Runtime Governance for Autonomous AI Systems
Why the question moved from "is this code safe to ship?" to "which agent is doing what, and should it be allowed?"
The question changed
For years the category was AI code review: grade what an assistant generated, then decide whether to ship it. That answers "is this code safe to ship?" It does not answer the question that matters once an autonomous agent can act on its own: which agent is doing what in your environment, and should it be allowed to do that? Runtime governance is the discipline of answering the second question. DeepSweep is built around it (ADR-RGOV-000).
The progression
Runtime governance is delivered in stages, each building on the one before it: Review (what can this agent do?), Identity (which agent is acting?), Authorization (what should it be allowed to do?), Protect (enforce the boundary), Control, and Runtime. You cannot enforce a boundary you have not authorized, and you cannot authorize an agent you have not identified — so the order is the product, not an accident.
Why runtime, not pre-ship
A pre-ship review inspects code that already stopped changing. An autonomous agent keeps acting after you ship — it pushes to a repository, runs shell commands, calls MCP tools, reads secrets, triggers deploys. Governance has to live where those actions happen: at the action boundary, before execution (ADR-RGOV-011). That is the difference between an advisory report and a control.
Where you start: the Agent Environment Review
The entry point is an Agent Environment Review — a read of one agent in one workspace that surfaces its capability profile, its identity, and the authorization gaps where its authority is ungranted. It is local, deterministic, and produces plain findings you can act on. See the live example on the homepage, then install the extension to review your own environment.
Website install extension Agent Environment Review findings upgrade
Govern your own environment
Install the extension, run an Agent Environment Review, and act on the findings — upgrade when you want unlimited reviews.