For the last couple of years, an often-repeated piece of advice about working with AI has been some version of the same line: treat it like your intern. Treat it like your entry-level employee. Give it small, well-scoped work, and check everything it hands back.
As far as advice on interacting with AI tools goes, it is fine. However, as a mental model for how AI fits into a company, it is fundamentally wrong – and the way it’s wrong matters more than it first appears.
This Isn’t An Argument About Capability
Let me get this obvious objection out of the way: It’s not that AI isn’t capable of completing this or that kind of task that we would like it to handle for us. Nothing I’m about to say depends on AI’s effectiveness.
The problem with treating AI like your entry-level employee is simply this: an AI isn’t an employee.
AI is not an individual who you are helping to grow into a professional capable of driving the future of your company: it will never take on more and more responsibility, be accountable for its decisions, or have a standing in the company. The bright-eyed intern you hired will, and the future of your company depends on them having the support to be able to do so.
The Credit Problem
Let’s look at a practical example of the difference between our relationship with our AI and our intern: saying “AI did this” vs. saying “the intern did this”.
Every one of us was taught early not to put our name on someone else’s work. This is especially true for the work of junior employees. The rule exists for a good reason: you nurture talent by giving credit, and you destroy talent by denying credit. However, this rule exists for the sake of people, as part of a company culture that realizes the full benefits of growth and diversity. It does not extend to software, since there is no career on the other side of the sentence being helped or harmed, no future being aided or hindered.
Attributing your work to AI isn’t generosity, it’s more akin to a CPA giving the calculator credit for doing someone’s taxes.
The Agency Problem
There is an additional cost hidden in the intern framing, one that directly saps the value from AI tools.
When you ask a colleague how they produced something impressive and the answer is “oh, AI did this for me,” the conversation is over. Compare that with the conversation that follows from “oh, I used AI to do this.” The difference is subtle but key. Our colleague is now claiming the ownership of the process used to produce the result. When asked “How” the conversation will naturally flow: “Here are the tools I chained together. Here’s the draft it gave me and here’s what I edited to get this.” That version is repeatable and makes the next person on the team faster.
The intern framing quietly licenses the first answer. Nobody expects a detailed methodology when a person did the work – “Sarah handled it” is a complete sentence. If I need the details of how she did so, I can go talk to her. But a tool has no autonomy to defer to.
If you used a tool, how you used it is the contribution, and withholding it isn’t modesty, it’s a gap in the team’s knowledge.
Repeatability Is the Discipline
This matters more with AI than with almost any tool we’ve used before, because the models underneath are nondeterministic. The same request does not reliably produce the same result. That means the edges – the prompt, the context you supplied, the tool configuration, the review step, the edits you made after – are where the reliability actually lives. They’re the part worth writing down.
The good news is that the tool will help you write them down. Ask it to document the steps you just took, or to lay out the plan before you start, and you have most of the artifact already. Closing that loop costs very little and compounds quickly.
Say It Plainly
None of this is an argument for using AI less. But it hopefully is an encouragement to describe what you did accurately, because accuracy is what makes the work repeatable and the credit honest.
Not “AI did this.”
“I did this. Here’s how.”
