Agents generate PRs faster than teams can review them. Bots catch a lot of the issues - linters, CI, security scanners, review tools. A human then has to go fix it. Or copy the feedback into a coding