Agents definitely fix this. When you can run commands and edit files, the agent can test its code by itself and fix any issues.