LangChain wants your repo to write the docs for the agent, not you

LangChain shipped OpenWiki, and the pitch is blunt: point it at a codebase, it reads the whole thing and generates documentation built specifically for coding agents, then drops a line into your AGENTS.md and CLAUDE.md telling the agent to consult those docs before it goes spelunking through files. A GitHub Action keeps it fresh so the docs don't rot two commits later. Five thousand stars in days, and it hit the HN front page today.

Here's why it matters. Every time a coding agent opens a fresh session, a big slice of its context and token budget gets burned just re-discovering how your repo is laid out. grep, read, grep, read, over and over. OpenWiki front-loads that work once and sinks it into a durable knowledge layer the agent reads instead of rebuilding. It's the same bet as DESIGN.md and the agent-ready design systems showing up lately: the repo should carry a machine-readable map of itself, not force every agent to redraw one from scratch.

The tell is who shipped it. LangChain built its name on orchestration frameworks, the wire-the-models-together layer. Now it's shipping a documentation tool. That's a quiet signal about where the value moved. Stringing models together got commoditized; feeding them the right context cheaply is the part people still pay attention to. Link: github.com/langchain-ai/openwiki