AI answers the average problem.
The work now is knowing how yours differs, and checking the answer.
Latest writing
A flaky test is not a broken test. It is a broken gate.
Flakiness does not cost you the time the reruns take. It costs you the meaning of a red build, and everything downstream of that assumes red means something.
8 min readAn average expert answer is now cheap. Knowing when it's wrong isn't.
AI gives everyone the same confident answer to the typical problem. Two small experiments on what agents assume, what they lock in, and where judgement now sits.
15 min readWhy Your Architecture Matters More Than Your AI Budget
Enterprise AI adoption succeeds or fails based on architectural foundations, not tool spending.
6 min read
I’m a principal platform engineer working on developer platforms, CI/CD and AI enablement. I write about engineering maturity and what AI-assisted development does to it. The teams that pull ahead are the ones that know how their systems actually work, and can point AI precisely at the hard parts.
Selected work
A build signal people trust again
Integration tests on a multi-million-line open-source monorepo were failing constantly and being ignored. A merge queue, the flaky tests driven out, builds gated and green. The full CI cycle went from three hours to under one.
Agent tooling engineers actually used
Claude skills for CI failure triage and issue workflow, adopted by the dotCMS team, each one reporting where it failed and where it spent more tokens than it needed. Plus the context architecture that let agents navigate the monorepo.
AI Maturity Assessment
Eleven questions on how an engineering organisation actually works with AI, built from my maturity framework. Free, no sign-up, and the result lives in the link.