecessary producer decided you should be looking through one of these baseline cameras under the hoop: These are also not bad seats! I’ve had the good fortune of sitting under the basket as well. These are the seats where you really get a sense of not just the power but also the physicality of an NBA game: I would gladly watch an entire game from here. But alas, I was only granted a few seconds, before the camera changed again. This was absolutely maddening — so maddening, that I am devoting a front page Article to a device no one but me cares about, in the desperate attempt to get someone at your company to listen. What makes the Vision Pro unique is the sense of presence: you really feel like you are wherever the Vision Pro takes you. In other words, when I’m wearing the Vision Pro, and the camera actually stays fixed — like, for example, when you set up a special fourth camera specifically for the Lakers Girls performance, which I think was the single longest continual shot in the entire broadcast — I get the sensation of sitting courtside at Crypto.com Arena, and it’s amazing. Suddenly $3,499 feels cheap! However, when I’m getting yanked around from camera to camera, the experience is flat out worse than just watching on TV. Just think about it: would it be enjoyable to be teleported from sideline to baseline to baseline and back again, completely at the whim of some producer, and often in the middle of the play, such that you have to get your bearings to even figure out what is going on? It would be physically uncomfortable — and that’s exactly what it was in the Vision Pro. What is so frustrating is that the right approach is so obvious that I wrote about it the day you announced this device: one camera, with no production. Just let me sit courtside and watch an NBA game. I don’t need a scoreboard, I can look up and see it. I don’t need a pre-game or post-game show, I can simply watch the players warm-up. I don’t need announcers, I’d rather listen to the crowd and the players on the court. You have made a device that, for this specific use case, is better than TV in every way, yet you insist on producing content for it like it is TV! Just stop! There will be more games this year; from your press release last October : Basketball fans will soon be able to experience NBA games like never before in Apple Immersive on Apple Vision Pro, with a selection of live Los Angeles Lakers matchups during the 2025-26 season, courtesy of Spectrum SportsNet. Viewers will feel the intensity of each game as if they were courtside, with perspectives impossible to capture in traditional broadcasts. The schedule of games will be revealed later this fall, with the first game streaming by early next year, available through the forthcoming Spectrum SportsNet app for Vision Pro. That schedule was announced last week , and there are six games total (including last Friday’s). Six! That’s it. I get it, though: producing these games is expensive: you need a dedicated studio host, a dedicated broadcast crew, multiple cameras, a dedicated production crew, and that costs money. Except you don’t need those things at all . All that you need to do, to not just create a good-enough experience but a superior experience, is simply set up the cameras and let me get from the Vision Pro what I can’t get from anything else: the feeling that I am actually there. And, I would add, you shouldn’t stop with the Lakers: there should be Vision Pro cameras at every NBA game, at every NFL game, at every NHL game, at every MLB game — they should be standard issue at every stadium in the world. There should be Vision Pro cameras at every concert hall and convention center. None of these cameras need a dedicated host or announcers or production crew, because the Vision Pro isn’t TV; it’s actual presence, and presence is all you need. $3,499 is a lot of money for physically uncomfortable TV; it’s an absolute bargain if it’s a way to experience any live experience in the world on demand. But, alas, you refuse. So nope, I still can’t recommend the Vision Pro, not because it’s heavy or expensive or has an external battery, but because you, Apple, have no idea what makes it special.
I realised something when I started thinking about my goals for 2026 (as I mentioned I would in my first week notes ) - it’s that I don’t really have any. Not in the way I usually do, anyway. Normally, I like goals. I like SMART goals, systems, structure, clarity, and measurable things. This time, when I genuinely asked myself what I wanted to accomplish this year, nothing specific came up. Apart from losing the three kilos I gained over the last few months, there were no big achievements I wanted to set and measure. It turns out, I don’t need to do anything this year. But I want to be. I want to inhabit my life a little more fully, with a bit more ease and a bit less self-generated pressure. I think this might be a middle-aged thing. I’m turning forty-nine this year (in May!). What eventually surfaced for me was a direction rather than goals. The phrase that resonated with me is “let myself be happier”. I first came across it a long time ago reading an article about The Five Regrets of the Dying. It didn’t resonate at the time at all. I mean, why wouldn’t you let yourself be happier? What kind of weirdness is that? And yet here I am. I don’t let myself be happier, or even happy, very often. I have a long-standing habit of thinking there’s something else to fix or achieve before I’m allowed to be truly happy. But I do know there isn’t . So, as I said, instead of goals, I wrote myself a direction and a set of “do less of this” and “do more of that” intentions. After hours of tweaking, I ended up with a few (4) main areas. The first is letting myself be content. Doing less overthinking, less managing everyone, less getting stuck in negative loops or self-imposed rules about how things should be. More fun. More presence. More horizontal relationships instead of always being the responsible one who holds everything together. This one feels deceptively simple and probably the hardest. The second is moving daily and being kind to my body. Gentle movement. Daily yoga. Walking. Choosing food that actually supports my energy and how I want to feel. I already know what those foods are. I also had to be honest with myself that there is a weight range where I feel physically better, lighter, more like myself. I don’t love that this is true, especially at this age, when metabolism, lifestyle, and the abundance of food everywhere all make this harder than it used to be. But pretending it doesn’t matter, or that I have to accept weight gain as I age, also doesn’t help. The third is going with what genuinely feels good in the moment, not what I think should feel good. Less waiting for the perfect time to enjoy things. More small pleasures now. Taking myself out for walks and coffee. Journaling somewhere I actually enjoy being. Letting myself have solo time without guilt instead of always trying to drag my kids into “adventures” (that may or may not be fun for them) because I hate seeing them glued to screens. More small family trips. More hosting. More bringing people together. And being more intentional about saving for those experiences. The last piece is staying focused at work and continuing to grow professionally. Protecting thinking time . Capturing lessons and mistakes so I actually learn from them. Staying sharp and curious. And careful about getting too busy. It took me much longer than I expected to distill these “goals” into something simple. I had pages of notes, overlapping ideas, and half-formed intentions. At one point I ran the whole thing past my husband, partly to sanity-check myself and partly because he’s very good at cutting through my overthinking. That conversation helped me shorten and clarify what actually mattered, rather than keeping every idea just because I’d spent time thinking about it (and writing it down). Interestingly, he didn’t even think I needed the “stay focused at work and keep growing professionally” piece. His view was that I already do that naturally. I kept it in anyway, because I know how easily time (and focus/attention!) gets eaten by noise and busyness if I don’t actively protect them. At this stage of my life, direction, as opposed to specific goals, feels kinder and more realistic. If I genuinely let myself be happier (more grateful for what I already have? ), not later, not once everything is sorted, but inside ordinary days, that will be enough. If my energy goes a little more into living and a little less into managing, optimising, and worrying, that will be good enough. "Let myself be happier" A year of trusting the flow of life and choosing what genuinely feels good/fun— physically, mentally, and emotionally. Less fretting about what the kids are doing or how they “should” spend their time. Less trying to manage or optimise everyone else. Less complaining and negative narrative loops. Less overthinking. Less self-imposed rules. More separation of tasks and horizontal relationships [[ core ideas in The Courage to Be Disliked ]] More fun - do what’s fun, do what I want to do in the moment. More contentment — being happy with what I already have Eating food that doesn’t make me feel good or support my energy/wellbeing. Ignoring movement when I’m tired or busy. Daily yoga practice (my year-long commitment) [[Keeping a daily yoga practice for a year]]. Regular walking and gentle movement. Choosing food that supports how I want to feel (wfpb). Weight: 65-69 kg range (where I feel my best) Doing what I think I “should” - what others are doing. Waiting for the perfect moment to enjoy life. Solo time (e.g. coffee by myself in the mornings while writing, walking, solo lunches even) Remember that what is fun and fullfilling for me doesn’t have to be “productive”. Small adventures with family or just with Quentin: day trips, weekend road trips. Hosting gatherings and bringing people together. More time with Anna while she still wants the time - shared rituals and little traditions. Save intentionally for bigger travel goals. Taking work home mentally and physically. Deep focus while working. Protect daily thinking time Keep my idea system alive. Capture lessons learned and mistakes. Revisit my five-year direction. Keep learning and staying sharp professionally. [[ actions to take from How to Become CEO ]] The first is letting myself be content. The second is moving daily and being kind to my body. The third is going with what genuinely feels good in the moment, not what I think should feel good. The last piece is staying focused at work and continuing to grow professionally. Less fretting about what the kids are doing or how they “should” spend their time. Less trying to manage or optimise everyone else. Less complaining and negative narrative loops. Less overthinking. Less self-imposed rules. More separation of tasks and horizontal relationships [[ core ideas in The Courage to Be Disliked ]] More fun - do what’s fun, do what I want to do in the moment. More contentment — being happy with what I already have Eating food that doesn’t make me feel good or support my energy/wellbeing. Ignoring movement when I’m tired or busy. Daily yoga practice (my year-long commitment) [[Keeping a daily yoga practice for a year]]. Regular walking and gentle movement. Choosing food that supports how I want to feel (wfpb). Doing what I think I “should” - what others are doing. Waiting for the perfect moment to enjoy life. Solo time (e.g. coffee by myself in the mornings while writing, walking, solo lunches even) Remember that what is fun and fullfilling for me doesn’t have to be “productive”. Small adventures with family or just with Quentin: day trips, weekend road trips. Hosting gatherings and bringing people together. More time with Anna while she still wants the time - shared rituals and little traditions. Save intentionally for bigger travel goals. Taking work home mentally and physically. Deep focus while working. Protect daily thinking time Keep my idea system alive. Capture lessons learned and mistakes. Revisit my five-year direction. Keep learning and staying sharp professionally. [[ actions to take from How to Become CEO ]]
