ChatGPT is like an intern

May 31, 2023

ChatGPT can produce an impressive amount of work in very little time. That makes it tempting to treat the output as a finished product. A better mental model is to treat the system like a capable intern: fast, broadly knowledgeable, eager to help, and still dependent on clear direction and careful review.

That framing makes AI more useful because it puts responsibility in the right place. The tool can accelerate the work, but a person still owns the goal, the judgment, and the result.

Start with a clear brief

An intern cannot read your mind, and neither can an AI system. A vague request usually produces a vague answer. Useful direction includes the audience, desired outcome, relevant context, constraints, and the form the answer should take.

Instead of asking for “ideas for a customer email,” explain who the customer is, what changed, what they need to do next, and how the message should sound. The quality of the brief determines how much useful work the system can do before it needs correction.

This does not mean every prompt needs to be long. It means the important information should be explicit. If a detail would change the answer, include it.

Delegate the right work

AI is strongest when the task has a clear shape and a human can recognize a good result. It can organize rough notes, compare options, draft variations, summarize documents, extract recurring themes, or turn an outline into a first pass.

Those tasks save time without asking the system to make an unreviewed decision on behalf of the business.

Open-ended strategic choices require more care. The system does not share your accountability, customer relationships, or knowledge of the consequences. Use it to expand the set of possibilities and pressure-test your thinking, then make the decision yourself.

Review the work, not just the writing

Polished language can create false confidence. A response may sound certain while missing context, inventing a detail, or reasoning from a weak assumption. Review should cover more than grammar.

Check the claims. Confirm that the recommendation fits the real constraints. Look for what was omitted. Ask whether the output serves the original goal. For important work, compare the answer with a trusted source or a subject-matter expert.

This is the same habit a good manager uses with any delegated work: evaluate the reasoning and the result, not merely the presentation.

Improve through feedback

The first answer is a starting point. Point out what is wrong, explain why, and ask for a targeted revision. Share an example of what good looks like. Break a complicated assignment into stages so you can correct the direction before the system produces a large amount of unusable work.

Over time, reusable instructions, examples, and review checklists turn one-off prompting into a repeatable workflow. That is where the larger business value appears. The goal is not to have a clever conversation with AI; it is to build a reliable way of getting useful work done.

Keep a person accountable

AI can draft, transform, classify, and recommend. It cannot own the outcome. A person should remain responsible for approving customer-facing communication, sensitive decisions, and anything where an error carries meaningful risk.

Treating ChatGPT like an intern is not a criticism of the technology. It is a practical operating model. Give it context, delegate deliberately, review the result, and improve the process. When those habits are in place, AI becomes less of a novelty and more of a dependable part of the team.