Hey HN — I'm Torben, I've been building software for 40 years. For the past 6 months I've been running AI agents in real delivery pipelines and kept hitting the same problem: there's no governance model for agent-driven delivery. CI/CD assumes humans write code. That assumption is broken now. GitHub calls their thing "Continuous AI," PwC published on the "Agentic SDLC," but neither gives you an actual framework you can implement. So I built one. This is a gate-driven process framework — 6 steps, 5 gates, production feedback loop — designed for the transition from CI/CD to what I'm calling AI/CD (Agentic Integration / Continuous Delivery) and eventually AI/AD (Autonomous Delivery). What makes this different from a whitepaper: every gate has real evidence. You can see the actual PRs, issue comments, and gate sign-offs in the repo. Example: https://github.com/torbenanderson/icebox-cli/issues/23 It's platform-agnostic (GitHub, GitLab, Azure DevOps, Jira + any SCM combo) and open source. Fork it, adapt it. Keen to hear from anyone running agents in their delivery pipelines — what's working, what's breaking, and whether this kind of governance makes sense for your team.