This is a re-post of an article I wrote in 2019 for the Gocode Colorado blog which is no longer available. Thank you Wayback Machine ! “So, we need this feature to work, and it has to tie into this API, and we should put it all on the blockchain.” “What about feature X? And we need admin screens, and roles and groups for different kinds of users.” “You’re right! Let’s add those to the list. We need to make something we’re proud of.” I heard some version of this conversation over and over again at my last Go Code Colorado mentoring session. And I sympathize with the sentiment, I really do. But instead of hitting it out of the park the goal should be to create a piece of software that achieves the bare minimum, or a minimum viable product (MVP). With a high-risk venture team members should aim to show features to the end user as soon as possible, and to let their interactions guide the future of development. It’s far too easy to get distracted by all the possibilities and build features that won’t be used. Even worse, developers may compare their application to other applications they use and find it wanting. The level of polish an MVP needs is far lower than a production application like Gmail, but because you use production ready apps every day, their UX and polish can feel like a requirement. Building features or adding UX polish can delay shipping an MVP. You want to wait until you are “ready”. You are never “ready”, my friend. If you’re not embarrassed by the first version of your product, you’ve launched too late. – Reid Hoffman , LinkedIn Founder Keep your focus not on the software but on the user. Spend time talking to your target market and putting either mockups or working code in front of them as frequently as you can. Finding these people is another blog post entirely, but hopefully, you have some idea who they are and where they hang out. It can be scary to put what you’ve built in front of people. It’s often much easier to sit back and build new features than it is to ship. I have felt it myself–as a startup co-founder, I built a web app that I was, frankly, embarrassed to show potential customers. It was missing features I considered crucial, was full of holes and bugs, and didn’t have a consistent user interface. But showing it to potential customers early and often was the best choice. They pointed out missing features and also explained what was unnecessary. We got great feedback and I was better able to understand the types of problems the customer faced. There are many ways you can show potential users what you are planning to build or are building without having a fully finished product. When building an MVP, use tools you know. Whenever I’m working on a project, I balance technical risk and business risk. If you’re building a true MVP, the business risk is very high, because you don’t know if the market actually exists. Therefore, minimize the technical risk by building with what you know. But wait, aren’t customers expecting a polished application? Some may. Early adopters who are looking to have a problem solved often can look past the rough edges and see the potential. It also depends on the domain and the competition. For instance, if you are starting an Instagram competitor aimed at consumers, the quality bar will be pretty high. If you are building a scheduling tool for tattoo parlors and your main competition is a spreadsheet, a web application built with any modern framework will likely wow your potential customers. It’s also important to show your customers that the application is continuing to improve–that will make them more forgiving of the inevitable issues. You’d be surprised by how forgiving people can be, especially if you are building something to help them do their job better. Remember, if you aren’t a little bit embarrassed when you show someone your application, you should have shown it to them sooner. Additional resources: You can show them mockups, either a paper draft, a series of powerpoint screens or a clickable prototype (Adobe XD or Balsam IQ are solutions). This is the cheapest way to get feedback because changing a screen in powerpoint is far easier than changing a screen in code. Enroll potential customers in a beta program. Customers are more forgiving if they know this isn’t the final product, and they’ll give you suggestions. Don’t take each suggestion as truth, but do try to find out what problem the suggestion is aiming to solve–that’s gold. Offer people in your beta program a discount when you start charging–that gives them the incentive to give good feedback and can seed your customer base. Build out mock features. Instead of building a full file upload facility, I have added a screen to an app with a link to “email us to add a file”, and fulfilled it manually. If enough people mailed a file, we’d know we needed to build the feature. Have someone walk through the application in a video chat (using GoToMeeting or Zoom). Similar to a beta, people are more forgiving when they are shown something and you will be able to see where issues arise (“how do I do task A?”). This experience can be humbling and frustrating at the same time, like watching this . Lean Startup Talk @google Wikipedia on MVPs
Why DRAM shortages, not capital, will define AI infrastructure growth through 2027
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 Netflix and the Hollywood End Game . Will AI Replace Humans? Dorm room discussions are not generally Stratechery’s domain, but for the terminally online it’s getting harder to escape the insistence in some corners of the AI world that humans will no longer be economically necessary, and that we should make changes now to address what they insist is a foregone conclusion. I disagree : humans may be doomed, but as long we are around, we’ll want other humans — and will build economies that reflect that fact. More generally, as we face this new paradigm, it’s essential to remember that technology is an amoral force: what is good or bad comes down to decisions we make, and trying to preemptively engineer our way around human nature guarantees a bad outcome. — Ben Thompson The Future of Power Generation. Through the second half of last year it became clear that one of the defining challenges of the modern era in the U.S. will be related to power: we need more than we can generate right now, electrical bills are skyrocketing, and that tension figures to compound as AI becomes more integral to the economy. If these topics interest you (and they should!), I heartily recommend this week’s Stratechery Interview with Jeremie Eliahou Ontiveros and Ajey Pandey , two SemiAnalysis analysts who provide a rundown on what the biggest AI labs are doing to address these challenges, the important of natural gas, and how the market is responding to the most fundamental AI infrastructure challenge of them all. Reminder on that last point: Bubbles have benefits ! — Andrew Sharp What China Thinks of What Happened in Caracas. On this week’s Sharp China, Bill Bishop and I returned from a holiday that was busier than expected and broke down various aspects of China’s response to the upheaval in Venezuela after the CCP’s “all-weather strategic partnership” with the Nicolás Maduro regime was rendered null and void by the United States. Topics include: A likely unchanged calculus on Taiwan, a propaganda gift, questions about oil imports, and Iran looming as an even bigger wild card for PRC fortunes. Related to all this, on Sharp Text this week I wrote about the logic of the Maduro operation on the American side, and the dizzying nature of decoding U.S. foreign policy in a modern era increasingly defined by cold war objectives but without cold war rhetoric. — AS AI and the Human Condition — AI might replace all of the jobs; that’s only a problem if you think that humans will care, but if they care, they will create new jobs. Nvidia and Groq, A Stinkily Brilliant Deal, Why This Deal Makes Sense — Nvidia is licensing Groq’s technology and hiring most of its employees; it’s the most potent application of tech’s don’t-call-it-an-acquisition deal model yet. Nvidia at CES, Vera Rubin and AI-Native Storage Infrastructure, Alpamayo — Nvidia’s CES announcements didn’t have much for consumers, but affects them all the same. An Interview with Jeremie Eliahou Ontiveros and Ajey Pandey About Building Power for AI — An Interview with Jeremie Eliahou Ontiveros and Ajey Pandey about how AI labs and hyperscalers are leveraging demand to build out entirely new electrical infrastructure for AI. Notes from Schrödinger’s Cold War — Why the U.S. captured Nicolás Maduro, and the challenge of decoding U.S. foreign policy in an era defined by cold war objectives, but without cold war rhetoric. CES and Humanity Tahoe Icons High-Five to the Belgrade Hand The Remarkable Computers Built Not to Fail China’s Venezuela Calculations; Japan’s Rare Earth Access; A Reported Pause on Nvidia Purchases; The Meta-Manus Deal Under Review A New Year’s MVP Ballot, The Jokic Injury and a Jaylen Brown Renaissance, The Pistons and Their Possibilities The Economy in the 22nd Century, Amoral Tech and Silicon Valley Micro-Culture, What Nvidia Is Getting From Groq
An Interview with Jeremie Eliahou Ontiveros and Ajey Pandey about how AI labs and hyperscalers are leveraging demand to build out entirely new electrical infrastructure for AI.
