What Broke When Your Organization Gave Developers AI Coding Agents?
Unexpected issues arising from deploying AI coding tools for developers.
Supporting over 200 developers, we rolled out agentic coding tools this year. The failure modes were unexpected, not limited to AI writing insecure code, but included agents committing credentials from local environment files, installing unvetted dependencies, and handing overly broad pipeline tokens to autonomous processes. We’re particularly interested in surprises faced by teams in regulated industries.