Ask most people what AI is and you get some version of a brain. A thinking machine. Something that knows things.
That metaphor is doing damage, and not because it overstates the technology. Because it points your expectations in a direction that makes you careless.
Here's a more useful one. AI is an intern who started Monday.
What is actually happening
Underneath the conversation, the mechanism is three steps.
Input. What you give it. Your prompt, your documents, your context.
Analyze. It runs that against patterns in what it was trained on, plus whatever direction you gave.
Output. An answer, a draft, a plan, an image.
It doesn't think for itself. It performs based on what it was trained on and what you asked for. That isn't a criticism. Performing well against a request is genuinely valuable and most of what any of us need on a given Tuesday.
But notice what's absent. No judgment about whether the request was a good idea. No knowledge of your specific situation beyond what you supplied. No stake in the outcome.
Which is exactly the profile of a bright intern on day three.
What the intern metaphor gets right
They are capable and fast. A good intern can produce a lot of work quickly, and much of it will be fine.
They do not know your context. They have not met your clients. They don't know that the board hates bullet points or that this vendor is a sore subject. You have to tell them, and if you don't, they will fill the gap with something reasonable and wrong.
They won't tell you when they're out of their depth. This is the important one. A new intern who does not know the answer often produces something confident rather than admitting they're lost, because admitting it feels worse. AI does the same thing, for different reasons, with the same result: fluent, well-formatted, incorrect.
They get better the more clearly you brief them. Vague instruction, vague output. This is a management problem, not a technology problem, and you already know how to solve it.
You are still accountable for their work. If an intern sends a client something wrong under your firm's name, "the intern did it" is not a thing you get to say. Same here.
Why the oracle metaphor hurts
If you believe you're consulting something that knows, then checking its answer feels like an insult. You wouldn't fact-check a calculator.
So people don't check. And AI's failure mode is specifically designed to slip past someone who isn't checking, because it isn't wrong in a way that looks wrong. It is wrong in complete sentences with confident structure and plausible detail.
The intern framing fixes this without requiring any technical understanding. Nobody feels rude reviewing an intern's first draft. It is expected. It is the job.
Where the metaphor breaks
It is imperfect and worth naming where.
An intern learns. Your corrections today change their work next month. Most AI tools do not carry that forward between sessions, which means you re-brief every time. Frustrating, and a real cost.
An intern also has skin in the game. They want to keep the job, and that motivates accuracy. AI has no such thing. It is not trying to be right; it's producing what fits the pattern. Those coincide often enough to be useful and not reliably enough to trust unsupervised.
And an intern will eventually tell you they're stuck. AI won't.
The practical version
Treat it like a new hire and everything else falls out.
Brief it properly. Give it the context it couldn't possibly have. Show it an example of what good looks like. Check the work before it goes anywhere. Don't hand it the tasks where a confident mistake would be expensive.
None of that's a new skill you have to learn. If you have ever managed a junior person well, you already have the whole toolkit. The mistake isn't lacking the skill. It is not realizing that this is the situation where it applies.
The metaphor you pick determines how much you check. Pick the one that keeps you checking.