We have a housing shortage in the U.S., and it is arguably a major cause of long-term unrest about the economy. Putting aside whether AI will eliminate jobs on net, it will certainly displace a lot of them. And the displaced people are unlikely to be the same people who will secure the higher-tech jobs that get created. For example, are most displaced truck drivers going to get jobs in new industries that require a lot of education? Put these two problems together and maybe there is a solution hiding in plain sight: create millions of new jobs in housing. Someone has to build all the affordable homes we need, so why not subsidize jobs and training for those displaced by AI? These jobs will arguably offer an easier onramp and are sorely needed now (and likely for the next couple of decades as we chip away at this housing shortage). Granted, labor may not be the primary bottleneck in the housing shortage, but it is certainly a factor and one that is seemingly being overlooked. There are many bills in Congress aimed at increasing housing supply through new financing and relaxed regulatory frameworks. A program like this would help complete the package. None of this has been happening via market forces alone, so the government would therefore need to create a new program at a large scale, like the Works Progress Administration (WPA) at the end of the Great Depression, but this time squarely focused on affordable housing (and otherwise narrowly tailored to avoid inefficiencies). There are a lot of ways such a program could work (or not work), including ways to maximize the long-term public benefit (and minimize its long-term public cost), but this post is just about floating the high-level idea. So there you have it. I’ll leave you though with a few more specific thought starters: Every state could benefit since every state has affordable housing issues. Programs become more politically viable when more states benefit from them. Such a program could be narrowly tailored, squarely focused on affordable housing (as mentioned above), but also keeping the jobs time-limited (the whole program could be time-limited and tied to overall housing stock), and keeping the wages slightly below local market rates (to complement rather than compete with private construction). It could also be tailored to those just affected by AI, but that doesn’t seem like the right approach to me. The AI job market impact timeline is unclear, but we can nevertheless start an affordable-housing jobs program now that we need today, which can also serve as a partial backstop for AI-job fallout tomorrow. It seems fine to me if some workers who join aren't directly displaced by AI, since the program still creates net new jobs we will need anyway and to some extent jobs within an education band are fungible. We will surely need other programs as well to help displaced workers specifically (for example, increased unemployment benefits). Thanks for reading! Subscribe for free to receive new posts or get the audio version . We have a housing shortage in the U.S., and it is arguably a major cause of long-term unrest about the economy. Putting aside whether AI will eliminate jobs on net, it will certainly displace a lot of them. And the displaced people are unlikely to be the same people who will secure the higher-tech jobs that get created. For example, are most displaced truck drivers going to get jobs in new industries that require a lot of education? Put these two problems together and maybe there is a solution hiding in plain sight: create millions of new jobs in housing. Someone has to build all the affordable homes we need, so why not subsidize jobs and training for those displaced by AI? These jobs will arguably offer an easier onramp and are sorely needed now (and likely for the next couple of decades as we chip away at this housing shortage). Granted, labor may not be the primary bottleneck in the housing shortage, but it is certainly a factor and one that is seemingly being overlooked. There are many bills in Congress aimed at increasing housing supply through new financing and relaxed regulatory frameworks. A program like this would help complete the package. None of this has been happening via market forces alone, so the government would therefore need to create a new program at a large scale, like the Works Progress Administration (WPA) at the end of the Great Depression, but this time squarely focused on affordable housing (and otherwise narrowly tailored to avoid inefficiencies). There are a lot of ways such a program could work (or not work), including ways to maximize the long-term public benefit (and minimize its long-term public cost), but this post is just about floating the high-level idea. So there you have it. I’ll leave you though with a few more specific thought starters: Every state could benefit since every state has affordable housing issues. Programs become more politically viable when more states benefit from them. Such a program could be narrowly tailored, squarely focused on affordable housing (as mentioned above), but also keeping the jobs time-limited (the whole program could be time-limited and tied to overall housing stock), and keeping the wages slightly below local market rates (to complement rather than compete with private construction). It could also be tailored to those just affected by AI, but that doesn’t seem like the right approach to me. The AI job market impact timeline is unclear, but we can nevertheless start an affordable-housing jobs program now that we need today, which can also serve as a partial backstop for AI-job fallout tomorrow. It seems fine to me if some workers who join aren't directly displaced by AI, since the program still creates net new jobs we will need anyway and to some extent jobs within an education band are fungible. We will surely need other programs as well to help displaced workers specifically (for example, increased unemployment benefits).
☕ Welcome to The Coder Cafe! I sucked at product management 1 . Early in my career, I was only passionate about how a product works under the hood, not about what the product actually does. Over time, I began to change and open up. Today, I want to share a concept from a book recommended on X by that I really loved: The Cold Start Problem . Get cozy, grab a coffee, and let’s begin! Network Effects Let’s consider a dating app. If there are only three people who installed the app in New-York, anyone new will probably try it for a few seconds and uninstall it. But if a big part of the city is on it, then someone single will probably stick around. Said differently, the more people use the product, the more valuable it becomes. There’s a term for that type of product, and it’s called the network effect: when a product gets more valuable as more people use it . Delving into the network effect, it’s not a single force but actually a trio of forces: The acquisition effect : How a product can use its own network to attract new people. The engagement effect : The more people join, the more useful and sticky it becomes. The economic effect : A larger network reduces costs, improves monetization, and strengthens the business model. Now comes the real challenge. If our product relies on network effects, how do we launch it? Do we start from zero, or wait until enough people are on board? It’s a chicken-and-egg problem, and that’s the cold start problem. One common mistake to solve the cold start problem is the big bang launch: releasing to everyone before any community exists. Google+ is the perfect example. Launched in 2011, it was Google’s attempt at a social network with a Facebook-style feed. The problem was that when people joined, they found empty timelines and left. Google later admitted that 90% of user sessions lasted less than five seconds. At one point, Google even tied YouTube comments to Google+, requiring an account just to comment. The platform eventually reached more than 500 million users, but the issue was never sign-ups. The real problem was that a newcomer’s first session didn’t feel like walking into a lively room. It was forced growth instead of real networks. Google+ didn’t fail because Google couldn’t build a social network. It lost because it never created a place where a new user could land in a live network. In short, Google+ failed to solve the cold start problem. A solution to the cold start problem was applied by Tinder, and it involves focusing on the concept of atomic networks. Back then, dating apps were not very popular. Yet, Tinder was ambitious and wanted to succeed in that market. Instead of launching worldwide, they did the complete opposite. They organized a party at a college that required installing the application to attend. The next day, most students there had Tinder installed. This college was an atomic network: the smallest self-sustaining cluster of users where network effects actually work. Soon after, they repeated the same process in another college in the same city, with the same result: within days, most students there had joined Tinder. What’s powerful about atomic networks is simple: if we can build one network, we can build two. If we can build two, we can build thousands. Tinder repeated this strategy college by college, then city by city, eventually growing into entire countries. The takeaway is that when we start a product with network effects, the first step is to build a single, tiny network that’s self-sustaining on its own. We just want to get started. If we can create one stable, engaged network, then we can build a second one next to it. From there, we can replicate the process and eventually connect them into one large network that spans the whole market. Network effects happen when a product gets more valuable as more people use it. How do we launch such a product with zero users? That’s the cold start problem. Big bang launches fail when newcomers find empty networks. One effective approach is to build atomic networks: the smallest self-sustaining clusters where network effects work. If we can build one atomic network, we can repeat it and scale across a market. Missing direction in your tech career? At The Coder Cafe, we serve timeless concepts with your coffee to help you master the fundamentals. Written by a Google SWE and trusted by thousands of readers, we support your growth as an engineer, one coffee at a time. Don’t Forget About Your Mental Health The XY Problem Lateral Thinking The Cold Start Problem Project Strobe: Protecting your data, improving our third-party APIs, and sunsetting consumer Google+ Google+: Communities and photos ❤️ If you enjoyed the post, please consider giving it a like. 💬 The book is really great and covers many more aspects. I definitely recommend it. Are you into product management? What resources would you recommend? Leave a comment Right, Afroditi? The acquisition effect : How a product can use its own network to attract new people. The engagement effect : The more people join, the more useful and sticky it becomes. The economic effect : A larger network reduces costs, improves monetization, and strengthens the business model. Network effects happen when a product gets more valuable as more people use it. How do we launch such a product with zero users? That’s the cold start problem. Big bang launches fail when newcomers find empty networks. One effective approach is to build atomic networks: the smallest self-sustaining clusters where network effects work. If we can build one atomic network, we can repeat it and scale across a market. Don’t Forget About Your Mental Health The XY Problem Lateral Thinking The Cold Start Problem Project Strobe: Protecting your data, improving our third-party APIs, and sunsetting consumer Google+ Google+: Communities and photos
Nvidia's CES announcements didn't have much for consumers, but affects them all the same.
