When BJ Fogg wrote Tiny Habits, his central argument was deceptively simple: behavior change doesn't come from motivation or knowledge. It comes from designing small successes that make the next attempt feel worth making. Celebration — feeling good in the moment — is what wires the habit in.
Most GenAI rollouts ignore this completely.
Access gets granted. Training gets delivered. Use cases get circulated. Then, firms wait for adoption to follow.
But there is a gap between "I attended the training" and "I reach for this automatically" — and that gap doesn't close on its own. It has to be designed.
The user is part of the system
GenAI is unlike any software tool that came before it. A document management system can be underused and still function correctly. A timekeeping system can be poorly adopted and still produce the expected output.
GenAI doesn't work that way.
The quality of the output depends directly on the quality of the engagement. A confident user and an uncertain user can sit in front of the same tool and have completely different experiences. The confident user iterates, adds context, refines. The uncertain user gets a generic answer, concludes it doesn't work, and stops.
That conclusion spreads. Not through formal resistance — through hallway comments.
"I tried it. It wasn't that helpful." "I had to redo most of it anyway." "I don't really see how it fits my work."
These aren't irrational responses. They're often accurate descriptions of a poorly designed first experience. And for CTOs, CInOs, and COOs managing a firm-wide rollout, a cluster of those early impressions is far more damaging than a failed pilot.
Confidence comes from doing, not knowing
Training tells people what GenAI can do. It doesn't necessarily give them the experience of getting good at it and feeling good about using it— and that experience is what actually changes behavior.
Fogg's insight applies directly here: people don't build habits from understanding possibilities. They build habits from feeling progress. From the moment a prompt produces something genuinely useful. From realizing they got a better answer because they asked a better question.
That feeling — small, specific, real — is what makes someone open the tool again tomorrow.
The implication for firms is that the early use cases matter enormously. Not because they showcase capability, but because they determine whether attorneys accumulate small wins or accumulate disappointment.
A use case that's too broad ("draft a client alert") produces generic output and a shrug. A use case that's tightly scoped — "suggest three possible structures for this alert based on these facts" — produces something evaluable, improvable, and instructive.
The difference isn't the tool. It's the design of the experience.
The 30-day question
The right question isn't "how do we train them?" It's "How do we design the first 30 days of usage?"
That means choosing initial tools and use cases where visible improvement is likely. It means helping attorneys understand not just what to ask, but what to do when the first answer isn't good enough. It means creating enough low-stakes repetition that the habit has a chance to form before the pressure of a real deadline arrives.
When that's done well, something shifts. The tool stops being something attorneys are supposed to try. It becomes something they reach for because it helped last time.
That's not a training outcome. It's a design outcome.
And it's where the return on GenAI investment is actually won or lost.