Sunny

← All weeks

What I’ve been reading (W39, 2026)

Where agents shine, the wisdom AI can't learn, why AI keeps not taking jobs, what replaces work as a source of respect, and just showing up

  1. Agents are bad at things built for humans and great at things nobody has built yet. Ask one to book an appointment on an existing website and it stops at the login page or gets blocked by the shop. Ask it to build you a small custom tool from scratch, say a web page that runs a battery test and logs the results, or to turn a school PDF into a shared family calendar, and it’s live in minutes. New things are easy because there’s nothing in the way, and that is where agents shine. The problem is nobody can imagine what to ask for. Muse is far more capable than people realise, with a real 8GB computer per user, yet it launches with examples like “go buy me some stuff”, exactly where it will fail, so people will hit the wall and leave. If it takes off it will be as a social media trend, people showing each other what’s possible. The bigger miss is that Silicon Valley doesn’t understand normal people: the demo is always a bot booking your flights, and most people don’t fly. Software engineering fell first not because it was the hardest job but because it meant thinking like a computer, and the rest of the world is much harder than that. stratechery.com/2026/an-interview-with-joanna-stern-about-the-iphone-duo-and-ai-for-normal-people
  2. AI has no wisdom, and neither will you. Maintainability and good architecture have no good measurements, because it takes months or years to notice the effects of bad code. Most engineers rely on context-dependent intuition built from past mistakes, which can’t be turned into rigid rules i.e, they don’t follow the rules, they make them. AI can’t learn this: reinforcement learning needs a reward signal that can be measured immediately, there’s no fitness function for maintainable code, and it learns from beginner rulebooks and mostly bad code in the wild. The risk is that if we lean on AI to write and read code without making the choices ourselves, we miss the mistakes that build that intuition, so it’s worth staying in the loop alexn.org/blog/…/ai-has-no-wisdom-and-neither-will-you news.ycombinator.com/item?…
  3. AI keeps refusing to take our jobs. Prime-age employment is near all-time highs, software developers are a near-record share of the US workforce despite Codex and Claude Code, translator employment is flat, and radiologists and truckers are fine. Of companies telling the Census Bureau that AI changed their headcount, more reported an increase than a decrease, and firms that adopt more AI tend to hire more humans, including at entry level. The Economist puts AI’s job creation at about 1 million in America against roughly 200,000 AI-attributed layoffs, much of it the data-centre build-out, where installation and maintenance work pays about 40% more than comparable jobs elsewhere. The mechanism is that automation does three things, not one: it displaces tasks, it raises productivity, and it creates new tasks, and right now the new tasks are matching or beating the replaced ones. Engineers write less code but spend more time deciding what to build, checking what the model wrote, and integrating it. The sharpest finding is about fear: the more of their day people spend using AI at work, the more they expect to be replaced, yet AI exposure has no correlation with actual job loss. People don’t actually know how they produce value at their jobs. Replacing tasks isn’t the same as replacing jobs, because modern jobs are messy bundles of tasks, which is why travel agents survived the internet and elevator operators didn’t. noahpinion.blog/p/ai-keeps-stubbornly-refusing-to-take
  4. People need respect and a place in society. Work gives people a way to earn that respect through competence and contribution. If that disappears, belonging to the right social group might become much more important. Groups can demand conformity in exchange for belonging. If paid work becomes less central, what would give people independence, recognition, and the freedom to leave a group? asteriskmag.com/issues/15/after-work-we-ll-have-each-other
  5. One of the most important skills for excellent performance (and honestly, for an excellent life) is showing up. Relationships. Sports. Art. Music. Business. Creativity. Craft. Parenting. Coaching. It doesn’t matter the pursuit; you’ve got to show up. People think intensity drives progress, but it’s much more about consistency. It’s not about hitting home runs, but stepping into the arena and putting the ball in play. It’s about being there over and over again. There’s this myth that you need to feel good to get going, that you need to be super inspired and super motivated all the time. But more often than not, in reality, what you need is to get going to give yourself a chance at feeling good. bradstulberg.org/p/an-ode-to-showing-up
  6. It takes 5 cloud services to hear the doorbell sometimes. blog.vghaisas.com/rube-goldberg-doorbell