• Knowledge Priming AI coding assistants default to generic patterns from their training data. • I propose treating project context as infrastructure-versioned files that prime the model before each session-rather than relying on ad-hoc copy-pasting. • This is essentially manual RAG (Retrieval-Augmented Generation), and I believe it fundamentally changes the quality of AI-generated code. • 24 February 2026 This article is part of a series: When I onboard a new developer, I don’t just point them at the codebase and say âgo.â I walk them through our conventions. • I show them examples of code we consider good. • I explain why we made certain architectural choices-why we use Fastify instead of Express, why services are functional instead of class-based, why validation happens at the route level.

Article Summaries:

  • The article proposes “Knowledge Priming” as a way to improve AI coding assistants. Instead of relying on generic patterns from training data, the author suggests treating a project’s versioned files-conventions, architecture notes, naming rules-as priming documents that are loaded before each AI session. This manual Retrieval‑Augmented Generation (RAG) step would override the model’s default internet knowledge, reducing the “frustration loop” where developers repeatedly regenerate code that doesn’t fit the codebase. The author outlines a three‑tier knowledge hierarchy-training data, conversation context, and priming documents-to explain how context should dominate AI output. The approach aims to make AI assistants behave like well‑onboarded human developers.

Sources: