Blog
How we build structural evidence for AI coding agents — what the graph can resolve, what it cannot, and how we measure the difference instead of asserting it.
RSS feed →Why an AI agent can't tell you what calls a function
A source file records everything it depends on and nothing that depends on it, which is why "find all the callers" is a search problem, not a read problem.
How to measure whether an MCP server actually saves tokens
Agent benchmarks fail silently: the environment mutates under you and the numbers still look fine. Here is a harness built to fail loudly instead, and the result that survived it.
Blast radius: what to check before and after an agent edits
Two moments in a review loop where structural evidence changes a decision, and the architectural choice — the working tree as the primary universe — that makes both possible.
Running Euthynos with Claude Code: setup and what it writes
Registration is one line; the three audits that should precede it — disk, network, trust — are the rest of this post, including a wrong command in our own README.