AI coding agents are becoming increasingly powerful, but writing better prompts is only one part of working effectively with them. As projects grow, the bigger challenge becomes helping agents understand context: past decisions, debugging history, architecture choices, and lessons learned over time.

Inspired by Andrej Karpathy’s llm-wiki concept, I experimented with building my own AI wiki for coding agents using automation, hooks, and documentation workflows. I wanted to explore a question many developers are starting to face: how can we help AI assistants build on accumulated knowledge instead of starting from scratch?

In this talk, I will share what worked, what broke, and what I learned from building this workflow — from capturing useful context to understanding the limitations of relying on AI-generated knowledge.

As AI changes how we build software, managing context might become one of the most important skills in collaborating effectively with coding agents.