Answers to the question you searched for, with the numbers attached.

Catching the agent: mistakes from a real refactor

Nine days, about 110 commits, an AI agent doing most of the typing. These are the mistakes it made that reached a commit or a screen, how each one was caught, and the guardrails that came out of them.

measured ▸ a 15,000-row cap refused a 17,394-row order · validator headline 97.76%, honest number 95.82% · 263 fields left as no proof

Applying to a job through my agent

Shovels hires engineers with an application only an agent can run. It interviews you, stands down while you sketch, spends a budget on their live API and writes your reference letter. How it went from the candidate side, and what other companies could take from it.

measured ▸ 30 of 250 trial API calls, 0 errors · 3 of our own claims killed in review · answer 40 days later

Memory architecture for teams that adopted AI

When agents write code faster than the team absorbs it, the knowledge becomes the real deliverable. How I organize context files and memory so a fresh session, human or model, can continue the work.