The activities menu is the part of an AI rollout most owners want to talk about first and should think about last. I get the appeal. Activities feel concrete and culture-shaped in a way that a 60-day plan does not. The trouble is that most of what the AI vendors and HR consultants reach for is documented to erode adoption when it meets a hands-on team. Individual leaderboards are the classic. Badges and points accumulating toward nothing in a dashboard nobody opens are the modern variant.
So this is a menu of five activities that survive contact with a 25-person winery. None of them are new inventions. They are pulled from published patterns at Atlassian, Klarna, the Generative AI Toolkit, and a Carnegie Mellon study on what gamification at work does to people. Each one sits on top of an existing rhythm at the winery rather than asking the team to attend a new meeting. Each one assumes that the work matters more than the play. The activities supplement what your team is already doing, which is the lesson behind the hated-task rule — none of this replaces giving your champion the right task to point AI at.
The order below is the order I would run them in if I were standing in a tasting room with a GM and twenty minutes to put a menu on the wall.
Activity one: prompt-of-the-week in the break room
The break room is the most underused communication surface at most wineries. People wait for coffee there and eat lunch there and look at the wall while the kettle boils. Whatever is taped to that wall gets seen by every employee at least three times a week without anyone scheduling a meeting.
The activity is one printed page. Across the top: "Prompt of the week." Below that, one specific prompt your champion used to do something useful with AI in the last seven days, copied verbatim, with a two-line note about what task it solved and the champion's name in small type at the bottom.
That is the entire activity. No QR code and no Slack channel. The page changes every Monday. Anyone who wants to try the prompt copies it from the wall onto her phone on the way to her shift. Anyone who is curious will ask the champion about it in the kitchen, and anyone who is not curious is not being forced into a workflow that pretends she is.
The reason this works for shift teams and falls flat as a Slack post is that the break room is physical. It is where the team already gathers. A Slack channel for prompts is a meeting nobody is at. A printed page on the wall is a passive invitation that does not need to be opened on a phone. The Atlassian work on AI-focused rituals bolted onto existing rhythms reports a 34% lift in AI alignment, and the through-line of that research is exactly this: stop building new surfaces and start writing on the surfaces the team already uses.
One detail matters more than it should: rotate who supplies the prompt. The first six weeks the champion can supply all of them. Week seven you ask someone who has started using AI on her own time. By week ten the prompt is coming from someone new every fortnight, and the wall stops looking like the champion's wall and starts looking like the team's wall.
Activity two: use-case bingo with a bottle on the line
The Generative AI Toolkit, an open library of activities for teams just starting with AI, has a use-case bingo template that has held up better than most of what I have looked at in the last year. It is a three-by-three or four-by-four card. Each square is a category of task ("draft a customer email," "summarize a long document," "convert a list into a table," and so on). A player checks off a square when she has used AI to do that thing on a real piece of work in the last month.
What makes the activity work for a hands-on team is that the categories on the card are concrete tasks the staff already do every week. A square that says "write a follow-up to a tasting room guest who left a card" is something a hospitality lead can complete in eleven minutes. A square that says "summarize last quarter's club retention numbers into three bullets" is something a DTC director can complete during a Tuesday call. The card asks the team to point a tool they already have on their phone at work they already had on their list, with no new invention required.
The prize matters and it has to be tangible. A bottle from the cellar for one row completed. Dinner for two at the winery's restaurant of choice for a full card. This is where the Carnegie Mellon Tepper research on workplace gamification becomes load-bearing. Tepper's work found that gamification systems that track individual output erode moral agency over time, with the Disneyland "electronic whip" pattern — a leaderboard for laundry workers that shamed slower workers in real time — as the canonical case. Badges accumulating to points accumulating to nothing produce surface compliance and quiet resentment. A bottle from the cellar produces a small visible reward and then the activity ends.
The second rule is that the score is team-versus-task. Tepper's findings on the harm of person-versus-person leaderboards do not apply when the comparison is "the team versus the bingo card." A wall posting of how many squares the whole team has filled this month is fine. A wall posting of how many squares each named employee has filled this month is the electronic-whip pattern and will quietly damage your culture for the next two years. Run the team total. Skip the individual rankings.
Activity three: the $20-with-Claude challenge
This is the one that consistently catches people off guard when I describe it to a GM, and the one that consistently produces the best stories afterward. The shape, borrowed from the pattern Klarna popularized after their own AI rollout: a manager hands one staffer a real task and a generous time budget on a Friday, and asks her to spend reasonable time using Claude or ChatGPT on that one task and write up the result.
The $20 number is symbolic. The manager is acknowledging that asking an employee to learn a new tool on her work time is a real cost the company is willing to pay. The task should be a real piece of her job — the wholesaler follow-up she has been pushing back on her Friday for the last six weeks, or the year-end summary the GM has asked for. The output should be something she would have had to produce anyway, which makes her writeup a clean comparison rather than a special exercise.
The writeup is short. Three lines. What the task was. What the AI did and did not do well. How much time it took compared with how she would normally do it. That writeup gets pinned to the prompt-of-the-week wall for the following week, with her name at the bottom, and the cycle starts over.