AI workflows are technically impressive, but there’s a deeper reason people are really amped about AI agents. This isn’t just new tech, it’s new psychology. Until now, very few people have known what it feels like to delegate to total competency. If you manage great people, or lead great teams, you know how it feels to put someone in charge who will get it done, get it done right, and get it done without drama. That kind of delegation — that depth of trust — is pure joy. Delegating to competency lets you forget about it completely. That’s real leverage. And now anyone can experience that. What was rare is now widely distributed. Everyone can feel it. And it feels fucking great. That’s a big reason why the excitement is real, and fully justified. -Jason
Nvidia is licensing Groq's technology and hiring most of its employees; it's the most potent application of tech's don't-call-it-an-acquisition deal model yet.
Over the years, I've met and observed teams that were competent and yet failed: working on the wrong priorities, sometimes working on the right objectives, but with unsuitable solutions, sometimes working with no priorities at all. The root cause of these problems can be articulated around these 3 words: autonomy, alignment, and context . If these words seem relatively simple to understand, in practice, their implementation is not trivial. How do you ensure that everyone in a team is trying to solve the same problems? What does autonomy mean within a group, and what are its limits? Perhaps you've heard the term "empowered teams"? And thought it was yet another new consultant concept? The expression gained popularity via the book Empowered by Marty Cagan . In simple terms, an empowered team works on a problem to be solved, as opposed to a feature team which works on a list of functionalities (the famous roadmap). We also speak of a team that prioritizes outcomes (impacts) over outputs (deliverables). The first thing I'd like to stress is that this isn't just a topic for product managers. Any senior person in Engineering should be involved in decision-making, for example : I'm writing this chapter for any technical leader looking to create impact. To do this, we'll see that creating autonomy, alignment and knowing how to communicate the context are levers for achieving this. Here, as a Tech Leader you can be in two situations, but often in both at the same time: To tackle this subject, I'm going to start from a drawing by Henrik Kniberg about alignment and empowerment. "Empowered teams" being the teams in the top right-hand corner. This drawing is part of a keynote given in 2016 that I invite you to watch. In this keynote, it's about a group, whose goal is to cross a river to settle on the other side of the bank. The lower-left corner shows a group with little autonomy, no alignment and no precise directions. No one in this part of the quadrant has taken the initial problem of crossing the river personally. The symptoms of such an organization in a technological context : What can be done about it? It's a tough job on this part of the quadrant. Even before talking about autonomy, the first issue will be to get people to agree on a common goal. The role of a tech leader is to get close to the business, or its closest representative, and understand the issues that need to be addressed. In this type of organization, the business is often far away and/or difficult to access. Breaking the distance can be difficult, and is not part of the company's culture. Technique and "best practices" are likely to be the least of your worries in this context, as the importance of connecting Engineering to the business is paramount. If we move to the right, we come across organizations made up of individuals or groups of individuals, each of whom has defined objectives, but all of which are different. If the group's objective is to cross a river, everyone is working on different things: some are growing vegetables, others are fishing, and a few leaders are wondering whether someone is working on crossing the river. There's a misunderstanding of autonomy here. Autonomy does not mean independence. A team that decides its priorities in its own corner doesn't make an impact, or does so by accident, and this success is hard to repeat many times over. And this doesn't mean that individual groups can't be highly effective. Unfortunately, impact is measured in global terms. The symptoms of such an organization in a technological context: What can be done about it? The observation is fairly similar to the previous one, but the work will be less arduous due to the highly entrepreneurial nature of the company. Here, the tech leader will be able to rely on committed teams. We'll need to work on alignment. But let's define this notion a little better. Alignment defines the ability of a group to all seek solutions to the same problem. Objective: A European company decides to enter the US market. Each team will seek to contribute to the same objective. However, in the drawing in the upper left corner, we can see a clear alignment without autonomy. One or more leaders can decide exactly what needs to be done to achieve an objective. The leader explicitly asks to build a bridge across the river. This mode of operation can be very effective, especially on a small scale with a visionary leader. But scaling up can cause problems, as the visionary leader can't spread out across the company and follow all the decisions. On several occasions, I've observed companies with very good leaders, but with great difficulty in managing rapid growth. They recruited people who were more junior than they were, more capable of carrying out a task than making decisions. These people then failed to scale up in the time available. Symptoms of such an organization in a technological context: What to do about it: If you're a tech leader, one of your responsibilities is recruitment. You have to recruit people who are better than you in certain areas. You have to learn to share your legos . The second important thing is to get out of the vicious circle: At stage 2, you have a major responsibility for coaching/mentoring to educate and provide the necessary context for more effective decision-making. If you're a tech leader, but you're in a situation where the work is delegated to you, you'll need to show that the team can be autonomous. To do this, two things are important: In our example above, a group that chose to build a catapult to hurl individuals to the other side would be showing a certain lack of understanding of the effect of gravity on human bodies :) Hopefully, the initiative would be stopped. Don't be the one to suggest the catapult . Bad: there are too few of us in the team, so we can't finish the project on time. Good: given the size of our team and our current speed, I've decided to cut back on some of the functionalities to meet the deadline. Bad: We're drowning in support. The team can't deliver project A on time. Good: We carried out a support analysis and noted a sharp increase in support cases linked to a bug in a recent feature. We decided to assign one person for a week to fix this bug and regain velocity on project A. Before I talk about the last quadrant, I need to talk a little about the "context". "Lead with context, not control" is a quote you'll find in the excellent book "No rules rules" by Erin Meyer and Reed Hastings. For a person or a group to make a good decision, they need to know the whole context. If we go back to our group, which has to cross a river, we lack the context to propose initiatives. Maybe we have to cross the river to find food. In which case, why didn't we decide to settle on this side? The leader failed to mention that we're in the territory of another tribe who has asked us to leave before autumn. Okay, and if the leader had told us that this tribe wasn't against trading with us after all, having a bridge would probably be more interesting in the future than just building a boat. Having the context allows you to make the right decisions. And in the corporate world, there are plenty of occasions when you neglect to give the full context, sometimes for the sake of speed, sometimes to protect your team, sometimes because you've forgotten. But without context, you can't expect people to make the right decisions. Context gives you the constraints that will enable you to propose the most appropriate technological solution. In this corner of the quadrant, the objective is clearly defined, such as crossing the river. The constraints are known, and each group is invited to contribute to the common goal. In theory, this is the most efficient type of organization, especially at scale. I'd like to point out two pitfalls at this stage: It doesn't make sense to hire smart people and tell them what to do; we hire smart people so they can tell us what to do. You have to accept the unexpected, accept that the solutions you come up with may not be the ones you originally imagined. If a group suggests crossing the river with a boat or a tunnel, or has spotted a ford a hundred meters downstream, it may be different from what you had in mind, but it may be more effective. Accept that not everything is as it seems in your head. BUT, if there are several groups working, one on building a tunnel, another on building a bridge, another on building a boat, you've got a problem. If everyone is working on the same problem with different solutions, that's also a form of misalignment . Again, autonomy doesn't mean independence, and there needs to be regular discussion around the initiatives launched to ensure overall coherence. Don't be the one working on creating a tunnel while everyone else is trying to build a bridge. Certain methodologies, such as OKRs for example, help to materialize this periodic discussion. Objective: The product should enable the acquisition of 10,000 prospects in the USA. Quarter 1 initiatives: Quarter 2 initiatives: The coherence of the initiatives is linked to a discussion at the beginning of the quarter of what each group has come up with to contribute to the initial lead acquisition objective. This step, designed to ensure the coherence of initiatives, can be found in many methodologies: I must stress once again that autonomy does not mean independence. If, during these discussions, a decision is made to favor a bet, and it's not the solution you had in mind, too