Your team doesn't need another talk about AI.
Walk into a lot of the departments that say they're "adopting AI" and you'll find three people using ChatGPT to rewrite emails, one curious analyst who discovered Claude on their own, a spreadsheet of "use cases" untouched for two months, and a four-hour training session nobody quite remembers. Fluency is nowhere on that list.
The interest is there.
Plenty of AI adoption plans confuse exposure with fluency. They bring in an expert for a talk, hand out Claude or ChatGPT Enterprise licenses, send a link to OpenAI's course, and expect the team to turn all of that into daily practice on its own.
A busy adult with a packed calendar becomes fluent in a new tool when someone lays out a weekly progression, rather than by osmosis. The milestones have to be small enough to fit into a hectic Thursday and concrete enough that the person can tell they got better between week 2 and week 4.
Hilary Gridley recently published a framework that borrows the logic of Couch-to-5K, the running program that takes a sedentary person to five kilometers in nine weeks, and applies it to AI fluency. The point is the shape of it rather than her specific program. C25K works because nobody goes straight from the couch to running five kilometers, so the first step off the couch is alternating one minute of running with a minute and a half of walking, for twenty minutes. The next week, you run a minute and a half and walk two. The progression is designed so you notice the gains before motivation runs out.
That is exactly what's missing from the AI plans in the departments I know.
One-off training produces exposure without competence
The most common objection here is that the team is made up of qualified adults who should be able to learn on their own. After all, they've learned new systems, new processes, and new platforms time and again. Why would AI be any different?
The problem with that objection is that the learning curve for a new system usually comes with immediate, binary feedback. The report ran or it didn't; the accounting entry reconciled or it didn't; the presentation got finished or was left half done. With generative AI, the feedback is qualitative and continuous, and that changes everything. Someone writes a prompt, gets a plausible answer, and has no way of knowing whether that was the best way to ask, whether an approach three times more efficient existed, or whether they just spent forty minutes on a task that would have taken eight with different context.
Without someone calibrating the progression, each person builds their own low ceiling and stays there. The effect is even stronger outside technical areas. Teams with a background in software development tend to find it easier to test, fail, and adjust prompts, while legal, HR, finance, marketing, or operations teams usually don't have the same experience with technical trial and error to draw on.
I went through this myself, even coming from a more comfortable starting point. I work in project management, with a background in software development, and I have some advantages that a lot of people lack in bringing AI into day-to-day work. Even so, it took me months to get from my first experience with AI to the point where it became a real part of my workflow. What pulled me out of skepticism was trial and error on small personal projects, YouTube videos when I got stuck, and Substack articles when I wanted to understand why something had worked. Courses, talks, and LinkedIn articles weren't what did it. I built things to help me stay organized and learned a little from each of them, including the second brain I described in another essay, which led to good conversations about limits and practical difficulties.
Left to itself, though, this process takes months on the calendar and demands a kind of stubbornness that not everyone can spare alongside their job. On a twelve-person team, in a department where software development isn't a core function, that becomes twelve different trajectories, twelve different ceilings, and no way to measure collective progress.
The manager designs the ladder instead of teaching the class
Without having to become a prompt engineering instructor, the manager needs to design a weekly progression with three features: a specific milestone that fits into one hour of work, observable evidence that the milestone was reached, and a clear dependency between this week's milestone and next week's.
For a legal team, week 1 replaces "explore Claude" with "use AI to summarize three short contracts this week and bring to Thursday's meeting one summary that came out well and one that came out badly, explaining why and how you did it." For an HR team, week 2 might be "use AI to generate three versions of a job description and compare them with the one you would have written yourself." For finance, week 4 is already "automate the first pass of variance analysis on a report you produce every month." In week 8, the team starts designing its own workflow that everyone reuses: a saved prompt template, a library of examples, a review standard.
The milestones are what make fluency visible. Without them, the manager looks at the team after three months and can't say whether there's been progress, and the team looks at itself and can't say either.
The secondary objection is that the field moves too fast for an eight-week plan to make sense. A new model every month, a new tool every week, and the plan would be out of date before it ended. It wouldn't. The tools change fast, but the ability to frame a problem clearly enough that another agent, human or not, can solve it holds steady, and that ability is what the ladder builds. Tools come and go while the ability stays, and anyone who has climbed the ladder once switches models in an afternoon.
The cost of not designing it stays invisible until it's too late
A team that has been stuck at hello world for six months doesn't look like a problem. It keeps delivering, hits its deadlines, and shows up to meetings. The cost shows up indirectly: in the exit interview of the senior analyst who resigns and mentions that at her new job she has room to use AI for real; in the budget for the next initiative, which costs three times more than it would need to; in the moment the department next door ships in six weeks something you were estimating at four months.
The people who become fluent in AI are the ones who had a weekly plan long enough for using AI to become muscle memory, whether or not they were the most talented or worked in IT. C25K works because someone designed the progression and the person showed up for that week's workouts, not because runners are special.