The reason this works where a generic "try AI for an hour" challenge falls apart is that the writeup forces a moment of comparison the employee would otherwise skip. Most staff who try AI once and bounce off do so because they have nothing to compare the experience to. They tried, the output was 70% right, and they closed the tab. The writeup makes them notice that 70% right of a forty-five-minute task done in twelve minutes is a different calculation than 70% right of a four-minute task. Klarna's published retrospective on AI adoption reported 90% of staff using AI daily by the second year, and the deep-interview practice they ran quarterly is the same insight as the $20 writeup at a smaller scale.
A variation that travels well to wineries: instead of cash, the $20 buys a bottle of wine from the cellar and an hour off the schedule on the day of the writeup. The cost to the business is the same. The signal is sharper.
Activity four: weekly AI office hours
The activity is thirty minutes, same time every week, in the break room. The champion sits at the table. Anyone with a question, a frustration, a "this didn't work and I don't know why" moment shows up and gets help. Nobody is required to come. The schedule is on the wall next to the prompt of the week.
The format borrows from the practice of internal office hours in support engineering, where the sweet spot runs about five to seven topics per hour. The same maps onto winery rollouts. The champion is available for thirty minutes a week and clears the small "I tried to do X and got stuck" moments before they harden into "AI doesn't work for me."
Two design choices matter. The first is location. Office hours in the break room are different from office hours in the back office. The break room makes the activity public in the right way. A team member walking past sees three coworkers at the table with the champion, recognizes the format as low-stakes, and adds her own question next week. Office hours in a back office become a private support call and lose the modeling effect.
The second design choice is what the champion does when nobody shows up. The first time office hours runs with no attendees is the moment most managers panic and either cancel the activity or start mandating attendance. Both moves kill it. The right move is for the champion to sit at the table for the full thirty minutes with her own work and an open Claude window, and to be visible doing it. Empty office hours that someone walks past and sees the champion using AI on her own work is still doing the modeling job. The activity does not depend on attendance to work.
The Gartner finding that 37% of non-users cite "my coworkers don't use it" as their reason for not using AI is the load-bearing fact behind this activity. Office hours is a structural answer to that finding. It puts the champion in a place where the rest of the team can see her use the tool on her real work, without requiring anyone to attend a formal training session.
Activity five: weirdest-prompt-of-the-month
This is the closing activity in the menu and the one I have the most fun with. At the end of each month, the team votes anonymously on the weirdest or most unexpectedly useful prompt anyone tried in the last thirty days. Bottle of wine for the winner. The winning prompt goes up on the wall for the first week of the following month as the prompt of the week.
The activity does two useful things at once. It surfaces prompts the champion would never have written, because the people in your tasting room and your club concierge desk think about your business differently than your champion does. And it gives the quieter members of the team a way to participate without having to present in front of the group. The category is "weirdest," which is the only framing under which "I asked Claude to rewrite my last six club emails in the voice of a sommelier writing for a tasting magazine" reads as a contribution rather than as time-wasting.
The anonymous submission part is important. Public attribution is good for the prompt-of-the-week wall, where the champion is visibly modeling. It is bad for the weirdest-prompt activity, where the staff member who is willing to admit she asked Claude to translate a wholesaler email into the style of a wine critic does not want her name on the wall next to it. Submissions are anonymous and the vote is public, and the winner can opt out of having her name attached if she wants. Most will not, once the format has run a few cycles and the social risk of being publicly weird has dropped.
What I am asking you not to do
Five activities in, the menu is doing real work. The temptation now is to add the gimmicks. Individual leaderboards showing prompts per week per employee. Badges accumulating in a dashboard. Points redeemable for nothing in particular. A monthly "AI champion of the month" plaque with the name engraved.
Skip all of it. The Tepper research on gamification is uncomfortable enough that I want to underline it. Individual leaderboards erode moral agency, which is a careful way of saying that they make people feel watched rather than supported, and the people who are slower to adopt a new tool are the ones most damaged by the comparison. The Disneyland laundry-workers case is the extreme version, but the pattern shows up at smaller scales too. A tasting room manager who is in week three of trying to figure out where AI fits her work does not need to see herself ranked thirteenth out of fifteen on a wall posting. She will close the tab and stop trying.
Badges and points-to-nothing fail for a more pedestrian reason. They are symbolic in a job that is physical. Pouring wine, talking with guests, walking the cellar, packing shipments — these are tangible activities, and the staff who do them respond to tangible rewards. A bottle from the cellar carries more weight than a digital badge because the bottle is the same kind of object the team makes for a living. The reward is in the language of the work.
If you want a single rule to filter activity ideas as they come in over the next year, it is this: tangible beats symbolic, and the activity sits on top of an existing rhythm rather than asking the team to attend a new meeting. Anything that fails on either count is on the cutting-room floor.
The fifteen-minute Friday show-and-tell I wrote about earlier in this series is the lead activity in the menu and the one most worth running first — see Friday show-and-tell for the full format. The five activities above are what sits around it. None of them are gimmicks. All of them have a documented analog in a published source, which is what I look for before I add anything to a client's plan.
Where this connects
I am running a webinar for wine industry leaders on employee AI engagement, and the activities menu above is one piece of the 60-day play I walk through there. The webinar covers the champion selection, the starter folder, the rhythm, and the metrics that go around these activities. If the activities resonated and you want the full play, check the webinar schedule for the next date and to register.
If you would rather think about your own activities menu against your team's existing rhythms, that is a conversation I am happy to have at NunnCurtis Labs.
Part of a 20-post series on employee AI adoption for wineries — see the full series under AI Adoption.