The Hard Part of AI Adoption Isn't the AI
What I keep relearning about getting people to actually use the tools.
A firm can buy AI in an afternoon and still not use it a year later. The licences are easy. The habits are the whole problem.
I led our move from one AI tool to another, wrote the onboarding, and ended up teaching most of it one person at a time. The tool was never the obstacle. It is good, and it improves on its own. The obstacle is that "we adopted AI" and "we use AI" are different sentences, and the distance between them is entirely human.
Buying the licence is not the project
The two are easy to confuse. A firm buys seats, announces the rollout, ticks the box. Then you look at who is actually using it and it is the same three people who would have used anything, plus a long tail who logged in once, met a small snag, and slipped quietly back to the old way.
That first snag is where adoption dies. Not some grand objection to AI, just five irritating minutes: a connector that will not link, a setting that is not obvious, a first prompt that returns something flat and confirms the suspicion that this was hype all along. People do not push through that alone. They have a job to do, and the old way still works.
The README as a change-management tool
So the most useful thing I made was not clever. It was a setup guide that deleted those five minutes.
It walked through the install, the connectors worth linking, mail and calendar and files and chat and the document apps, and a short checklist of the extras worth turning on. The aim was narrow and deliberate: carry a non-technical colleague from "I have a login" to "I just did a real task" without their having to ask anyone. That is the entire trick. It looks like documentation, but it is change management, because the thing standing between a team and a new habit is almost never the concept. It is the onboarding nobody bothered to write.
Teaching up is the awkward, valuable part
A good share of the teaching went upward, which has its own texture. I made a walkthrough for our founder on building custom assistants, sat with a senior colleague through context windows and project folders and when deep research is worth the wait, and did the one-to-one troubleshooting that does not scale and is the only thing that actually moves a person from curious to capable.
There is a small discomfort in teaching your own manager a tool. There is also a great deal of value in it, and not the cynical kind. Senior people are usually the most time-starved and the least able to spend a Saturday prodding at settings, so a little hands-on help has an outsized effect on whether the thing takes hold across a team. It also means the people setting direction understand what they are directing.
The move I would recommend to anyone
Before recommending the tool firm-wide, I bought a personal account with my own money and lived with it for a while. Partly impatience, partly that I did not want my name on a recommendation I had not stress-tested.
It turned out to be the highest-leverage thing I did. A recommendation from someone who has actually used the tool, knows where it grates, and can answer the second and third question lands nothing like a forwarded feature list. The few pounds were the cheapest credibility I have ever bought.
What it comes down to
Nobody adopts a tool. They adopt a habit, and habits are taught one slightly annoying question at a time. The technology keeps getting easier, which only sharpens the point: the people who decide how much value a team gets from AI are the ones willing to do the unglamorous work of getting everyone else past the first five minutes. That is not an engineering job. It is a patience job, and it is the one I keep ending up with. ▮
“Nobody adopts a tool. They adopt a habit, and habits are taught one slightly annoying question at a time.” THE WHOLE ARTICLE IN ONE LINE