AI should sharpen your thinking, not replace it.

You’re reading The AI Directive — practical AI intelligence for business leaders. Subscribe here if someone forwarded this.

The previous article dealt with a recognisable problem: a presentation can look generically AI-generated and leave the audience wondering whether the person delivering it genuinely thought the material through.

There is an awkward irony behind that reaction. The deck may be the result of hours of very human work: feeding in hard-won experience, explaining the context, correcting the model’s assumptions, pushing back on weak conclusions and sharpening the argument through repeated exchanges. AI may have produced the final slides, but the thinking was not produced in a single prompt.

All of that can be obscured by a generic AI layout or a few familiar phrases. #024 addressed the presentation problem. This article is about the more important process that may sit behind it: how to work with AI so the output is genuinely human-steered before making the language sound like the human.

That order matters. A style prompt can imitate the surface. It cannot supply the experience, context or judgement that gives the work its point of view.

The weakest pattern is simple delegation: provide a short prompt, receive a plausible answer and tidy it until it looks acceptable. It is quick, but most of the intellectual choices have been made by the model.

The stronger pattern is closer to sparring. Start with substantive source material and a clear outcome. Ask AI to explain what it thinks the problem is, then challenge anything it has misunderstood. Test the argument against awkward exceptions, introduce context the model could not know, reject generic recommendations and amplify the points that fit the evidence and your experience.

The process moves backwards and forwards. The human is not merely approving text; they are changing the model’s understanding until the output faithfully represents a view they are willing to own.

That can be dramatically faster than starting alone with a blank page, without being intellectually passive. AI accelerates research, structure, comparison and drafting. The human supplies direction, relevance and judgement.

Microsoft Research interviewed 19 professional writers as they worked with personalised and non-personalised AI writing tools, then gathered responses from 30 avid readers. The readers could not reliably distinguish AI-assisted work from writing produced by the authors alone and were generally positive about writers experimenting with the technology.

The writers focused less on detection and more on whether the work still expressed a self they had actively constructed. Personalisation helped, but copying the finished style was not enough. They wanted tools that supported their development and thinking rather than simply producing text on their behalf.

That is a better definition of authorship for AI-assisted business work. The question is not how many words the person typed. It is whether their context changed the answer, whether their challenges changed the argument and whether the final conclusion reflects a decision they actually made.

Microsoft Research describes a similar distinction between AI that simply finishes work faster and AI designed to deepen human reasoning. In that second role, the system can highlight tensions between sources, surface alternative interpretations and ask questions that push the person beyond their first answer.

This is where AI becomes more than a writing assistant. A senior leader can use it to interrogate an emerging position, explore the strongest counterargument, compare what the data says with what experience suggests and expose gaps before the work reaches an audience. The value comes not only from speed, but from having a tireless thinking partner that can be challenged repeatedly without losing patience. A useful executive colleague, in other words, but with fewer diary conflicts.

There is still a boundary. A 2026 study involving 20 science journalists found that participants were more comfortable with AI gathering information or offering feedback than generating the central idea or draft. They saw the distinction as one of agency: support the decision-maker without quietly becoming the decision-maker.

Once the argument is genuinely yours, AI can help express it in your style. Give it examples, correct the rhythm, remove language you would never use and make the final draft sound natural when read aloud. That is useful work, but it is the final mile rather than the source of authenticity.

A piece can sound exactly like you and still contain none of your thinking. It can also be substantially produced by AI while faithfully carrying your experience, context and conclusions. The difference is the quality of the human contribution throughout the process.

This also explains why #024 matters. If an organisation uses AI well but allows the final presentation to collapse into generic imagery, repeated layouts and familiar phrasing, the packaging can undermine the human work behind it. The answer is not to disguise AI involvement; it is to ensure both the thinking and the presentation reflect the person and organisation that produced them.

Before asking AI to “make this sound like me”, ask three questions: what context did I provide that the model could not know, what changed because I challenged its first answer, and which conclusion in the final output is genuinely mine?

If those answers are weak, style is not the problem yet.

This is the second of three articles exploring how organisations earn trust in AI-assisted work.

Next: How to make AI-assisted work trustworthy.

See you Tuesday.
— Toby

These are tools I use personally. Affiliate links marked - I earn a small commission if you sign up, at no extra cost to you.

Some links in this issue are affiliate links. I only recommend tools I actually use.