// writing

Production AI, real numbers, no hype.

Long-form essays on the production engineering of AI. Each post leads with the number that mattered, then walks the changes that produced it.

• 2026-04-12 // legacy · port pending

  How I cut our LLM bill 28% without changing models

  Six specific production optimisations — task routing, semantic caching, prompt compression, structured output, batching, gatekeeping — with per-change contribution numbers.

• 2026-04-12 // legacy · port pending

  Bridge Sourcing: how I moved scrape accuracy from 82% to 96%

  The 8 specific engineering changes that moved a production LLM-extraction pipeline from 82% to 96% field-level accuracy over 3 months, with per-change accuracy impact.

Newsletter ships weekly starting May 9. Email me if you want the early issues.