Docker, alongside Keycard, Socket, Tessl, Tracebit, Browserbase and Tailscale, has published the Secure Software Factory reference architecture. The document extends the work started in July with Agent Baseline, an open, vendor-neutral framework that defined security outcomes for running AI agents in the enterprise. Where Agent Baseline answered three core questions for adopters — what is operating and with what capabilities, whether it stays inside approved boundaries, and whether you can prove what happened and stop it — the new architecture applies those same outcomes to one of the most common agentic workflows: the AI-powered software factory.
The Secure Software Factory was written by practitioners and shaped by conversations with security leaders at large enterprises in regulated industries such as finance and insurance. The changes it introduces include:
- A vendor-neutral reference architecture that maps Agent Baseline security outcomes onto the systems that build software, rather than onto individual agents alone.
- Focus on the AI-powered software factory pattern, where agents take a request and carry it through the engineering workflow.
- Input from Keycard, Socket, Docker, Tessl, Tracebit, Browserbase and Tailscale, combining perspectives across supply chain, sandboxing and network access.
- Guidance aimed at regulated environments, where proof of what happened and the ability to stop activity are operational requirements, not optional extras.
- A shared frame of reference so teams can evaluate tooling against defined security outcomes instead of ad hoc checklists.
For developers, the significance is that the systems building your software are now treated as a first-class security surface. As agents take on more of the request-to-code path, the boundaries around their capabilities, the evidence trail they leave, and the controls that can halt them become part of your architecture. The reference architecture gives teams a common vocabulary for those controls and a way to compare implementations, which should make it easier to adopt agentic workflows in environments where auditability matters.