autoresearch

Automated codebase optimization runs iterative experiments, applies code changes autonomously, measures numeric outcomes, and keeps or reverts changes based on budget-guided exploration until a goal is reached.

Autonomous experimentation requires a clearly defined Goal, a Metric command producing a numeric value, a Metric extraction method, a Direction setting like lower_is_better, and in-scope or out-of-scope files to confine edits.

Metric-driven exploration can optimize CI build times by running automated experiments that apply code changes, measure the resulting build duration, and retain improvements that lower the measured metric.

You restrict automated code changes by defining in-scope and out-of-scope files before starting the experiment loop, confining edits to targeted areas and preventing modifications to the rest of the codebase.

When a code change does not improve the target metric, the autonomous loop reverts the changes and continues hill-climbing improvements through further iterative experiments within the defined budget.

Autonomous codebase optimization removes manual trial-and-error by automatically running code changes, extracting metric values, and deciding to keep or revert modifications based on the specified optimization direction.