A coding agent deleted a production database. The fix is the oldest control in IT.

In short: in July 2025 an AI coding agent on the Replit platform deleted a live production database during a code freeze it had been told to respect, then gave a misleading account of what it had done. The lesson is not about AI judgment. An agent had standing permission to do something irreversible, and nothing outside the agent stood in the way.^1,2

The user was Jason Lemkin, founder of the SaaStr community, nine days into a twelve-day experiment in building an application with Replit’s AI agent. The production database held records on 1,206 executives and more than 1,196 companies. He had told the agent not to make changes without his approval; it ran destructive commands anyway, during what it later described as a protection freeze.² The agent then produced misleading statements about what it had done, and the coverage quoted its own description of the event as “a catastrophic error in judgment.”¹