Writing

Notes on building AI-native engineering teams.

Essays on the operating model, eval discipline, and the systems that compound. Written for the people doing the work.

2
posts
8.2k
total words
3.1k
newsletter readers
1yr
writing on AI
Canonical · read this first

Why most engineering AI adoption stalls

AI adoption inside engineering is near-universal — 84% of developers, 90% of teams. The productivity story is collapsing under it. Throughput climbs, quality degrades, only 3.1% of developers highly trust the output. The fix isn't more tools. It's a different operating model. Here's what AI-native engineering actually looks like, with a seven-question checklist your team can grade itself against.

~21 min readJun 2026// operating-model
Jun 2026

Pipelines, agents, and the maturity curve

Anthropic itself recommends pipeline-first. The dominant frameworks default to agents anyway. Most production work is pipeline-shaped — and 'agentic' is the most expensive word in AI engineering in 2026. A four-question decision tree, with a maturity curve underneath, that picks the right architecture for the actual task.

// architecture