Treat AI like your sharpest new hire, not a search engine

Google trained you to type three words and skim three links. That habit is the exact reason most CEOs get mediocre results from a genuinely extraordinary tool.

A CEO I know has had a paid Claude account for eight months. He uses it almost daily. His most common prompt is some version of "what do I need to know about X?" He gets back something decent, skims it, closes the tab, and moves on. He is convinced the tool is only marginally useful. His verdict is completely logical given how he is using it. And it is completely wrong.

The problem is not Claude. The problem is that he is running a 2006 Google habit on a 2025 tool built on entirely different physics.

Search and AI are not the same operation

Google indexes documents that exist somewhere on the web. You give it keywords, it ranks existing pages, it surfaces links. Short queries work because Google is pattern-matching your words against a database of words that humans already wrote. The shorter and cleaner the query, the less noise in the signal. Skimming three links works because the information is already there, formatted, and someone else already organized it.

A large language model does something structurally different. It generates a response by predicting the most useful continuation given everything you have given it. Nothing is being looked up. No documents are being retrieved. The model is producing new text, from scratch, on the basis of your input and its training. Understanding what an LLM actually is changes how you use it, because the two implications fall out immediately: the quality of the output is almost entirely a function of what you put in, and a thin input produces an average-case output, the same average output anyone else would get with the same thin input.

Google rewards the short query. AI punishes it.

The two decades of Google habit that live in your hands are not a neutral starting condition. They are the thing you have to override. You are not going in cold. You are going in trained in the exact opposite direction.

The model that actually works: a sharp new hire

The frame I have found most useful, and the one that unlocks all the right behaviors, is this: AI is like an extremely sharp generalist you just hired.

They know their craft cold. They can write, analyze, synthesize, draft, structure, pressure-test, and adapt fast. They know more about most topics than anyone you have on staff. But they started today. They know nothing about your business, your customers, your competitors, your strategy, your voice, or what you care about. And they will never tire, never have a bad week, and never resent being asked to redo something.

That framing is not a metaphor. It is a working model with direct behavioral consequences.

A new hire who is brilliant but has zero context is only as useful as the onboarding you give them. The CEO who hands a smart new employee a vague task and expects a brilliant output has made a management error, not a hiring error. The tool is not the variable. The brief is.

Joseph Fuller at Harvard Business Review put it plainly in his 2026 research on AI at work: the real challenge is not figuring out how to adapt to a new technology, it is primarily about managing work. Anthropic's own research on how AI is changing professional tasks reinforces the same frame: the people getting the most from these models are the ones who treat accountability for the output as theirs, not the model's. They delegate, they review, they redirect. The people who get the least are the ones who treat it as an oracle to consult.

The five things a good manager does, applied here

The new-hire model is only useful if it translates into different behavior. Here is what it actually means in practice.

Onboard it with context. A new employee on day one needs to know the situation before they can help. The same is true for AI. "Summarize my Q3 strategy" is a vague task handed to someone who has never seen your company. "Here is our Q3 strategy doc. Flag the three risks an outside investor would ask about in a board meeting" is a brief. The brief takes thirty additional seconds and produces something usable. The vague ask produces something generic.

Delegate outcomes, not keystrokes. The CEO who writes a strong job description gives the hire a deliverable, an audience, and a reason. "Produce a two-page memo for our board summarizing the competitive landscape in Southeast Asia, for people who know the industry but not our positioning" tells the model what success looks like. "Tell me about competition in Southeast Asia" does not. Describe what good looks like and who it is for.

Treat the first output as a draft, not a verdict. When a new hire turns in their first piece of work, you mark it up. You do not accept it wholesale and you do not throw it away. You send it back with specific notes: this section buries the lead, the third paragraph should be the first, you are missing the cost dimension. AI responds to exactly the same behavior. The second pass is almost always better. The third is often excellent. The CEO who stops at one mediocre output and concludes the tool is useless is the manager who never gave feedback.

Review it like a manager, not a student accepting a grade. This is the single most important behavioral shift. Your job is editorial judgment. The model is confident but not always correct. It does not flag uncertainty the way a junior hire would say "I'm not sure about this number, you might want to check." A hallucinated statistic arrives in the same confident voice as a real one. When you read AI output, you are reading a report from a capable but new hire: does the argument hold, are the facts ones I can verify, what is missing that matters. You are not grading it. You are directing it.

Build a standing brief so it is not permanently day one. The biggest tax on AI use, for most CEOs, is re-explaining context they have explained a dozen times before. Your voice, your priorities, your industry, your company's strategy, the terms you use and the ones you hate. A two-page brief that you paste at the top of every session, or that lives in a file AI can read automatically, is the investment that pays every single time you open it. An hour of writing it saves five minutes per session. By week four, you are ahead. The new hire is no longer day-one.

The before and after is not subtle

If the five behaviors above feel abstract, here is what they look like in practice.

None of these require any technical skill. They are management skills you already have, applied to a new kind of direct report.

The reason this actually lands for CEOs

Here is the part I want to say directly: you already know how to do all of this.

You have written job descriptions. You have given feedback on drafts. You have done 1:1s where you told someone their work was directionally right but needed sharpening. You have handed a smart person a deadline and a deliverable and let them figure out the approach. You have reviewed a report and asked "what is the source on this?" Those are the exact moves. The only thing between you and getting dramatically more from AI is recognizing that the skills are not new, only the context is.

The CEO who writes a clear brief, delegates a real outcome, marks up the draft, and asks where the evidence comes from already has every move. The only blocker is the Google habit, the reflex that says: short query, quick scan, close tab. Unlearn that specific reflex and the rest follows.

One honest caveat: what I am describing is for your own knowledge work. How you personally use AI on the work you do every day. This is not a prescription for dropping AI into org-chart boxes or replacing roles with a model. Those are different questions with different answers. Start with your own desk.

How to actually brief it

I am going to stop here instead of turning this into a prompt-anatomy tutorial. That exists: there is a whole piece on how to write a prompt that works, and it is worth reading once you have the mental model. The structure matters. But structure without the mental model is just a template. The mental model is the thing.

The short version, before you go read that: give it the situation, the deliverable, the audience, and the constraints. That is the brief. Everything else is refinement.

This is a port, not a new skill

A year from now, the CEOs who got the most out of AI will not be the ones who figured out a clever trick. They will be the ones who recognized that they already knew how to do this, applied an existing skill to a new context, and got sharper at it every week.

The generational asymmetry is real. The CEO who builds a working relationship with this tool, who has standing context loaded and gives real feedback and reviews output like a manager, is compounding leverage that the CEO who still types three words and skims three links is not. In eighteen months, that gap is hard to close.

If you want the sequence, start with where to actually begin as a CEO. If some of the beliefs in your way are on the list of common CEO myths, that piece is worth a read (it also covers the search-engine myth, but briefly; this is the piece that develops the mental model).

The move is simple. The next time you open AI, instead of typing a short question, paste the relevant document, name the deliverable, name the audience, and tell it what good looks like. See what comes back. Then mark it up and send it again.

That is the whole thing.

Andrew

Related reading

• How to write a prompt that works · the briefing mechanics, once you have the mental model
• What is a large language model? · why the physics of AI reward rich context instead of short queries
• The things non-technical CEOs get wrong about AI · six beliefs that slow you down, including the search-engine myth
• Where to start with AI as a CEO · the actual first steps and highest-value use cases
• What is a prompt? · the brief, defined