Writing

What we are learning, written down.

Notes on building with AI — what works, what quietly does not, and the mistakes worth skipping. Practical rather than promotional.

What does the OneShopAI blog cover?

Posts here are about the part of building with AI that does not fit in a demo — why a retrieval system returns the wrong document, what changes when a prompt becomes an agent loop, how to tell whether a change actually improved anything. The bias is towards things we had to learn the expensive way.

What you will not find is coverage of every model release. That is well served elsewhere and it dates within a fortnight. The aim is writing that is still useful in a year, which mostly means writing about the problems rather than the products.

If you are earlier on and want structure rather than essays, the learning paths give an ordered route through the same material, and the AI glossary explains any term these posts assume you already know.