Blog

Why AI multiplies judgement rather than replacing it

---

I have been using AI as part of how I work for several years now, well before it became the thing everybody talks about. Not as a subject I advise on. As a tool I actually use, most days, on real client work with real deadlines.

The most useful thing I have learned from that is unglamorous. AI gave me the output of a team. It did not give me the decisions.

That distinction sounds like a nice line until you watch an organisation discover it the expensive way.

What actually changed

The constraint used to be production.

If you wanted forty pieces of content, a full competitor analysis, a set of user journeys and a documented content architecture, you needed people, and people take time and money. That constraint did real work. It forced prioritisation. You could not commission everything, so you had to decide what mattered, and the deciding was where most of the value sat even though nobody billed for it separately.

That constraint has largely gone for a certain class of work. I can now produce in a fortnight what would have taken a team of four a couple of months. That is not an exaggeration and it is not a sales claim; it is just what the tooling does now.

Here is what nobody mentions. When production stops being the bottleneck, the bottleneck moves. It does not disappear. It relocates to the thing that was previously hidden behind it, which is knowing what is worth producing at all.

The failure mode this creates

An organisation buys capacity it has no strategy to direct.

You can see it happening. Content volumes go up and nothing improves. Reports get longer and decisions do not get better. Somebody generates forty pieces of collateral in a week and nobody can say which three of them mattered, because that question was never the tool's to answer.

I find this genuinely interesting rather than depressing, because it is such an old pattern in new clothing. The same thing happened when desktop publishing arrived, and again with content management systems, and again with marketing automation. Each time, the constraint that was removed turned out to have been doing useful work, and the organisations that thrived were the ones that already knew what they were trying to say.

The tool amplifies whatever judgement is already in the building. If the judgement is good, the amplification is remarkable. If it is absent, you get a great deal of confident output pointing in no particular direction, produced faster than anyone can evaluate it.

What AI is genuinely good at

I want to be specific rather than sceptical in general, because the sceptical-in-general position is as lazy as the enthusiastic one.

Research and synthesis, with checking. It will get through volumes of material in an afternoon that would have taken a week. It will also produce plausible things that are wrong, so the checking is not optional and the time saved is real but smaller than the raw figure suggests.

Production of things whose shape is already decided. Documentation, first drafts, structured content, repetitive analysis. This is where the leverage genuinely is.

Interrogating your own thinking. This one gets underrated. Asking it to argue against a position I hold has changed my mind more than once, mostly by surfacing an objection I had been avoiding.

What it is not good at

Knowing what matters to this organisation, this year, given what happened last year and who is in the room.

That sounds like a soft claim, so let me make it concrete. The decisions that determine whether a piece of work succeeds tend to be things like: which of these two problems do we solve first, given that the finance director is sceptical and the operations lead has been burnt before. Or: this recommendation is correct and the organisation will not execute it, so what is the second-best answer that will actually happen.

Those are not information problems. They are judgement problems, and judgement is built from having been wrong before in similar circumstances. The tool has no stake in the outcome, no memory of the last three attempts, and no sense of what this particular group of people will actually do.

What this means if you are buying

Two practical consequences.

Be sceptical of capacity as a value proposition. If somebody is selling you AI on the basis that it will produce more, ask what you will do with more. Most organisations are not currently constrained by output volume. They are constrained by decision quality, and adding output to an organisation with unclear priorities makes the priorities harder to see, not easier.

Look for where judgement gets applied, not where the tool gets applied. The useful question about any AI proposal is not what it automates. It is who is deciding what it should work on, how often that decision gets revisited, and what happens when it produces something confidently wrong.

The honest version of my own position

I use AI heavily and it has changed what I can deliver as one person. That is a real commercial advantage and I am not going to pretend otherwise.

But I have been doing this for more than forty years, and the thing clients are buying is not my production capacity. It is the accumulated experience of having watched a lot of technology decisions go well and badly, in commercial businesses, in government programmes and on charity boards. The tool makes me faster at executing that judgement. It does not supply it.

I would be suspicious of anybody whose pitch is mainly about the tooling. The tooling is available to everyone now, which is exactly why it has stopped being a differentiator. What has not become available to everyone is knowing which of the forty things you could now produce are the four that will make a difference.

---

If you want an independent view on where AI could usefully be applied in your organisation, an AI opportunity audit assesses that specifically. If the question is about oversight and responsible adoption rather than opportunity, AI strategy and governance is the closer fit.

---