AI Solved the Tasks We Gave It. It Exposed the Decisions We Had Not Made.

I started looking at this seriously in February 2024, and I started from a position of no advantage whatsoever.

I had been at an executive event where somebody was presenting on how this would change business. I am not a technical person. I had no idea how to start the process or where to start it. The question I kept hearing asked, and could not answer for my own business, was who is supposed to drive this: the technology function, the business owner, or some expert nobody has hired yet. I realised I was well outside my capability.

Two years later it runs through everything I do, and almost nothing I have learned is about the technology.

We did not start with AI. We started with frustration

The first useful decision was to stop asking what this could do and start asking where we were wasting time.

Where are people spending hours on things they hate. Where do errors keep happening. What would I hand to an assistant tomorrow if I had one. Nobody in the room needed to understand anything technical to answer those, and the answers pointed at unglamorous places: research before client meetings, meeting summaries, marketing content.

That is still the advice I give, and I give it because the alternative fails so reliably. Organisations that start from the tool end up with a subscription and a pilot. Organisations that start from a frustration end up with something people use, because nobody gets excited about AI and everybody gets excited about saving an hour.

The early projects were deliberately unimpressive. No agents, no chains, nothing anybody would present at a conference. They worked, everyone could see they worked, and that produced belief, which turned out to be the scarce input.

Marketing was not a marketing problem

Then it started telling me things I had not asked it.

We pointed it at marketing because marketing was slow. In 2024 we had no in-house marketing capability, so content creation fell back to me. I hated doing it, and I became the bottleneck.

What it revealed was that marketing was not slow for any reason to do with marketing. Every post, every article, every piece of content had to come through me. I was the constraint, and I had been the constraint for years without ever having decided to be.

AI did not solve a marketing problem. It exposed a dependency problem.

That sentence is the closest thing I have to a general lesson from two years of this. The technology did the task we gave it. What it uncovered was a decision nobody had made: in this case, that content required my judgement, which was never true and had simply never been examined.

Once you have seen it in one place you start seeing it everywhere, and almost none of what you find is technical.

I have argued elsewhere that where experience is not there, a leader has to get in and get their hands dirty, and that if you do not handle something directly it does not happen the way it needs to. I still hold that, and it does not sit against what I have just described.

The difference is the stage the organisation has reached. Where capability is genuinely absent, involvement is the job, and standing back in the name of empowerment hands somebody a mess they then have to carry. Where capability exists, that same involvement stops being support and hardens into dependency. What I had missed about marketing was that we had crossed from the first stage to the second some years earlier and my own behaviour had not moved with it.

Reading the stage wrong is expensive in both directions. Most of the leaders I meet are wrong in the second one, and are certain they are being helpful.

Everything it did next uncovered something

We built meeting summaries to save time after client conversations. Transcript in, summary out, the client's problems set out in order. We started sending them back to the client, and the client would either say that is exactly our challenge or no, you have misunderstood this part.

The value was never the summary. The value was finding out whether we had actually understood the client, which we had previously assumed and never tested. What looked like an efficiency tool turned out to be a diagnostic on the quality of our own listening.

Reporting on a coaching programme with more than twenty participants used to take me over a week. It now takes under an hour, and the quality went up rather than down. Because coaching involves sensitive material, the whole thing runs inside a closed environment, which is a decision I would make again before any other. That is a striking number and it is not the interesting part. The interesting part is what it revealed about the year before, when I had spent close to half of it building customised solutions by hand. I had not been spending my time where I created the most value, and no efficiency review had ever told me so.

We built something to support the people doing business development, because they were strong at relationships and less confident on the technical side of a proposal, so they came to me. It did the research and drafting we expected. What surprised us was that it became a development tool. Previously a junior person brought me a problem, I thought about it, I produced something, and nobody ever saw the thinking. Now the reasoning is visible: why this approach, what was considered, what was rejected. For the first time we could scale expertise rather than just output.

That is not a technology outcome. It is the correction of an organisational design fault that had been invisible because it was wearing my name.

I will put this more plainly than is comfortable, because the honest version is more useful than the flattering one. In the coaching work it now acts as a second brain. It holds everything that has been said across an engagement, helps me prepare a session, surfaces patterns I had not connected, and offers angles I would not have reached on my own. It has widened what I have to work with.

What it does not do is any of the things the work actually turns on. It does not decide which of those angles matters for this person, in this organisation, this month. It does not know what it costs someone to answer a question honestly, or when to press and when to leave something alone. It cannot read what is happening in a room, and it cannot carry the consequence of an intervention that lands badly.

So it has expanded the perspective available to me and left the responsibility exactly where it was. The judgement, the challenge, the reading of context and the decision to intervene are mine. They are what I am accountable for, and they are the part that was always the job.

What it exposes in other people's organisations

I have had a version of the same conversation with a great many senior leaders now, and two things come up almost every time.

The first is ownership. Most senior leaders already know this matters and have taken their first steps, so that is not the issue. The issue is that the organisation is not set up to implement it. No clear owner. No dedicated headcount. Limited visibility. And a great deal of activity happening in the shadows.

Nobody is clear on whose job this is. Not the technology function alone, not human resources alone, not a few enthusiasts. Everybody owns a piece and nobody owns the whole thing, which is a reliable way to produce a year of activity and no change. When everyone is partially responsible, nobody is fully accountable.

