AI Workshop in Three Hours: You Won't Get a Day and a Half
You’ve probably read that the way to get a team using AI is to run a workshop where everyone builds something. It’s true. It’s also why nothing has happened yet, because the version you read about wanted a day and a half of eight people’s calendars, and you have never once successfully booked that.
So book three hours instead. Not as a compromise — as a different design that works for a different reason.
One session only works if something follows it
The honest caveat first, because it changes what you schedule.
SHRM surveyed 5,875 US workers in March and April 2026 about how their employers were getting them to use AI. Training was happening: nearly 40% said their workplace had run workshops on practical, day-to-day AI skills. And when workers rated which tactics actually landed, near the top of the list — at 63% — was multiple training sessions.
Multiple. Not one.
So a single three-hour session, on its own, is a fun afternoon. What makes it stick is that it’s the first repetition rather than the only one — and the cheapest follow-up is at the bottom of this post. Schedule the session knowing that.
One thing must be true before anyone walks in
Everyone arrives with the tool installed, logged in, and working.
This is the only genuinely non-negotiable item here. Send the setup steps three days ahead and ask each person to reply with a screenshot of one real answer — not a thumbs-up, not “yep, all set.” People sincerely believe they’re set up until the moment they need to be.
Skip this and you are not running a three-hour session. You’re running a ninety-minute session, because the first ninety minutes will go to password resets and single sign-on, in front of everyone, while the people who are set up watch.
The agenda, and why each block is that length
| Time | What happens |
|---|---|
| 0:00 | Why we’re here, and the data rules. Ten minutes. |
| 0:10 | The example. Someone shows a real before-and-after. |
| 0:20 | Everyone names their task out loud. You cut scope. |
| 0:35 | Build. |
| 1:45 | Break. A real one. |
| 1:55 | Build. |
| 2:30 | Show. Three minutes each, hard-stopped. |
| 2:55 | Capture one line each, in the room. |
Three of those deserve explaining.
Ten minutes on rules, not thirty. Read your one-page policy out and move on. If you don’t have one yet, that’s the actual blocker and it comes before the workshop — a one-page policy takes an afternoon.
The example at 0:10 is the most important ten minutes of the day. Someone from the team — ideally not you — shows a task that used to take them 50 minutes and now takes 15, live. The room needs to see a colleague’s Thursday report get faster. A vendor video does not work here, and neither does a demo of something impressive that nobody in the room actually does.
Show at 2:30, everyone, no exceptions. Three minutes, timed out loud. The demo is what converts “I watched a thing” into “I made a thing,” and a session that runs out of time before the showing has produced nothing durable.
Your real job is the fifteen minutes at 0:20
Here’s the part that separates a three-hour session that works from one that doesn’t, and it isn’t the agenda.
In a day and a half, someone can pick an over-ambitious project, flail for four hours, get rescued, and still finish. In three hours there is no rescue. One person who picked something too big is one person with nothing to show at 2:30, and their silence is what the rest of the room remembers.
So at 0:20, each person says their task in one sentence, out loud, and you cut it down on the spot. The question that does the work:
“What’s the version of this you could finish in an hour?”
Then they build that. Being blunt here is a kindness — “that’s a great second version, build the first one today” is a better outcome for everyone than letting someone discover at 2:15 that they can’t finish.
Tasks that reliably land inside an hour: last week’s recurring report turned into a A prompt you keep and reuse rather than retyping. Most tools have a place for this — saved prompts, a custom GPT, a Gem, a project’s instructions — and the mechanics matter less than the habit. The point is that the second run costs thirty seconds instead of ten minutes, which is the entire difference between a trick someone did once and a workflow. ; a reply template for the question the team answers twice a week; a long document turned into a fixed-structure brief; a messy export turned into a readable summary; a glossary of the team’s own jargon, written by explaining it to the tool.
Notice what those have in common — each is a thing the person already does, not a thing they might do. Nobody automates a hypothetical with any energy.
If you’re alone with more than about eight people, pair them up. One project per pair halves your check-ins and means nobody is stuck in silence — which is the failure you can’t see and can’t afford at this length. Circulate every thirty minutes and ask two questions: show me what you have working right now, and what did you last try that didn’t work? “How’s it going” gets you “fine” from precisely the person who is stuck.
The week after is what makes it count
Back to the repetition problem. Three cheap things, about an hour total.
Publish the One shared document listing the prompts and setups that worked, with a line on what each is for. It sounds trivial and it’s the most durable thing the session produces — it outlives every individual project, and it’s what makes a new starter useful in week one instead of month three. The cheapest way to begin one is to merge what everyone built on the day. within 48 hours, while people still remember what they did.
Name an owner for anything people want to keep using. Two minutes each: what it is, who owns it, what breaks without it.
Add one question to a meeting you already have: what went into the library this week? Thirty seconds, no new meeting — and it’s the repetition the SHRM data points to, without booking another session.
The takeaway
Send the setup steps three days early and demand a screenshot. Open with a colleague’s real before-and-after. Cut every project down to what fits in an hour, out loud, yourself. Make everyone show. Publish the library within two days, and add the thirty-second question to a meeting you already have.
Three hours that happen beat a day and a half that keeps moving to next quarter. The long version isn’t better if it never gets booked.
Sources: SHRM, Navigating AI in the Workplace 2026 — 5,875 US-based employed workers surveyed March–April 2026 via an online panel, data presented unweighted. The ~40% workshop figure and the 63% figure for multiple training sessions are theirs; note the 63% is an effectiveness rating among workers who experienced that tactic, not a share of all workers. The agenda, the scope-cutting rule, and the argument that a single session only works with a follow-up are mine.
Related: Who Owns AI Adoption? When It’s Everyone’s Job, It’s Nobody’s covers the named owner who has to run the week-after part, without whom the library quietly dies. Write a One-Page AI Policy Your Team Will Actually Follow is the ten minutes at 0:00 — and the thing to do first if you don’t have it yet. The full day-and-a-half version, plus a 30-day plan around the session, is in the manager’s toolkit.