bad but you have to commit to this decision. This is the famous "disagree and commit" . And that doesn't take anything away from team autonomy. Teams can be the driving force behind initiatives to solve a given problem, but in the end, we'll be looking for alignment. Otherwise, we're back in the bottom right quadrant. If your aim is to create more autonomous product teams, you may come up against the following difficulties: These 3 subjects create deadlocks. A good PM and/or Tech Leader will be reluctant to join an organization where he/she feels a lack of autonomy. Management will find it hard to trust a team that they don't feel is business-oriented enough. Top management who don't have confidence in a team will continue to demand a high degree of predictability in development via detailed roadmaps. As a Tech Leader, you have your hands on several levers that I've already covered in the first chapters: The aim of all these actions is to increase the trust placed in Engineering's actions. This confidence will be transformed into the delegation of autonomy . Would you say that you have all the constraints in mind for your current projects? How much time do you spend explaining the context to the people who work with you? What is your team/department's objective for the year? How are you contributing to this objective this quarter? an empowered team's mission is to work on acquiring new users a "feature team" has to come up with a sponsorship program an "empowered team" is tasked with improving the developer experience by reducing dev cycle times a feature team is tasked with implementing a new CI (continuous integration) solution. avoiding the construction of unnecessarily complex solutions, demonstrating a misunderstanding of objectives checking that these priorities are built with the right constraints in mind ensuring that these priorities contribute to the group's objectives. the person seeking to develop his or her group's autonomy the person who belongs to a group with limited autonomy, and is seeking to gain independence. reaction-only teams, i.e. whose work is driven by support and bugs, upgrades to end-of-life components, and so on. no hyper-clear objectives, but a kanban (a big todo list) that feeds itself organically micromanagement, with tasks assigned by a manager/project leader a technical team that works in isolation from users, often in supplier mode for a team that defines the product, but itself far from the business. several groups work on defined topics, but without collaboration with other teams. a large number of projects are underway, stretching out over months. This is linked to the "one-man army" phenomenon: if a single person can move fast, he or she ends up burning out faster than a larger group. This slowness makes teams weary, as they continue to pile new subjects on top of the list. the multiplicity of topics creates vagueness throughout the organization, which becomes a black box for the rest of the company the perceived impact is low The sales department will invest to open a sales office in the USA. The finance department will open a subsidiary in the USA and modify its invoicing model to comply with current regulations. The engineering department will study how to reduce product latency through a better geographical distribution of its infrastructure the leader is the most experienced member of the group, with a significant gap between them there is a strong imbalance between junior and senior staff, to the detriment of the latter the team is not involved in decision-making, and the backlog is fed by a single person the people I work with don't make the right decisions or don't know how to make decisions, so I step in to make them for them for the sake of speed I don't spend much time on the explanation the people in question end up concentrating more on their expertise than on decision-making. Develop your business understanding . You need to step outside the team to better understand the challenges and constraints of the company and/or your customers. Develop your decision-making skills. To re-establish the symmetry of exchanges with your management, you need to keep one thing in mind: you need to provide more solutions than problems . You won't be trusted if you only come up with problems to solve, and your discussions will systematically turn into basic reporting, so you'll have to make choices. Team A works on improving SEO through multi-domain management and regionalization. Team B sets up several new domain names for multi-domain management Team C works on SEA to refine geographic targeting of campaigns. Team A works on a sponsorship program Team B works on sharing mechanisms on social networks Team C works on affiliation mechanisms The objective is defined in advance. Teams propose initiatives to contribute to it After discussion, a set of initiatives is chosen. in OKR via the OKR Quarterly review fixing appetites in the ShapeUp method the construction of "bet boards" by Henrik Kniberg Program Increment Planning in SAFe recruitment. To get started, you need very good PMs and Tech leaders who have an entrepreneurial mindset as well as a technical one. Transformation is all about people. top management buy-in and a change of mindset are essential. They must agree to delegate the "how" to product teams, and focus on the "what" (objectives). And that's a lot harder than it sounds. Many say they want to move to autonomous product teams, but are very frustrated at not being able to tell them what to do agreeing to abandon traditional roadmaps in favor of strategies based on objectives (OKR or other, it doesn't matter). Develop your understanding of the business Measuring everything Know how to prioritize Create time for ideation Know how to communicate Give away your legos and other commandments for scaling startups No rules rules keynote from Henrik Kniberg en 2016 Empowered from Marty Cagan
Almost a year ago, I found that N100 Mini PCs were cheaper than a decked-out Raspberry Pi 5 . So comparing systems with: Back in March last year, a GMKtec Mini PC was $159, and a similar-spec Pi 5 was $208. Today? The same GMKtec Mini PC is $246.99, and the same Pi 5 is $246.95: Today, because of the wonderful RAM shortages 1 , the Mini PC is the same price as a fully kitted-out Raspberry Pi 5. 16GB of RAM 512GB NVMe SSD Including case, cooler, and power adapter
Listen to this post : Pity the paradox of the content producer in the age of AI. On one hand, AI is one of the greatest gifts ever in terms of topics to cover. The 2025 Stratechery Year in Review was, just like 2024 and 2023 (plus a few bangers in 2022 ) completely dominated by AI; my Sharp Tech co-host Andrew Sharp wrote The Definitive Ranking of Tech Company Takeability , and OpenAI was number one with a bullet: OpenAI may or may not be the most important company of the future. There can be no doubt, however, that we are witnessing one of the most takeable enterprises in the history of the world. From the day it was founded — with a non-profit corporate structure that sought to build AGI and then control it themselves “to ensure artificial general intelligence benefits all of humanity” — this company has divided the audience and invited either passionate support or aggressive eye-rolls. On the other hand, LLMs in particular are quite literally content producers! What’s the point of writing analysis when ChatGPT or Gemini or Claude will deliver analysis on demand, about any topic you want? Is this one of those situations like the early web, where the possibility of reaching everyone seemed like a boon but was actually a ticking time bomb for the viability of the traditional publishing model ? I’m actually pretty optimistic about Stratechery Plus’ fortunes for reasons I laid out in last year’s Content and Community : So, are existing publishers doomed? Well, by-and-large yes, but that’s because they have been doomed for a long time. People using AI instead of Google — or Google using AI to provide answers above links — make the long-term outlook for advertising-based publishers worse, but that’s an acceleration of a demise that has been in motion for a long time. What I think is intriguing, however, is the possibility to go back to the future. Once upon a time publishing made countries; the new opportunity for publishing is to make communities. This is something that AI, particularly as it manifests today, is fundamentally unsuited to: all of that content generated by LLMs is individualized; what you ask, and what the AI answers, is distinct from what I ask, and what answers I receive. This is great for getting things done, but it’s useless for creating common ground… Stratechery, on the other hand, along with a host of other successful publications, has the potential to be a totem pole around which communities can form…There is a need for community, and I think content, whether it be an essay, a podcast, or a video, can be artifacts around which communities can form and sustain themselves, ultimately to the economic benefit of the content creator. There is, admittedly, a lot to figure out in terms of that last piece, but when you remember that content made countries, the potential upside is likely quite large indeed. This might, to be fair, be wishful thinking: maybe I’m doomed, but if I’m doomed, probably everyone else is too, particularly when you think about the very long run. It’s the very long run that a fellow content producer, Dwarkesh Patel, considered alongside Philip Trammell in a widely discussed post over the winter break entitled Capital in the 22nd Century : In his 2013 Capital in the Twenty-first Century , the socialist economist Thomas Piketty argued that, absent strong redistribution, economic inequality tends to increase indefinitely through the generations — at least until shocks, like large wars or prodigal sons, reset the clock. This is because the rich tend to save more than the poor and because they can get higher returns on their investments. As many noted at the time, this is probably an incorrect account of the past. Labor and capital complement each other. Wealthy people can keep accumulating capital, but hammers grow less valuable when there aren’t enough hands to use all of them, and hands grow more valuable when hammers are plentiful. Capital accumulation thus lowers interest rates (aka income per unit of capital) and raises wages (income per