AI raises the individual and narrows the collective

The decision that arrived before the meeting

At 9:18, the capital allocation review is effectively over.

The CFO has a one-page recommendation. The COO has a cleaner version of the same case. The product lead adds a market risk, but it is already covered on page two. Nobody reaches for the appendix. Nobody asks for the option that was cut from the deck.

The team approves the expansion in nine minutes. It will consume much of next year’s hiring plan and set the product roadmap for eighteen months. The presenter closes with twenty-one slides still hidden. It looks disciplined.

It may be. It may also be that the team has mistaken a shared starting point for a collective decision — and from inside the room, on the day, you cannot tell which.

Each of them prepared alone. Each used AI to test the logic, close the gaps and turn a rough view into a defensible case. The work genuinely improved. But if the same class of system helped four people frame the problem, rank the options and anticipate the objections, what reached the room was four polished versions of one starting point.

The contrary argument was not defeated. It was edited out before anyone met.

That is the leadership risk. When AI enters preparation, the quality of each contribution can rise while their independence falls. From that point on, fast agreement is no longer reliable evidence that the team converged.

A leadership team is not valuable because it can assemble several competent answers. It is valuable because different judgements can meet, be tested, and become a decision no individual brought in complete. If that difference disappears before the meeting, better preparation has removed part of what the meeting was for.

Better work can still make a narrower set

The gains are real. Experiments have found that people complete professional writing faster and to a higher standard, and that consultants working inside AI’s capabilities produce more, faster, and better. That is why preparing with AI is rational. It is also why the risk is easy to miss: every document can improve on its own.

Anil Doshi and Oliver Hauser measured what happened when the outputs were considered together. Writers with access to AI-generated ideas produced stories judged more creative, better written and more enjoyable — with the largest gains among writers who started with less creative ability. Those same stories were also more similar to one another.

The individual result improved. The range across the set narrowed.

They studied short fiction, not executives deciding. Extending it to leadership teams is an inference, not their finding. But it exposes a measurement error that matters in companies: we score the memo, the analysis and the recommendation one at a time. We rarely check whether the set still contains genuinely different judgements.

AI does not always narrow that set. A study published in 2026 put 791 Procter & Gamble professionals to work on real product-innovation challenges. Individuals using AI matched the performance of two-person teams without it, AI helped specialists reach beyond their usual functional positions, and variation in outcomes was preserved.

That finding does not weaken the argument. It locates it. AI can broaden what one person generates — in the P&G work its main benefit came during idea generation, while human judgement still mattered in selection. The study ran for one day, with individuals and newly formed pairs. It did not test a standing leadership team choosing under continuing authority, history and consequence.

So the claim is not that AI makes groups think alike. It is that polished inputs can no longer be assumed to be independent inputs. That assumption now has to be checked.

Sometimes the failure looks like quick unanimity. Sometimes it looks like grind: each function arrives with an AI-strengthened case that has already answered the obvious objections and has nowhere left to move. The surface differs. The loss is the same. The room receives finished positions when it needs judgements that can still change.

Get the unsynthesised version into the room

The answer is not less AI. It is a different sequence.

Before the synthesised deck appears, ask each person — before they see anyone else’s note — to write three lines: what they actually think, what evidence or experience they bring that may not be shared, and what would change their mind. If AI shaped the view, say what it added or removed.

Put those lines beside one another and you can see what was invisible before: whether the agreement arrived by different routes, whether one frame quietly set the boundaries for everyone, or whether four finished cases have turned up with nowhere left to go.

It also changes the leader’s job in the room. Do not reward agreement because it arrived quickly. Ask which judgement could only have come from finance, from operations, from product, from the customer, or from this particular history — and whether it survived long enough for the group to work on it. The point is not to manufacture disagreement. It is to see what is being synthesised before it has been.

Akheon works on the part that remains once the decision is live. It joins the conversation, reads whether distinct judgements are being worked into one owned decision or only repeated in common language, offers one thing when another move is still available, then reads what changes. It cannot restore a judgement that was never formed, or never reached the room.

The weekly leadership rhythm is where this risk becomes normal. Board preparation, the strategic pivot, the acquisition review, the session after a crisis — that is where it becomes expensive.

Before your next consequential meeting, don’t ask only for everyone’s best answer. Ask to see the judgements that existed before the answers were synthesised.

And if nothing in the room can be traced back to where it came from, don’t call the speed convergence. The team didn’t arrive aligned. It arrived narrowed.