Insights on product engineering, AI transformation, and the future of software development
A data-driven look at why frontier coding benchmarks are rising by orders of magnitude while real-world productivity gains remain uneven, bottlenecked, and highly context-dependent.
Read more →For fifteen years, the SaaS business model rested on two moats — code customers couldn't build, and know-how their teams didn't have. AI dissolved both. SaaS isn't dying, but the version that sold tools has to. Here are four real cancellations, replacements, and rewraps from our own stack, and what they say about what SaaS has to become next.
Read more →The studies showing AI tools making engineering teams slower aren't wrong — they're just measuring the wrong patient. Here's what really happens when you give AI to one role and nothing to the rest of the company.
Read more →Your company bought every AI tool on the market and productivity didn't move. The problem isn't the tools. It's that nobody built the scaffolding. Meet the role that fixes it.
Read more →A small mistake in your data model used to cost you a week of refactoring. With AI, it costs you every prompt you'll ever write. Meet the token tax — and the single mental model that stops it.
Read more →Slicing an epic used to be obviously correct. With AI-assisted development, the coordination overhead of four features can exceed the time to build the whole thing. The old rules don't fit the new economics.
Read more →Handing Cursor and Claude Code to developers without the right scaffolding is how you end up explaining a production incident to your CEO. Three layers separate teams that compound with AI from teams that explode with it.
Read more →A faster car doesn't beat traffic lights, school zones, and rush hour. A faster developer doesn't beat code review queues, deployment windows, and PM sign-offs. The uncomfortable math of why AI rollouts underwhelm.
Read more →In 2000, a Spanish utility giant billed its customers through a Microsoft Access database built by a non-technical employee and her husband. A year later, entire shopping centers hadn't been charged. The pattern repeats today—faster and with worse tools.
Read more →The Endesa story was one failure mode in one year. Here's the full catalog of ways non-technical vibecoding ends in disaster—twelve specific patterns you should print out, audit against, and hang on the wall of whoever thinks 'I'll just automate this'.
Read more →AI accelerates everything—including your mistakes. A deep dive into why most teams use AI to crash faster, not to build better.
Read more →Everyone's installing Clawdbot. I reviewed the code and found 14 security issues. But context matters more than vulnerability counts.
Read more →Every PM is in a relationship with their product. Three of these patterns are destroying it. One might save it.
Read more →We gave Claude Code to a domain expert and a senior developer. Only one finished the app. The result surprised us both.
Read more →The role that was supposed to bridge business and technology is now the bottleneck. Here's what's replacing it.
Read more →I asked Claude to play my executive team. Things got uncomfortably honest.
Read more →Developers dismiss AI code quality. PMs insist they're essential. Both are accelerating exactly what they fear most.
Read more →Paul Graham's controversial essay sparked debate across Silicon Valley. But the data supports his thesis.
Read more →GitHub's research shows massive developer productivity improvements from AI. But the gains aren't evenly distributed.
Read more →