Machinist measures coding-agent runs

Machinist is a software-factory control plane that queues coding tasks, dispatches workers to agent harnesses, and records run-level metrics such as time, tokens, commands, and outcomes. Why it matters: A shared agent workflow makes prompt, model, and harness changes comparable over time instead of relying on anecdotes. The practical gain is a repeatable path from run data to a specific workflow improvement, such as replacing a routine agent step with a deterministic script.

Try this: Run one recurring coding task through a fixed prompt and capture a trace with task time, token use, commands, and test outcome; compare that baseline with one changed model or harness.