Enable Shadow Mode on any Cerebral. Run it for a day, a week, or six months. Every decision is visible before it ever touches your systems. You see what it would have done, with full context. You review on your timeline.
When you see something you trust, approve that action class. Build confidence one verb at a time. reamaze:send_reply approved. shopify:cancel_order still shadowed. Nothing moves to production until you explicitly say so.
One toggle. The Cerebral that has been silently working beside your team starts executing for real — with the exact behavior you already verified over days or weeks of shadow operation. No surprises.
It's trust. Your team doesn't trust it yet. Your customers are real. Your operations can't absorb mistakes. Shadow Mode is the answer to that problem — not documentation, not demos, not sandboxes with fake data.
You run the Cerebral against real tickets, real orders, real customers — and you watch what it does. Then you decide. That's it.
Run Shadow Mode for 2-4 weeks alongside your existing team. Compare what the Cerebral would have done vs what your agents actually did. When the agreement rate is where you want it — go live.
Any action class you approved in Shadow Mode stays approved when you go live. Any action you kept shadowed stays blocked until you explicitly release it. Your governance configuration transfers completely.
Shadow Mode operates verb by verb — not all-or-nothing. A Cerebral can be live on read operations, shadowed on writes, and blocked on high-risk verbs simultaneously. Granular control at every step.