AI glossary · Agents
Human in the loop
In one sentence
Human in the loop (HITL) is a design in which a person reviews, approves or corrects what an AI system proposes before it takes effect, rather than letting it act alone. It lets you give an agent autonomy step by step.
Also known as: HITL · human oversight · human-in-the-loop AI · human review of AI
Before you let an AI agent reply to customers on its own, it makes sense for someone to see what it’s about to say. That, in essence, is human in the loop.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingWhat is human in the loop?
It’s a way of splitting work between AI and people. The system does the heavy lifting (reading, sorting, drafting, proposing) and a person steps in where their judgement counts: they approve, correct or reject. Their corrections also show where the agent’s instructions need adjusting.
Don’t confuse it with a human handoff. In a handoff, the agent recognises it shouldn’t carry on and hands the conversation over. With human in the loop, the person and the agent work on the same task together: the agent proposes and the person signs off.
It’s the opposite of a fully autonomous agent, the far end of agentic AI. Between the two there are degrees, and choosing the right degree is the key decision.
Related to this term
UTIO agents
- Answer in the customer's language
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
The usual approach is to define autonomy levels and assign one to each channel or type of task:
| Level | What the agent does | The person’s role |
|---|---|---|
| No AI | Nothing | Does all the work |
| Ask first | Checks before replying | Decides what happens |
| Draft | Prepares the reply | Reviews and sends it |
| Automatic | Replies and keeps a record | Spot-checks afterwards |
The middle levels are human in the loop in the strict sense. Automatic is closer to human on the loop: the person supervises and can step in, but doesn’t approve every action. Guardrails complete the picture: they set what the agent must never do at any level.
The rule of thumb is to move up one level at a time, task by task, once reviews show the agent gets it right, and keep human approval on anything sensitive.
Example
A small accountancy practice in Bristol gets dozens of client emails a day about deadlines, paperwork and invoices. It puts the agent on draft: the agent reads each email, prepares a reply using the client’s details and leaves it ready. An assistant reviews and sends it, often without changing a word. After a few weeks, replies about opening hours and paperwork move to automatic; anything about amounts owed or penalties still goes through a person.
Human in the loop at UTIO
At UTIO, email has gradual autonomy: no AI, ask first, draft or automatic (automatic by default). With the WhatsApp agent you can start with no AI, using just the shared inbox and the CRM, and switch it on whenever you like. With the email agent, you can start on draft: the agent writes the reply in your tone and you review and send it. Once you see it gets things right, you move it up a level. On WhatsApp and voice, a person can pause the agent or take over the conversation at any time.
The 14-day trial lets you watch it work before giving it more room: with email, starting in draft mode if you like; on WhatsApp and voice, with the inbox in view so you can take over the conversation yourself.
Frequently asked questions
What's the difference between human in the loop and a human handoff?
In a handoff, the agent passes the conversation to a person and stops taking part. With human in the loop, the person doesn't replace the agent: they review or approve what it proposes, such as a draft reply, and the agent keeps working.
When should you keep a human in the loop?
When an agent is new, for decisions with real consequences (money, health, personal data, complaints) and whenever a mistake would be costly. Over time, tasks the agent gets right consistently can move to automatic.
Does the EU AI Act require human oversight?
The AI Act requires human oversight for systems it classes as high-risk. A customer service agent doesn't usually fall into that category, but oversight is still good practice. For your specific case, check the official text and your adviser.
Related terms
- Human handoff
Human handoff is the point at which a chatbot or AI agent passes the conversation to a member of staff, together with the context it has gathered, because the case needs it, a rule requires it or the customer asks for it.
- Guardrails
Guardrails are the safeguards placed around an AI model to keep it within agreed limits: checks on what goes in and comes out, limits on topics and actions, and rules for handing a case to a person when needed.
- Agentic AI
Agentic AI is AI that pursues a goal instead of just answering: it plans steps, uses tools, checks the results and decides what to do next with a degree of autonomy, within rules set by a person.
- AI agent
An AI agent is a system that uses a language model to understand a request, decide which steps to take and carry them out with real tools, such as your calendar or CRM, until the task is done rather than just replying.
Updated on