unit of labor). This effect has tended to be strong enough that, though inequality may have grown for other reasons, inequality from capital accumulation alone has been self-correcting. But in a world of advanced robotics and AI, this correction mechanism will break. That is, though Piketty was wrong about the past, he will probably be right about the future…If AI is used to lock in a more stable world, or at least one in which ancestors can more fully control the wealth they leave to their descendants (let alone one in which they never die), the clock-resetting shocks could disappear. Assuming the rich do not become unprecedentedly philanthropic, a global and highly progressive tax on capital (or at least capital income) will then indeed be essentially the only way to prevent inequality from growing extreme. Without one, once AI renders capital a true substitute for labor, approximately everything will eventually belong to those who are wealthiest when the transition occurs, or their heirs. Or more precisely, it will belong to the subset of this group who save most and most invest with a view to maximizing long-run returns. There is an aspect to this argument that is of the dorm room discussion variety: even if we are approaching the point where AI can create AI (Claude Opus 4.5 appears to be a major leap forward in coding capability in particular), there is still a lot of work to do in terms of making it possible for AI to break out of its digital box and impact the real world via robotics. Part of the thinking, however, is that once AI can create AI, it can rapidly accelerate the development of robotics as well, until robots are making robots, each generation more capable than the next, until everything humans do today — both in the digital but also the physical world — can be done better by AI. This is the world where capital drives all value, and labor none, in stark contrast to the approximately 33% share of GDP that has traditionally gone to capital, with 66% share of GDP going to labor. After all, you don’t pay robots for marginal labor: you build them once…check that, they build themselves, from materials they harvested, not just here on earth but across the galaxy, and do everything at zero marginal cost, a rate with which no human can compete. I get the logic of Patel and Trammell’s argument, but I — perhaps, once again, over-optimistically — am skeptical about this being a problem, particularly one that needs to be addressed right here right now before the AI takeoff occurs, especially given the acute need for more capital investment at this moment in time. First, the world Patel and Trammell envisions sounds like it would be pretty incredible for everyone. If AI can do everything, then it follows that everyone can have everything, from food and clothing to every service you can imagine (remember, the AI is so good that there are zero jobs for humans, which implies that all of the jobs can be done by robots for everyone). Does it matter if you don’t personally own the robots if every material desire is already met? Second, on the flipside, this world also sounds implausible. It seems odd that AI would acquire such fantastic capabilities and yet still be controlled by humans and governed by property laws as commonly understood in 2025. I find the AI doomsday scenario — where this uber-capable AI is no longer controllable by humans — to be more realistic; on the flipside, if we start moving down this path of abundance, I would expect our collective understanding of property rights to shift considerably. Third, it’s worth noting that we have seen dramatic shifts in labor in human history. Consider both agricultural revolutions: in the pre-Neolithic era zero percent of humans worked in agriculture; fast-forward to 1810, and 81% of the U.S. population worked in agriculture. Then came the second agriculture revolution, such that 200 years later only 1% of the U.S. population works in agriculture. It’s that decline that is particularly interesting to me: humans were replaced by machines, even as food became abundant and dramatically cheaper; no one is measuring their purchases based on how much food cost in 1700, just as they won’t measure their future purchases on the cost of material goods in a pre-robotics world. That’s because humans didn’t just sit on their hands; rather, entirely new kinds of work were created, which were valued dramatically higher. Much of this was in factories, and then, over the last century, there was the rise of office work. All of that could very well be replaced by AI, but the point is that the history of humans is the continual creation of new jobs to be done — jobs that couldn’t have been conceived of before they were obvious, and which pay dramatically more than whatever baseline existed before technological change. Like, if I might be cheeky, professional podcaster! Podcasts didn’t even exist thirty years ago, and yet here is Patel — and me! — accumulating capital simply by speaking into a mic and taking advantage of the Internet’s zero marginal cost of distribution, a concept that itself was unthinkable fifty years ago. It’s possible, of course — and to return to my perhaps self-interested and potentially misplaced optimism above — that robots will be better at podcasting than Patel or I. I’m skeptical, though: my experience — and I’ll only speak for myself here — is that the human element is essential in creating compelling content. Sure, sometimes I say “uhm” or “like” or “sort of”, or I get facts wrong, but that’s a feature, not a bug: what I have to say is by definition unique to me, and that is interesting precisely because I am flesh-and-blood, not a robot. 1 Indeed, another way to frame the optimism I have around my career is that the dynamics are the exact inverse of AI: Right now it is individual humans who are uniquely capable to reach audiences at scale; AI, on the other hand, is about scaling compute to deliver results to individuals. Patel and Trammell were, to be sure, talking about the 22nd century, while this is a depiction of the first quarter of the 21st, but I think the desire for a communal experience will persist, and I think those experiences will continue to be organized around other humans, not machines. More generally, I don’t think that this will be limited to podcasting (if such a concept even exists in one hundred years). Consider the most base example: sex. I have no doubt that there will be human-like robots with which you can have sex; I also have even stronger conviction that the overwhelming preference of humans will be to have sex with other humans. And that, by extension, means that all of the courtship and status games that go into finding a lover will persist, and that that itself will be an entire economy all its own. One will not impress a partner with commodity robot-generated goods, no matter how objectively perfect they might be: true value will come from uniqueness and imperfections that are downstream from a human. In fact, I have great optimism that one potential upside of AI is a renewed appreciation of and investment in beauty. One of the great tragedies of the industrial era — particularly today — is that beauty in our built environment is nowhere to be found. How is it that we built intricate cathedrals hundreds of years ago, and forgettable cookie-cutter crap today? That is, in fact, another labor story: before the industrial revolution labor was abundant and cheap, which meant it was defensible to devote thousands of person-years into intricate buildings; once labor was made more productive, and thus more valuable, it simply wasn’t financially viable to divert so much talent for so much time. Perhaps it follows, then, that the devaluing of labor Patel and Trammell warns about actually frees humans up to once again create beauty? Yes, robots could do it too, but I think humans will value the work of other humans more. Indeed, I think this is coming sooner than you might think: I expect the widespread availability of high quality AI art to actually make human art more desirable and valuable, precisely because of its provenance. 2 It’s also worth noting the relative popularity of human-generated content versus AI-generated content. Sora is down to 59 in the App Store, and I count double-digit human-denominated social apps that rank above it. Yes, I get the argument that this is the worst that AI will ever be, but it also will never be human, which is what humans want most of all. This gets at what I found the most frustrating about Patel and Trammell’s point of view: the core assumption undergirding their argument was also about the human condition; it just happened to be negative. Louis C.K., in an October 2008 appearance on Late Night with Conan O’Brien , delivered one of the most incisive and eternal observations about human nature: “Everything is amazing right now and nobody’s happy.” You’ve almost certainly seen this clip, but if not, it’s worth watching in full; Louis C.K. focuses on three incredible technological innovations and how quickly we took them for granted: smartphones, Internet access on planes, and the act of flying itself. It’s certainly a sentiment I can relate to: just in the last 72 hours I have chafed at slow airplane WiFi, complained about jet lag from having literally traversed the globe, and gotten frustrated at an iPhone bug that is sapping my battery. It’s all so terrible, until I remember I have access to anything everywhere, can be anywhere anytime, and oh yeah, can achieve both simultaneously. If anything, you can make the case that technological innovations, by virtue of conferring their benefits on everybody, has actually had the perverse effect of making everyone feel worse off. When I was a child growing up in small town Wisconsin, I had some sort of vague sense that there were rich people in the world, but from my perspective taking my first airplane flight around the age of ten was a source of great wonder, and even provided a sense of status; after all, many of my friends had never flown at all. That was the comparison set that mattered to me. Social media — or, more accurately, user-generated content feeds, which are increasingly not social at all — has completely changed this dynamic. All I or anyone else needs to do is open Instagram to see beautiful people on private jets or on beaches or at fancy restaurants, living a life