Writing

Engineering lessons from production AI systems

A short list of lessons that survive contact with production. No invented metrics.

2026-07-15notes

Lessons I am willing to stand behind without dressing them up as case-study numbers:

  1. Evidence before interpretation. Store what happened. Then generate.
  2. Inconsistency is usually timing or retrieval. Debug those before you fine-tune a story.
  3. Idempotency is a product feature. Users experience duplicate actions, not your worker’s retry policy.
  4. Eval the path, not the paragraph. Tool traces and retrieved ids belong in the score.
  5. A prototype is a liability the day it meets retries.

This page will grow only with lessons that can be explained without confidential detail.