Notes from production.
What our teams learn shipping AI and product systems that have to survive contact with real users, real data and real audits. No thought leadership — field notes.
Will AI replace outsourced developers? What actually changes about team economics
AI compresses the mechanical middle of software work. It does not compress judgement — and that changes who wins in outsourcing.
Buy or build? A straight answer on internal copilots
Buy when the workflow is generic and the data is simple; build when permissions, domain data or audit dominate. And the hybrid that usually wins.
May 2026
The pilot graveyard: why enterprise AI pilots die, and the org-chart reasons nobody admits
AI pilots rarely fail on technology. They die from missing owners, missing budget lines, and permissions nobody untangles.
May 2026
Golden datasets before golden demos: building eval suites clients trust
What a golden dataset is, why domain owners must adjudicate it, and how regression evals in CI make AI systems safe to change.
May 2026
Agent autonomy is earned per workflow, not granted per project
Approval gates wherever money, data or risk moves — and an autonomy ladder each workflow climbs on eval evidence, with an audit trail per action.
May 2026
Your RAG demo works. Here’s the six months between that and production.
The gap between an impressive retrieval demo and a production system is six kinds of work — and most of them are not model work at all.
Feb 2026