Approval Gate
A point in an AI workflow where an action is suspended until a human reviews and approves, rejects, or modifies it.
Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.
What it is
An approval gate is a deliberate pause point in an agent workflow where a high-impact action (sending an email, modifying production data, paying an invoice) is held until a human signs off. The agent prepares the action with full context; the human reviews; the workflow either proceeds, gets sent back, or is closed.
Why it matters
Not every AI decision is safe to ship unattended. Approval gates concentrate human judgment on the decisions that need it, leaving the agent to handle everything routine. Done right, they shift human review from 'check every output' (which collapses under volume) to 'check the cases that matter' (which scales).
How it works
The workflow declares which steps require approval, by what role, with what SLA. A pending decision lands in an approval inbox with the trigger event, the proposed action, the agent's reasoning, and the trace. The reviewer approves, rejects with reason, or modifies. Reject and modify events feed the eval set so similar cases become self-handling over time.
Related resources
Approval gates that put a human in the loop where correctness, risk, or accountability actually require human judgment — designed as part of the workflow, not as a panic button bolted on after launch.
One tidy inbox where every action the AI wants to take — but is not allowed to take alone — waits for a human yes or no, with all the context attached.
Before a workflow change goes live, someone checks not just whether it works, but what new things it could touch — and the changes that could do real damage get an extra set of eyes.
Written rules for when an AI agent stops trying and hands the case to a human — so the handoff is consistent, explained, and adjustable, instead of a mood.