Yesterday I went to E-COMMERCE LIVE! Full of marketers, techies (a lot of them!) and business owners. And everywhere: AI.

One of the talks that stuck with me was from Just Eat Takeaway.com (Danke Felix).

They do not call or see their AI agents "tools". Not "models". Not "solutions".

They called them interns.

And honestly: that is the best explanation I have heard in a long time.

The intern metaphor fits exactly.

A good intern is motivated, quick and eager to learn. He picks up the tasks you give him and works harder than you expect. But he also needs guidance. Clear instructions. And someone to check his work before it goes out the door.

AI agents are exactly the same.

They work at enormous speed, "intern on speed", as Just Eat put it. But give them vague instructions, bad data or no boundaries and they take shortcuts that cost you dearly later on. That is not a bug. That is a property of the system. One you have to allow for as the person in charge.

What Just Eat learned, and what you can apply tomorrow.

There were three hard lessons on their slide.

I am adding my own observations:

1. Quality varies when instructions are vague An agent does what you say, not what you mean. The more detailed your work instructions, the better the result. Treat your prompt as an induction programme.

2. Too much data is a problem too Faced with data overload, agents take shortcuts to save tokens. Give them focused input, not an encyclopaedia.

3. Results are not always reproducible That feels uncomfortable if you are used to Excel. But it is the reality. Always ask for a "backsheet", let the agent show its reasoning, not just the conclusion.

4. Build the checks in, not afterwards Validation is not a luxury. It is part of the workflow. Same as with a work placement: you do not let a junior send customer emails without a review.

5. Start small, make it concrete The companies winning with AI today are not the ones with the biggest budgets. They are the ones that set up one process properly, share the wins internally and then scale up.

The real question

You do not have to become an AI expert to understand this.

You do have to realise that an agent is only as good as the leader behind it.

Because AI is not an IT project.

It is a leadership question.