// 2 ESSAYS · MORE COMING

Writing/about/shipping/AI.

Methodology, technical deep-dives, and the occasional contrarian take. No newsletter · RSS only, on purpose.

// FEATURED
2026-05-09methodology
The compound failure problem: why your agent demo won't scale
Per-step accuracy is a misleading metric. A 95% step accuracy across a 10-step agent yields ~60% end-to-end. The fix isn't a better model. It's the boring infrastructure most teams skip.
10 MIN READread →
2026-05-09technical
Long-context vs RAG: a 2026 decision tree
With 1M+ token windows now standard, the old advice is wrong but the reactions are wrong too. A framework keyed on corpus size, latency budget, and cost ceiling, and why most systems converge on a hybrid.
11 MIN READread →
RSS
No newsletter. Subscribe via RSS to get every new essay the moment it drops, no inbox clutter, no unsubscribe friction.
subscribe → RSS