that seems dramatically better than one’s dull existence in the suburbs or a cramped apartment; never mind that the means of achieving that insight is a level of technological wealth that would have been incomprehensible to the richest person in the world fifty years ago. To put it another way, what Louis C.K. identified in this clip was the extent to which human happiness is a relative versus absolute phenomenon: what we care about is not how much we have, but how we compare. That, by extension, is what drives the technological paradox I noted above: more capabilities, more broadly distributed, has tremendously enriched the world on an absolute basis; the end result, however, has been the dramatic expansion of our comparison set, making us feel more immiserated than ever. 3 This, writ large, is what Patel and Trammell seem to be worried about: sure, everyone may have all of their material needs met, but that won’t be good enough if the price of that abundance is the knowledge that someone else has more. This might not be rational, but it certainly is human! If you assume that the negative parts of humanity will persist in this world of abundance, however, then you must leave room for the positive parts as well, the ones that I wrote about above. Even if AI does all of the jobs, humans will still want humans, creating an economy for labor precisely because it is labor. You can’t make the case that the potential for jealousy ought to drive authoritarian capital controls while completely dismissing the possibility that the prospect of desirability gives everyone jobs to do, even if we can’t possibly imagine what those jobs might be — beyond podcasting, of course. If you want a specific example, consider the rapturous response to Bill Simmons’ 50 Most Rewatchable Movies of the 21st Century episode, which was delightful precisely because it was pure Bill ↩ This isn’t idle talk: I’m encouraging my daughter to pursue art; granted, I’m also working quite hard to build up a store of capital for her as well! ↩ As an aside, this is why galaxy exploration would be a positive, not a negative: out of sight out of mind, just like it used to be. ↩ If you want a specific example, consider the rapturous response to Bill Simmons’ 50 Most Rewatchable Movies of the 21st Century episode, which was delightful precisely because it was pure Bill ↩ This isn’t idle talk: I’m encouraging my daughter to pursue art; granted, I’m also working quite hard to build up a store of capital for her as well! ↩ As an aside, this is why galaxy exploration would be a positive, not a negative: out of sight out of mind, just like it used to be. ↩
The Dictator’s Handbook is an ambitious book. In the introduction, its authors Bruce Bueno de Mesquita and Alastair Smith cast themselves as the successors to Sun Tzu and Niccolo Machiavelli: offering unsentimental advice to would-be successful leaders. Given that, I expected this book to be similar to The 48 Laws of Power , which did not impress me. Like many self-help books, The 48 Laws of Power is “empty calories”: a lot of fun to read, but not really useful or edifying 1 . However, The Dictator’s Handbook is a legitimate work of political science, serving as a popular introduction to an actual academic theory of government . Political science is very much not my field, so I’m reluctant to be convinced by (or comment on) the various concrete arguments in the book. I’m mainly interested in whether the book has anything to say about something I do know a little bit about: operating as an engineer inside a large tech company. Let’s first cover the key idea of The Dictator’s Handbook , which can be expressed in three points. Almost every feature of organizations can be explained by the ratio between the size of three groups: For instance, take an autocratic dictator. That dictator depends on a tiny group of people to maintain power: military generals, some powerful administrators, and so on. There is a larger group of people who could be in the inner circle but aren’t: for instance, other generals or administrators who are involved in government but aren’t fully trusted. Then there is the much, much larger group of all residents of the country, who are affected by the leader’s policies but have no ability to control them. This is an example of small-coalition government. Alternatively, take a democratic president. To maintain power, the president depends on every citizen who is willing to vote for them. There’s a larger group of people outside that core coalition: voters who aren’t supporters of the president, but could conceivably be persuaded. Finally, there’s the inhabitants of the country who do not vote: non-citizens, the very young, potentially felons, and so on. This is an example of large-coalition government. Mesquita and Smith argue that the structure of the government is downstream from the coalition sizes. If the coalition is small, it doesn’t matter whether the country is nominally a democracy, it will function like an autocratic dictatorship. Likewise, if the coalition is large, even a dictatorship will act in the best interests of its citizens (and will necessarily democratize). According to them, the structure of government does not change the size of the coalition. Rather, changes in the size of the coalition force changes in the structure of government. For instance, a democratic leader may want to shrink the size of their coalition to make it easier to hold onto power (e.g. by empowering state governors to unilaterally decide the outcome of their state’s elections). If successful, the government will thus become a small-coalition government, and will function more like a dictatorship (even if it’s still nominally democratic). Why are small-coalition governments more prone to autocracy or corruption? Because leaders stay in power by rewarding their coalitions, and if your coalition is a few tens or hundreds of people, you can best reward them by directly handing out cash or treasure, at the expense of everyone else. If your coalition is hundreds of thousands or millions of people (e.g. all the voters in a democracy), you can no longer directly assign rewards to individual people. Instead, it’s more efficient to fund public goods that benefit everybody. That’s why democracies tend to fund many more public goods than dictatorships. Leaders prefer small coalitions, because small coalitions are cheaper to keep happy. This is why dictators rule longer than democratically-elected leaders. Incidentally, it’s also why hegemonic countries like the USA have a practical interest in keeping uneasy allies ruled by dictators: because small-coalition dictatorships are easier to pay off. Leaders also want the set of “interchangeables” - remember, this is the set of people who could be part of the coalition but currently aren’t - to be as large as possible. That way they can easily replace unreliable coalition members. Of course, coalition members want the set of interchangeables to be as small as possible, to maximise their own leverage. What does any of this have to do with tech companies? The Dictator’s Handbook does reference a few tech companies specifically, but always in the context of boardroom disputes. In this framing, the CEO is the leader, and their coalition is the board who can either support them or fire them. I’m sure this is interesting - I’d love to read an account of the 2023 OpenAI boardroom wars from this perspective - but I don’t really know anything first-hand about how boards work, so I don’t want to speculate. It’s unclear how we might apply this theory so that it’s relevant to individual software engineers and the levels of management they might encounter in a large tech company. Directors and VPs are definitely leaders, but they’re not “leaders” in the sense meant in The Dictator’s Handbook . They don’t govern from the strength of their coalitions. Instead, they depend on the formal power they derive from the roles above them: you do what your boss says because they can fire you (or if they can’t, their boss certainly can). However, directors and VPs rarely make genuinely unilateral decisions. Typically they’ll consult with a small group of trusted subordinates, who they depend on for accurate information 3 and to actually execute projects. This sounds a lot like a coalition to me! Could we apply some of the lessons above to this kind of coalition? Let’s consider Mesquita and Smith’s point about the “interchangeables”. According to their theory, if you’re a member of the inner circle, it’s in your interest to be as irreplaceable as possible. You thus want to avoid bringing in other engineers or managers who could potentially fill your role. Meanwhile, your director or VP wants to have as many potential replacements available as possible, so each member of the inner circle’s bargaining power is lower - but they don’t want to bring them into the inner circle, since each extra person they need to rely on drains their political resources. This does not match my experience at all. Every time I’ve been part of a trusted group like this, I’ve been desperate to have a deeper bench. I have never once been in a position where I felt it was to my advantage to be the only person who could fill a particular role, for a few reasons: In other words, The Dictator’s Handbook style of backstabbing and political maneuvering is just not something I’ve observed at the level of software teams or products. Maybe it happens like this at the C-suite/VP or at the boardroom level - I wouldn’t know. But at the level I’m at, the success of individual projects determines your career success , so self-interested people tend to try and surround themselves with competent professionals who can make projects succeed, even if those people pose more of a political threat. I think the main difference here is that technical competence matters a lot in engineering organizations . I want a deep bench because it really matters to me whether projects succeed or fail, and having more technically competent people in the loop drastically increases the chances of success. Mesquita and Smith barely write about competence at all. From what I can tell, they assume that leaders don’t care about it, and assume that their administration will be competent enough (a very low bar) to stay in power, no matter what they do. For tech companies, technical competence is a critical currency