How we think about running AI coding agents in production — the problems, the architecture, and the hard-won lessons.
AI agents don't exfiltrate data the way attackers do. They do it by doing their jobs — fetching packages, calling tools, processing context. That's what makes it so hard to catch.
Read more →Engineering teams are afraid to run AI agents in CI/CD pipelines. The fear is rational — because the current security tooling doesn't provide the right guarantees.
Read more →Everyone says 'don't put secrets in the agent's environment.' Almost nobody talks about what safe injection actually requires — especially when your stack goes beyond HTTP.
Read more →