The Intern Test: a quick way to check whether your AI will actually work
Before you roll out any AI tool or agent, there’s a simple test worth running first. We call it the intern test, and it goes like this: could a brand new intern do this task with only the instructions, data and access you’ve handed the AI? If they couldn’t, the AI can’t either.
It sounds obvious. It catches more failed AI projects than almost anything else.
What the intern test is
Picture giving the job to a new intern on their first day. You hand over the instructions, the data and the access they need, then you say nothing else. They can’t come and ask you questions as they go. With just that, could they get the task done correctly?
That’s exactly what you’re asking of AI. It needs the same three things a new starter does, and no more than you’ve actually given it.
The three things AI needs
- Clear instructions – what you actually want, spelled out properly
- The data – the information it needs to do the job
- The access – the permissions to carry it out
Miss any one of these and the task falls over, whether it’s a person or a machine doing it.
An example
Say you want a Claude agent to check your project management software each morning and chase any project manager who’s fallen behind. Run it through the test:
- Instructions – have you told it clearly what “behind” actually means?
- Data – is the project manager labelled against each task, so it knows who to chase?
- Access – can it actually send the message?
If any of those is missing, your intern would be stuck on day one. So is the AI.
Why AI is different from an intern
Here’s the part that matters most. If an intern isn’t sure, they ask. Two contacts with the same name, data that’s obviously out of date, a step that doesn’t quite add up: a sensible intern checks before carrying on.
AI won’t. It doesn’t pause and it doesn’t ask. It just does the task with whatever’s in front of it, duplicates and old data and all. The intern would come and flag the problem. The AI quietly gets it wrong, and you might not spot it until it’s gone out the door.
How to use the test
Before you set any AI workflow going, ask yourself one question: could a sensible intern do this with nothing but the instructions, data and access I’ve given? If the answer’s no, that’s your job before you automate anything.
Nine times out of ten the gap isn’t the AI. It’s the instructions that were never clear, the data that was never tidied, or the access that was never sorted. Fix those and the AI has a real chance. Skip them and you’re just automating a mess.
So before you trust it with the task, run the test: would an intern manage this on their own?
If you want help steering your organisation in the right direction, get in touch.
