Digee / Wiki / Introducing AI
Explainer
Human-in-the-loop: a person decides.
Human-in-the-loop is the principle where AI prepares the work and a person checks and approves it. It is the safest way to start using AI in day-to-day operations – and often the fastest, too.
In short
- The agent prepares, a person decides – wherever money, terms or sensitive data are involved.
- Limits are set in advance, separately for each task.
- As trust grows, the agent can gradually do more on its own.
On this page
The principle in one example
An enquiry comes into an estate agency. The agent reads it, gets the missing details from the prospective buyer, matches it against the listings and drafts a reply with viewing times. The estate agent just checks it and sends it.
The AI agent did 80% of the mechanical work. The decision – what to offer and to whom – stayed with a person.
Three levels of autonomy
| Level | What the agent does | What the person does | When to use it |
|---|---|---|---|
| 1. Draft | Prepares a document or reply | Checks, edits, sends | At the start and for sensitive matters |
| 2. Exceptions | Handles routine cases on its own | Deals only with what the agent flags | Once the agent has proved itself on simple cases |
| 3. Oversight | Works independently | Spot-checks the summaries | Only for low-risk tasks with clear rules |
Most business tasks start at level 1. Moving up is a management decision, not a technical one.
Where a person should always decide
- Money – payments, discounts, quotes, complaints and refunds.
- Terms and commitments – contracts, promises to customers, deadlines.
- Sensitive data – health, personal circumstances, children.
- Unclear situations – complaints, conflicts, anything new.
- Decisions about people – choosing candidates, performance reviews, access to services.
The last point also has a legal dimension: GDPR restricts decisions based solely on automated processing, and the EU AI Act requires human oversight of high-risk systems.
How to set the limits in practice
- Describe the task step by step, as you do it today.
- For each step, decide: does the agent prepare it, or does a person decide?
- Define when the agent must always hand over (e.g. an amount above a limit, a complaint, missing information).
- Decide who approves and how quickly.
- After a month of use, review where the agent made mistakes and where it could do more.
This is exactly how we work when rolling out an agent – see how we work.
Frequently asked questions
Doesn’t the time saving disappear if a person has to approve everything?
Usually not. Checking a prepared draft takes a fraction of the time it takes to prepare it by hand. And over time, routine cases can go through without approval, leaving people to handle only the exceptions.
Who is responsible if the agent makes a mistake?
The business, just as it is for the work of an employee or any other tool. That is why important steps require human approval and the agent’s work is logged.
How do I know the agent can do more on its own?
When it has made very few mistakes on a task over a longer period and those mistakes are easy to fix. The decision should be based on data from real use, not on gut feeling.
Want to know where AI would save you time?
We go through a normal week with you and find the first task worth handing to an agent. No obligation.
Related articles
What is an AI agent
An AI agent does more than answer: it finishes a task. It reads an email, finds the details, prepares the output and hands it to a person to approve.
→Explainer · 3 minAI hallucinations: why AI makes things up
Why AI sometimes states something untrue with full confidence, and how to handle it in a business: your own documents, cited sources and human checks.
→Overview · 3 minAI Act: what the EU regulation on AI means for businesses
The EU Artificial Intelligence Act in plain English: risk categories, application dates including the 2026 delay, obligations for businesses and fines.
→