Engineering notes on code graphs and AI agents

This is where the Symvanta team writes about the hard part of AI coding: giving an agent an accurate model of a codebase it has never read. Most tools hand the model a pile of embeddings and hope the right file surfaces. We build a code graph instead: every symbol, every call edge, every import, resolved by a compiler-grade parser and served over MCP. The posts here get specific about what that changes.

Expect engineering notes written for other engineers: why agents fail on large repositories, how blast radius catches a breaking change before you merge, where embeddings help and where they mislead, and how the graph spans every repository in a project instead of stopping at one. Cross-repo context is the difference between an agent that guesses and one that knows, and it is what we care about most. New pieces land as we ship, and you can see the same analysis run on real open source projects in the architecture library.