The second is the shape of the conversation itself. Sit with an executive team and the discussion goes almost immediately to compliance, security, legal exposure and what must not be permitted. All of that is necessary. What is striking is the ratio. Many organisations are spending more time discussing what this cannot do than what it could do, and that is not a governance position. It is a way of postponing a decision about value while appearing responsible.

The governance conversation has largely settled on one principle, which is to keep a human in the loop. It is a sensible principle and I am not arguing against it. It is also a statement about output. It asks whether the thing that came back is correct.

That is a reviewer's question rather than a leader's. What determines whether any of this is worth the money is whether we are working on the right problem at all, and no amount of checking outputs will surface that. You can keep a human in the loop on a workflow that should not exist.

Neither of those is a technology problem. Both are decisions sitting unmade in front of people who are entirely capable of making them.

The question the agents asked me

For the first year most of what we built replaced small things. Admin. The specific tasks people named when I asked what they hated. Useful, unremarkable, and easy to explain to anybody.

Then we started linking agents together so they ran whole processes rather than single steps, and the character of the thing changed. We built one that can take in a client need, match it against what we do, and produce a proposal. In the past that would have involved several roles and several days.

I want to be careful about what I claim from that, because the interesting question is not the one people expect. The real leadership question is not whether this replaces jobs. It is how many people you need when the workflow itself is run by something else.

I do not have a settled answer. What I am certain of is that it is a question about organisational design rather than technology, that it arrives sooner than leadership teams are ready for it, and that a quiet hiring freeze is a way of answering it without admitting you have.

What did not work

I should be clear that this reads much tidier than it was.

We started a good number of projects and found the technology was simply not ready for what we wanted, so they went on hold or were abandoned. We got distracted a lot. We went down a great many rabbit holes, some of them for weeks, chasing something that turned out to be a demonstration rather than a capability.

I do not regard any of that as waste, and I would tell another leader to expect the same ratio. But it is the reason I am unmoved by organisations waiting for a clear business case before they begin. You do not get a clear business case from analysis. You get it from having built six things, four of which did not work.

The thing I got wrong

The most useful thing I can tell another leader is where my own position has had to change.

Through 2024 and into 2025 I believed the answer was distributed ownership. We had moved from me directing the work to a model where almost anybody in the business could pick up an AI project and run it, and I said so publicly. It was working.

Then 2025 got busy.

2024 had been a quieter year. People had space, and we made enormous progress. In 2025 the business was busier, we had the same people, and the AI work became secondary. Progress slowed and then largely stopped. Not because interest disappeared. Not because we ran out of ideas. The time disappeared.

What I had actually built was a model where this was everybody's second job. That holds until the first job needs all of somebody's day, and then client work wins. Every time. It is the correct decision by each person individually and it produces a stalled programme collectively.

This does not compete with your technology. It competes with your calendars.

I include myself in that, and it is the part I find least comfortable to say. I have several projects I want to build. I know precisely what they would do and roughly what they are worth. I have not built them. The challenge stopped being ideas a long time ago and became finding enough uninterrupted time to execute them.

So the exposed decision here was mine: whether this work gets dedicated capacity or is expected to happen in the gaps. I had never made it. I had assumed enthusiasm would substitute for it, and enthusiasm is not a resource allocation.

Saved time is not capacity until somebody decides it is

There is a related decision that almost every organisation skips, and I have been saying it since the first of these talks.

You will save time. That is the easy part. What happens to the time is a separate question and it will not answer itself. Left alone, saved time gets absorbed. The same work expands gently into the space and nothing about the business changes except that people are slightly less pressed.

So we started asking people directly what capacity they had recovered and where it had gone, and then putting it somewhere deliberately. Expect resistance to that, and expect it to feel intrusive, because you are effectively reclaiming something people have quietly come to enjoy. Do it anyway, or accept that you have bought an efficiency you will never see.

What changed about the work itself

The last thing, and the one I did not anticipate at all.

For the first year or so I used this before the work: preparing proposals, analysing clients, designing sessions, writing reports. Somewhere in the last twelve months it moved inside the work.

In a workshop I used to take extensive notes, capture the themes and analyse afterwards what people had said. That now happens as we go. More significantly, I used to ask a group to reflect, brainstorm and structure a problem between themselves. Now I will often ask them to put the problem to the machine directly, and then we critique what comes back together.

That changes what I am for in the room. Less time producing content, more time challenging thinking, making sense of what is in front of us, and deciding what to do about it.

The wider point is the one I would leave a senior audience with. A workshop used to be limited to the intelligence in the room. It is no longer. We are now able to draw on the intelligence outside the room as well, which means the constraint on the quality of a discussion has moved from what the participants happen to know to whether anybody present can tell which of the available answers actually matters.

That is judgement, and it has quietly become the scarce thing.

Why this is a leadership problem

I have said publicly that this is no longer a technology challenge, it is a leadership challenge, and I want to be precise about what I mean by that, because the phrase is also available as an excuse for a senior person who would rather not open the thing.

I do not mean that the technology is unimportant, or that a leader can have opinions about it without using it. I used it daily for two years before I trusted my own judgement on any of this.

I mean something narrower and more uncomfortable. Every single thing this uncovered in my business was a decision that had been sitting unmade, usually for years, in a place where only I could make it. Who content has to go through. What my own time is for. Whether expertise lives in a person or in a system. Whose job this is. Whether the work gets real capacity or the gaps in somebody's diary. What happens to the hours we save.

None of those questions were created by the technology. They were all made visible by it, and made harder to keep ignoring.

That is the actual change. It did not hand leaders a new set of problems. It removed the cover from an old set.