it depends on the codebase. if there's adequate test coverage, and the tests emit informative failures, coding agents can be used as constraint-solvers to iter

it depends on the codebase. if there's adequate test coverage, and the tests emit informative failures, coding agents can be used as constraint-solvers to iterate and make changes, provided you stage your prompts properly, much like staging PRs. claude code is really good at this.