"Human on the loop vs human in the loop" sounds like wordplay, but it's a real design choice for AI agents. In the loop, a person approves an action before it happens. On the loop, the agent acts on its own while a person monitors and can step in. Most well-run agent deployments use both — for different actions. This guide sits alongside our human in the loop for AI agents pillar.
Human in the loop
A person must approve before the action runs. The agent pauses, an approver decides, and the decision is recorded.
Use it when an action is:
- Hard or impossible to reverse (payments, deletions, external messages)
- Above a threshold (large refunds, bulk changes)
- Unusual for this agent or customer
Trade-off: safer, but slower — and approvers can tire if it's applied too broadly.
Human on the loop
The agent acts autonomously within limits. A person watches activity — dashboards, alerts, regular reviews — and can pause the agent or reverse actions.
Use it when actions are:
- Low-impact and reversible
- High-volume, where per-action approval would be impossible
- Already bounded by strict rules
Trade-off: fast, but oversight happens after the fact. It only works if limits are enforced automatically and activity is visible.
Human in command
A third term you'll see, especially in policy discussions: a person decides whether and how an AI system is used at all — setting its scope, limits and when to switch it off. For agents, this is the owner who sets the rules.
Human on the loop vs human in the loop, side by side
| In the loop | On the loop | |
|---|---|---|
| When the human acts | Before each action | During or after, by exception |
| Speed | Slower | Fast |
| Best for | High-impact, irreversible actions | Routine, reversible actions |
| Depends on | Clear routing and context | Enforced limits and good visibility |
| Failure mode | Approval fatigue | Problems noticed too late |
Combining them in one agent
A refund agent might be:
- On the loop for refunds under $100 — allowed automatically, reviewed weekly.
- In the loop for refunds from $100 to $1,000 — held for the finance lead.
- Denied above $1,000 — never allowed.
The split is defined in your AI agent policy, the in-the-loop part runs through an approval workflow, and the on-the-loop part relies on an audit trail someone actually reviews.
What regulators mean by oversight
Human oversight appears in AI regulation — the EU AI Act's Article 14 requires certain high-risk systems to be designed so people can oversee them effectively. It doesn't prescribe one model for every action. See the EU AI Act and AI agents, and ask your legal team how it applies to you.
Where Gaten fits
Gaten supports both: rules decide which actions run automatically (on the loop, with every decision recorded) and which wait for a named approver (in the loop). See human approvals.
Frequently asked questions
What is the difference between human in the loop and human on the loop?
In the loop: a person approves each action before it runs. On the loop: the agent acts on its own while a person monitors and can intervene.
Which is better for AI agents?
Neither on its own. Use in-the-loop for high-impact, irreversible actions and on-the-loop for routine ones, with enforced limits in both cases.
What does human in command mean?
A person decides whether, where and within what limits an AI system operates — including when to stop it.

