Notes on agents, codebases, and the work before the work.
Problem-first engineering writing about why agents keep starting cold, why notes are different from documentation, and why some context should be captured only after real work.
Where the tokens go in an agent session
Most of what a session bills for is not the answer it produced. It is the same context, re-sent on every turn. Decomposing that leaves exactly one lever worth pulling.
Why an index of your code cannot answer the same question twice
A graph makes each hop cheaper. It does not reduce the number of hops, and because it is regenerated from source, the second investigation costs exactly what the first one did.
The tool the agent does not call
Availability, documentation, and an explicit instruction still do not add up to adoption. Including the hook-gating mistake I shipped and then found in someone else's source.
From four tools to two
An origin story told through deletions. Two traversal tools that turned out to be missing fields, a set of labels that were guesses in the costume of facts, and a server that served nothing.
Notes about code should be written by whoever read the code
A transcript is not the same thing as context. The agent that did the work is usually the only one qualified to write down what was learned.