A site about product and software engineering. Read & Discover about building teams, org and impactful software products.
Most companies are pricing AI adoption like a platform migration — a multi-year programme before any value shows up. It isn't one. Every enterprise has already made this journey twice: information became findable, then knowledge. This is the third turn, where knowledge becomes insight and insight acts — a paradigm shift in how work gets done, not an upgrade to the stack you already run. Why the…
Vibe-coded software ships, demos, and even works. It still rots, because the engineering fundamentals underneath were never there. On cargo-cult engineering, the rent you pay for a facade, and why the model was never the moat.
Alex Honnold didn't remove the harness — he internalized it. For 100x engineering teams using AI agents, a load-bearing harness is the only way to climb safely.
The Anatomy of an AI-Native Org argued that the middle of the org chart collapses because translation work is the work that gets eaten. This is the role-by-role walkthrough for anyone who asked the obvious next question — what happens to my job specifically?
For thirty years we were glorified translators — business asked why, product defined what, engineering translated to how. AI just ate the translation step. The anatomy of the team that's left looks nothing like the one you have today.
I argued for tests, trunk-based development, and against the PR-rubber-stamp ritual for twenty years. Most teams didn't listen. Now AI is shipping in hours, the bugs are shipping in hours, and the industry is rediscovering — under new names — every practice we used to skip.
Three months into a sabbatical, six things became obvious. Most of them are about the damage we do to ourselves while calling it ambition.
In my last post I wrote about the small island of delight — the agent — surrounded by a sea of ops work. This is the map of that sea. Ten walls people keep hitting when they try to run their own AI agent, and what each one actually costs.
It started with a friend weighing a dedicated Mac against a VM full of config. It turned into a question I can't stop thinking about — what blockers are keeping people from actually using AI? ClawStation is my attempt at an answer.
Every conversation about AI agents celebrates speed. But the hardest problem isn't building agents — it's deciding what 'correct' means. When execution is nearly free, judgment becomes the expensive thing.