Product

Why our agents ask permission

Team Gridwork · April 2026 · 5 min read

The most common question we get isn’t about capability — it’s about control. “What if it says something wrong? What if it spends my money?” Fair questions. Here’s our answer.

Every action an agent can take — send a reply, issue a refund, launch a campaign — has a setting: automatic, ask-first, or never. On day one, almost everything defaults to ask-first. Trust is granted by you, one toggle at a time, with evidence.
The limits are real, not guidelines. Spend caps are enforced in code: an agent can propose a campaign over budget, but it cannot run one. A refund inside your limit processes instantly; a dirham over it becomes a request to you.
And everything is logged, append-only: who asked, what ran, and what it cost. Handing real work to software you can’t inspect is a leap of faith — and faith is not a control system.
Put this to work

Your AI team can run this playbook from day one — free for 14 days.

Start free
Read next
The Gulf buys at 11pm. Who’s answering?The first ten seconds of a lead