Writing
Working notes on AI coding tools, agents, evaluation loops, and the craft of shipping software that holds up.
Is Vibe Coding Real Programming?
Why the debate keeps repeating, and why prototype vs production is the line that matters.
The agent orchestration trap: what have you shipped this week?
Agent swarms that look impressive but ship nothing. Productivity theater versus actually building things.
You probably don't need an eval platform
What actually caught regressions while building an AI detection product: traces you can diff, plus a small golden set in CI.
I spent a year building AI features and shipped nothing
What over-engineering burnout actually looks like from the inside, and the ordering mistake that caused it.