Field notes from the studio — how we build, what works, what doesn't, and where AI is actually changing software engineering. Written by the engineers doing the work.
Nine in ten Retrieval-Augmented Generation projects we're asked to review have the same problem — nobody set up an evaluation pipeline before writing a line of code. A practical framework for measuring RAG quality on your own data.
· 10 min read
Read the article →Sprint by sprint, hour by hour: how RedFin AI compresses three months of typical agency work into 28 calendar days. What we automate, what we don't, and where the 80% speed-up actually comes from.
· 9 min read
Read the article →Almost every AI system we're asked to review is spending 3–5× more on LLM inference than it needs to. Here are the five techniques we apply — prompt caching, small-model routing, semantic caching, prompt compression and batching.
· 11 min read
Read the article →The 2018 answer doesn't apply anymore. React Native has matured, native tooling has changed, and the honest trade-offs now live in different places than most articles describe. Here's where they actually are.
· 8 min read
Read the article →Most UK teams think the EU AI Act is either a problem for someone else, or such a big problem they can't start yet. Both are wrong. What actually applies to you and how to ship this quarter.
· 10 min read
Read the article →Engineering implies measurement, repeatability, handover. Most "prompt engineering" today has none of them. Here's what the upgrade to real production LLM engineering actually looks like.
· 7 min read
Read the article →One email per post. No noise, no drip campaigns. Or subscribe by RSS to fold posts into your existing reader.