The question is usually put as whether AI will replace executive judgement. It is the wrong question, and the way it is framed gives away a misunderstanding of what was actually slowing organisations down.

Very few consequential decisions have been delayed because the analysis could not be produced. They are delayed because the consequences fall unevenly, the interests around the table genuinely differ, and somebody has to accept responsibility for a choice that cannot be proven right in advance.

It automates the part that was not the bottleneck

What AI does extraordinarily well is produce work that used to be slow: reading across more material than a team could hold, modelling scenarios in an afternoon, drafting the argument, finding the pattern nobody had time to look for. That is a real gain and it will keep improving.

But relieving a constraint that was not binding does not make an organisation faster. A board that sat on a decision for two quarters was rarely waiting on the modelling. Remove the analytical delay and what remains is the part that was always the difficulty, now with less cover. Organisations in this position often feel considerably busier while deciding at exactly the rate they did before.

It makes disagreement cheaper to defend

This is the effect leadership teams are least prepared for. Building a rigorous case used to take weeks of someone's time, and that cost imposed a quiet discipline: you only commissioned the work for a position you were serious about.

That discipline has gone. Every executive can now assemble a credible, well-evidenced argument for the conclusion they already preferred, before lunch. Where a team is genuinely aligned, this is excellent pressure-testing. Where it is not, analysis becomes ammunition, and the debate gets longer and better armed without getting closer to a decision. The technology does not settle a disagreement about what matters. It equips both sides of it.

The same mechanism operates privately. A model answers the question it is given, so a team looking for reassurance will now find it faster and in more convincing form than ever. Asking it to argue the opposite case is the obvious discipline, and almost nobody does it unprompted.

Where it genuinely helps

None of this is an argument for holding back. Used deliberately, AI is very good at the things leadership teams are reliably bad at: surfacing the option nobody raised, separating what is actually known from what is merely widely believed, exposing an assumption that has gone unexamined because it has been repeated for years, and producing a defensible read of an unfamiliar situation quickly enough to matter.

The word carrying that sentence is deliberately. A good deal of adoption runs the other way round, starting from the tool rather than the problem: a programme is stood up because the board wants evidence of momentum, and the use case is found afterwards to justify it. The work that follows is real enough and demonstrates progress convincingly, but nothing the organisation was actually struggling with moves. Teams getting genuine value from these tools can usually say which decision or which constraint they set out to improve, and would know whether it had.

It is also useful on the question of when to stop. Waiting has a cost, and a leader who keeps requesting more analysis can look rigorous while avoiding the decision. Better tools sharpen that temptation rather than removing it, because there is always another scenario available. Knowing how much uncertainty the organisation should accept before committing is a judgement about appetite and consequence, not a calculation.

What cannot be delegated

A model carries no exposure to the outcome. It will not sit in front of the board, explain the decision to the people affected by it, or live with the precedent it sets. The executive who will do all three reasons differently as a result, and that difference is not a flaw in their thinking. It is most of what judgement means.

Some choices are not optimisation problems at all. What the organisation is prepared to carry, what it will not do for commercially attractive reasons, how much disruption to impose on customers in pursuit of efficiency: these say what the business values, and answering them is the visible part of leading. So is accepting that the technically optimal path may be organisationally unachievable, or that the efficient one destroys a capability that will be needed later.

The practical implication is that accountability has to be placed deliberately rather than assumed. Which decisions can sit with a system and which must remain explicitly human. Where responsibility rests when a model informed a consequential call. How the team will notice when more analysis has quietly become a substitute for choosing. "The model recommended it" has never been a position a board accepts, and the more capable the tool becomes, the more clearly that line needs to be drawn in advance.

Where I drew the line

I have reason to be enthusiastic about these tools. The thinking behind this practice took years. Standing it up took under three months across technology, brand, legal, finance and risk, and it would not have moved at that pace without heavy use of them.

What is worth noticing is where they stopped being enough. A commercial and employment lawyer, a qualified accountant, a risk specialist, a branding specialist: each was brought in at the point where the question had stopped being what the options were and had become which one I was prepared to be accountable for. AI got me to that point faster and in far better shape than I would have managed alone. It could not make the call, and it could not carry the cost of getting it wrong.

The harder part was timing. These tools leave you feeling competent well past the point where you are, and the work they produce looks finished long before it is sound. The risk is not that you skip the expert. It is that you bring them in later than you should, having already made decisions you did not know you were making.

_____

Technology amplifies organisations. A leadership team that welcomes challenge will use these tools to test what it believes; one looking for confirmation will use the same tools to manufacture it, faster and more persuasively. The capability arrives identically in both cases. What differs is the judgement it is pointed at, which is the part that still has to be built rather than bought.

Written in September 2026. What these tools can do is moving quickly, and that part of this description will date. Where accountability sits is the part I expect to hold.