This is a roadmap preview, not a walkthrough you can run today.
Today every operational decision in VibeControls is a human reading a dashboard. An agent goes degraded and somebody notices the card. A sandbox burns quota past its TTL over a weekend and somebody eventually spots it. A catalog component drifts from what is actually deployed and nobody finds out until an incident. The platform has the data; it just does not act on it.
The pieces that make acting safely possible are already shipped, and they are the hard parts. Plan mode is a real human-in-the-loop gate that turns a proposal into something you read, annotate, and approve or deny before any tool call fires. The AI tool-event ledger already records what AI work happened, with a live stream into the app. Scheduled prompt dispatch already exists as a primitive for sending work to an agent session on a timetable. AI credits already provide a pre-flight budget check. And the AI workbench is already the surface those things live on.
What is designed and not yet built is the decision layer: a small set of proposal operations that reason over the state VibeControls already holds and emit a plan rather than a mutation. Agent remediation for a machine that is degraded or stuck awaiting configuration. Resource reclaim for sandboxes, tunnels and sessions past their TTL or plainly idle. Catalog reconciliation against what agents report is really running. Scorecard remediation for a component that dropped below its tier. Finding triage that proposes a severity and an owner drawn from catalog ownership.
The cadence on top of that is also designed and not built: an opt-in, default-off sweep per workspace, with a credit budget it will not exceed, dormant workspaces excluded so it does not spend money watching nothing, duplicate proposals suppressed while one is still open, and a kill switch. The rule for the first version is unambiguous — there is no unattended-execute path. Everything the loop produces is a plan waiting for a person.
We are describing this before it ships because the sequencing is the point. Plenty of tools add autonomy first and governance later. VibeControls built the review gate, the ledger, the identity boundary and the budget first; the decisions are what comes next, and they will arrive behind all four.
Ready to make this your story?
