Think back to your last busy season. Not to what you learned that year, or which tools you bought, or the training your team sat through. Think about what you actually did differently when the work was at its heaviest.
If the honest answer is that you used a tool sometimes, when there was a spare moment, then you have learning. You do not yet have a change to your practice. I know the difference precisely, because I have had the conversion moment twice and failed the test both times.
I am writing this from a ship with no wifi, and my practice is running without me. Being outside the work is the only place that question can be asked honestly. Last Friday I wrote about AI returns that arrive in a currency nobody measures. This is the harder one.
I had the moment. It did not survive April.
May of 2024, a few days after the 30 April personal tax deadline, when Canadian practitioners are mostly lying down. I had been listening to podcasts, and the subject was AI, and I had not so much as used ChatGPT. I had been too busy.
What struck me was not the technology. It was that my own firm had stopped being the one that moved first. Our stack had barely changed in five years. We had become an ordinary cloud practice in a comfortable space, the exact opposite of the rule we had written down for ourselves: lead with technology.
Then I had something close to a vision. I was sitting at the side of a road with my head in my hands. In front of me my desk and my computer lay destroyed, papers scattered everywhere. To my left, a truck was driving away from the wreck.
I understood immediately that the truck was AI, and that I had to decide whether I was the man on the verge or the man driving. I would rather drive it, and if that meant taking my practice apart and rebuilding it, so be it.
I spent six months on it. More podcasts, formal training, an AI consulting group, then training my team, then starting on clients.
Then busy season arrived, and everything went back. We knew how to prompt by then. We had a new tool in the toolkit and we used it when we could. Nothing about how work moved through the firm had changed at all.
Then I did exactly the same thing again
In January, Anthropic put Cowork on the Mac. I bought a Mac Mini specifically to run it and took my first Claude subscription, and within a day or two I could tell this way of working was different in kind, not degree: the first thing I had seen that could restructure accounting work rather than accelerate it.
It was also busy season, so I told myself I would come back when the work eased.
Two conversions. Two reversions. Two years apart, same practice, same owner, and by the second one I knew more about this subject than most accountants in the country. That is not a knowledge problem.
Learning is not a change to the operating model
This is the mechanism, and it took me two rounds to see it. You can teach every person in your firm to use AI well and change nothing whatsoever about how work moves through it. Capability sits in people. Operating models sit in process.
The custom GPTs we built are the clearest tell. They worked. People liked them. You can build those all year and still have the same firm at the end of it, because none of them changed who does what, in what order, with what handoff, to what standard.
Peak load is the only test that cannot be gamed
Adoption in a quiet month measures interest. Survival in a busy season measures structure. Under real deadline pressure, people fall back to whatever the process genuinely requires of them, and everything that depends on somebody remembering, caring, or finding a spare twenty minutes loses to the deadline. Every time.
There is a line in one of our own roundups from February of 2026, taken from somebody else's survey data: the firms without foundations are tinkering, nothing compounds, nothing sticks, and when busy season hits the experimentation stops entirely. We published that. Then we went and did it.
Where we actually are
I would like to tell you that we have since transformed the firm. That would not be true.
The instability is real: what works today sometimes breaks tomorrow. Uptake across the team is uneven, and that is its own piece of work. It is getting better every month, and the new products arriving now are built with AI at the centre rather than attached to the side.
But adding AI on top of a process you have not touched is not the change. That is the thing I keep having to relearn, and the reason I keep having to relearn it is that the old process is still there, waiting, and in March it is quicker.
So here is the question I would put to you, which is the same one I have to answer myself next spring. In your last busy season, what did you do differently? Not what you knew. What you did. And in the next one, will the answer describe something that changed in your process, or only something that changed in your head?
If the honest answer is the second, that is what the AI Practice Transformation program is built to fix. The current cohort closes on 21 August: theaiaccountant.ai/transformation