for leaders . Leaders who can attract and retain technical competence to their organizations are able to complete projects and notch up easy political wins. Leaders who fail to do this must rely on “pure politics”: claiming credit, making glorious future promises, and so on. Of course, every leader has to do some amount of this. But it’s just easier to also have concrete accomplishments to point to as well. If I were tempted to criticize the political science here, this is probably where I’d start. I find it hard to believe that governments are that different from tech companies in this sense: surely competence makes a big difference to outcomes, and leaders are thus incentivized to keep competent people in their circle, even if that disrupts their coalition or incurs additional political costs 4 . Still, it’s possible to explain the desire for competence in a way that’s consistent with The Dictator’s Handbook . Suppose that competence isn’t more important in tech companies , but is more important for senior management . According to this view, the leader right at the top (the dictator, president, or CEO) doesn’t have the luxury to care about competence, and must focus entirely on solidifying their power base. But the leaders in the middle (the generals, VPs and directors) are obliged to actually get things done, and so need to worry a lot about keeping competent subordinates. Why would VPs be more obliged to get things done than CEOs? One reason might be that CEOs depend on a coalition of all board members (or even all company shareholders). This is a small coalition by The Dictator’s Handbook standards, but it’s still much larger than the VP’s coalition, which is a coalition of one: just their boss. CEOs have tangible ways to reward their coalition. But VPs can only really reward their coalition via accomplishing their boss’s goals, which necessarily requires competence. Mesquita and Smith aren’t particularly interested in mid-level politics. Their focus is on leaders and their direct coalitions. But for most of us who operate in the middle level, maybe the lesson is that coalition politics dominates at the top, but competence politics dominates in the middle. I enjoyed The Dictator’s Handbook , but most of what I took from it was speculation. There weren’t a lot of direct lessons I could draw from my own work politics 5 , and I don’t feel competent to judge the direct political science arguments. For instance, the book devotes a chapter to arguing against foreign aid, claiming roughly (a) that it props up unstable dictatorships by allowing them to reward their small-group coalitions, and (b) that it allows powerful countries to pressure small dictatorships into adopting foreign policies that are not in their citizens’ interest. Sure, that seems plausible! But I’m suspicious of plausible-sounding arguments in areas where I don’t have actual expertise. I could imagine a similarly-plausible argument in favor of foreign aid 6 . The book doesn’t talk about competence at all, but in my experience of navigating work politics, competence is the primary currency - it’s both the instrument and the object of many political battles. I can reconcile this by guessing that competence might matter more at the senior-management level than the very top level of politics , but I’m really just guessing. I don’t have the research background or the C-level experience to be confident about any of this. Still, I did like the core idea. No leader can lead alone, and that therefore the relationship between the ruler and their coalition dictates much of the structure of the organization. I think that’s broadly true about many different kinds of organization, including software companies. Maybe there are people out there who are applying Greene’s Machiavellian power tactics to their daily lives. If so, I hope I don’t meet them. “Organizations” here is understood very broadly: companies, nations, families, book clubs, and so on all fit the definition. I write about this a lot more in How I provide technical clarity to non-technical leaders In an email exchange, a reader suggested that companies face more competition than governments, because the cost of moving countries is much higher than the cost of switching products, which might make competence more important for companies. I think this is also pretty plausible. This is not a criticism of the book. After five years of studying philosophy, I’m convinced you can muster a plausible argument in favor of literally any position, with enough work. When explaining how organizations 2 behave, it is more useful to consider the motivations of individual people (say, the leader) than “the organization” as a whole Every leader must depend upon a coalition of insiders who help them maintain their position Almost every feature of organizations can be explained by the ratio between the size of three groups: The members of the coalition of insiders (i.e. the “inner circle”) The group who could conceivably become members of the coalition (the “outer circle”, or what the book calls the “interchangeables”) The entire population who is subject to the leader Management are suspicious of “irreplaceable” engineers and will actively work to undermine them, for a whole variety of reasons (the most palatable one is to reduce bus factor ) It’s just lonely to be in this position: you don’t really have peers to talk to, it’s hard to take leave, and so on. It feels much nicer to have potential backup Software teams succeed or fail together. Being the strongest engineer in a weak group means your projects will be rocky and you’ll have less successes to point to. But if you’re in a strong team, you’ll often acquire a good reputation just by association (so long as you’re not obviously dragging the side down) Maybe there are people out there who are applying Greene’s Machiavellian power tactics to their daily lives. If so, I hope I don’t meet them. ↩ “Organizations” here is understood very broadly: companies, nations, families, book clubs, and so on all fit the definition. ↩ I write about this a lot more in How I provide technical clarity to non-technical leaders ↩ In an email exchange, a reader suggested that companies face more competition than governments, because the cost of moving countries is much higher than the cost of switching products, which might make competence more important for companies. I think this is also pretty plausible. ↩ This is not a criticism of the book. ↩ After five years of studying philosophy, I’m convinced you can muster a plausible argument in favor of literally any position, with enough work. ↩
This one hits a little too close to home. Like Tim Cain, I am done with making other people rich off my IP. I enjoying making small games like Death By Scrolling and I’m going to keep making games and having fun. You may like the games, you may not, but I’m making what I want. I’m not rich but I can pay for food and rent and make what I want. People often ask about a TWP2. TWP cost around $1.1M to make. $600k came from Kickstarter backers and $500k came from private investors I found later. Kickstarter for digital games is all but dead. I could not raise the money to make TWP2 on Kickstarter today. I couldn’t even raise the full amount back then. I have spoken to publishers and they have been willing to fund TWP2, but they get rich and I get very little and have to do most of the work. I am done with that. From now on I’m going to make the small games I want and have fun doing it. I’ve been in the games industry for 40+ years. I think I’ve earned that. MicroProse is publishing DBS, but unlike other publishers they offered an very fair deal. Also, unlike a game like TWP2, there wasn’t a lot of upfront money and that probably helps.
I love the energy of New Years. Resolutions and intentions thread through my day-to-day thoughts in the days and sometimes even weeks leading up to a new year. That has toned down quite a bit since having a kid, but I’m still holding on to this meaningless date that we attribute meaning to and I am looking forward to some fresh-starts. More than anything, this coming year, I’d like to steal back my attention, even more. Focus might be an apt theme for the year. As I work to continue to eradicate unwelcome, manipulative distractions from my life and redirect that stolen time and attention back to more important matters, I find myself thinking about what it means to be on the other side of distraction; that is, imagining what kind of person I am after "arriving" (he said, trying not sound extremely obnoxious). Arriving back at having my attention belong to me takes me to thinking about my post on priorities, and spending time—namely, what do you do with that new time that you have? One thing I’ve realized, is that what is on the other side of a doomscroll, is another kind of doom: facing up to what you actually want to do, and how uncomfortable that is. For me, it’s an abundant backlog of several creative art projects, idly standing by for me to just start (and keep going) . It is altogether scary to have even gotten to a point where I’ve realized I could actually just do these things once I’ve taken command of my attention. I don’t know what to call this petrifying position beyond analysis paralysis . But what I find interesting about it is that it exists at all. I would have thought that just having gotten at least some of my attention back would mean I’d be free and clear and the actual directing of this newfound time/energy/focus would be immediately put to use. Not so, of course. I expect that as this year progresses, I’ll be whittling down more and more things. I look forward to this, as hard as it can be to let go. But the things you love, and you want to love, are out there waiting for you—you have to drop a few of the heavier things to make it out there, though. I don’t know what shape writing on this site will have for me this year. I’ve enjoyed writing here in this all-too-non-fiction-way, and I hope to continue doing it. But really, it is a stand in for more important writing I want to do. I expect my writing here to be whittled down or change shape in some way or other, too. Everything is an experiment, everything is exploration. May you have peace, happiness, love, joy, and continued experiments in your own way this year.
Hardware
AI
Business
Mobile
Sports
Career
Go
UI
Agile
Gaming
Writing
Culture