Everyone wants AI transformation. Nobody wants to change Tuesday.

The kickoff meeting usually goes well. There is a slide with the word "transformation" on it, a sponsor who has read the right articles, and real appetite in the room. Then someone asks whether the new way of working will change the Monday pipeline review, and the temperature drops a couple of degrees. Not that meeting. That meeting works.

I have been consulting since 2015 and coding with AI assistants since 2022. Over that time the gap between what the technology can do and what organisations will let it do has only widened. I don't think that is anyone's fault in particular. I do think it is the most interesting problem in my line of work, and I have been on the wrong side of it more than once.

The brief asks for change. The calendar asks for continuity.

Most of the engagements I have been offered in the last few years share a shape. A leadership team wants AI to make the company faster, leaner or more responsive. They want it this year. And, without ever saying so, they want it to happen without moving a recurring meeting, retiring a report, or asking a single senior person to work differently.

Nobody puts that last part in writing. You learn it by watching where the energy goes. A proposal to let an agent triage incoming support tickets gets enthusiastic nods. A proposal that the head of support stops personally approving every refund above a small threshold gets a "let's park that for phase two". Phase two, in my experience, is where good ideas go to retire.

The trouble is that the parked item is usually the one that matters. An agent that sorts tickets in seconds achieves very little if every interesting ticket still waits for one person's sign-off. The speed piles up in front of the habit, like traffic at a level crossing.

I used to think the clients were the problem

For a few years I carried a quiet, slightly superior theory. Clients asked for change and then refused it, so the fault was theirs. I had the clear diagnosis; they lacked the nerve. It is a comfortable theory for a consultant, because it means every stalled engagement belongs to someone else.

It was also wrong, or at least lazy. When I looked honestly at the work that went nowhere, the pattern implicated me. I described the end state in loving detail and said almost nothing about the Tuesday afternoon when a sales manager has to stop doing the thing she has done for a decade. I treated habit as friction to be removed, rather than as something people had built, defended and often been rewarded for.

Habits at work are rarely stupid. The refund approval exists because, once, a refund went badly wrong. The weekly status meeting exists because a project once slipped without anyone noticing. Chesterton's fence applies: before you take something down, find out why it was put up. Most of the habits I was asking people to drop were fences, and I had been strolling past them with bolt cutters.

Advice is cheap to give and expensive to follow

There is a structural flaw in consulting that I didn't appreciate when I started. The consultant's work is finished when the advice is delivered. The client's work begins at exactly that moment. Everything hard happens after I have sent the invoice.

It is a version of the principal–agent problem, and it cuts both ways. I am paid for clarity, so I am tempted to produce a great deal of it. The client is paid, in salary and in standing, for keeping the place running, so they are tempted to absorb my clarity into a deck and keep the place running exactly as before. Both of us behave rationally. The company stays the same.

I have written recommendations I was proud of that changed nothing. You will know the type: a well-argued case for collapsing three approval steps into one, or a plan for letting an agent draft first-pass contract reviews so the lawyers can spend their time on the clauses that matter. Accepted, praised, filed. Nobody disagreed. Nobody had a spare week to act on them either.

The smallest habit is the honest test

These days I ask a different question at the start. Not "what would you like AI to do for the business?" but "which single habit are you personally willing to change next month?" It is a rude question, so I ask it politely. The answer tells me more than a day of discovery sessions.

If the sponsor names something real, such as dropping a weekly report they read but never act on, the work has a chance. If they name something that belongs to someone else, I know the change will be pushed downward until it lands on the people with the least authority to refuse it and the least reason to want it. If they can't name anything, I have learned to say so, and occasionally to step back. That is an awkward conversation for someone who earns a living from engagements. It is less awkward than six months of polite stalling.

The smallest habit works as a test because it is cheap. Nobody's job is at stake. It costs a little pride and a little convenience. If a leadership team won't pay that, they won't pay the larger bill that real change sends later, and it is kinder to everyone to find out in week one.

Why I started building instead of only advising

The other thing I changed was to stop handing over recommendations and leaving. Advice describes a different way of working. Software, if it is any good, is a different way of working. You can ignore a report. It is much harder to ignore a tool that already sits in the middle of your day, does the dull part of your job, and was shaped around how you actually do it.

That shift is most of why the studio now builds things rather than writing about them, and why we spend as much time on the weeks after launch as on the build itself. I don't think software beats persuasion because people are irrational. It wins because it changes the default. A habit is mostly a default nobody has had a reason to revisit. Put a better default in front of someone on a Tuesday afternoon and you no longer need them to summon willpower. You only need them not to go out of their way to avoid it.

It is not a magic trick. Built tools get resisted too, especially when they follow the official process rather than the real one. But the argument moves from the abstract ("should we work differently?") to the concrete ("this screen is missing the field I always check"). The second argument is one I can usually settle by Friday.

What I tell people now

I still run strategy sessions, and I still enjoy them. I have just stopped treating the session as the product. When we run workshops, the aim is not a vision. It is one workflow, one owner and one habit that will be different by the end of the month, written down with that person's name next to it.

I have also made peace with my own part in the dilemma. Clients are not irrational for protecting routines that have kept them safe, and consultants are not heroes for pointing at those routines from a safe distance. The useful work sits in between: close enough to the desk to see why the habit exists, and patient enough to replace it with something people prefer.

So before you hire anyone to change your company, try something cheaper. Pick the recurring meeting you would miss least, cancel it for a month, and see who complains. If nobody does, you have made your first real change for free. If someone does, you have found the person your AI project needs to win over, and you have found them before paying anyone to discover it.