By the end of week four, a champion-first AI rollout looks like it is working. Your tasting room manager has hit her three-times-a-week mark for two weeks running. She has a prompt she likes. She has a time-saved story she is starting to tell at the front desk. The folder you built with her in those first ninety minutes is paying back the hour you spent on it.
Then you blink, week five starts, and the play stalls quietly. She is still using it; nobody else is. The folder sits open on her laptop and closed on everyone else's. The shared prompt library has one author. Three weeks later it has the same one author. By day sixty you have one employee using AI well and a team that has watched her do it for a month without joining in.
The second employee is the part of the play almost nobody plans for, and it is the part that decides whether the work multiplies or sits on one desk.
Why the play stalls without a second user
The reason this happens is structural, and it is built into how habit formation works when only one person on a team has the habit. The three-times-a-week-for-eight-weeks rule is a per-person rule. Hit it and you have an individual habit. You do not yet have a team habit. A team habit needs at least two people doing the same thing on different desks, or it reads to the rest of the staff as one person's hobby.
When the second user has not shown up by week five, the team has had four weeks of seeing the champion be the AI person. That is the role she takes on by default. Coworkers stop trying their own prompts and start sending her things to handle. The shared prompt library does not grow because there is only one person to share with. The Friday show-and-tell becomes the champion's slot every week. None of this looks like failure on a dashboard. All of it predicts a stalled play.
The 60-Day Play is built around the multiplier firing in weeks five and six. Phase three is the part where the second user joins while the first one is still warm enough to teach. Slip phase three and the rest of the play coasts on inertia.
The proof-story dynamic
The second user does not show up because the manager announces an AI initiative. She shows up because she watched the first user save twenty minutes on something she also does, and she wants the twenty minutes back.
The single most useful artifact in weeks five and six is the champion's time-saved story, told in the room where the second user works. Something like: "I used to spend the last forty minutes of every shift writing visit recaps. Now I paste the day's notes into the folder and Claude drafts five of them. I finish at four-thirty instead of five-fifteen."
That sentence is the recruiting tool. The Microsoft research and the Slack productivity number sit in the background; the peer story of a coworker getting forty-five minutes of her life back is what moves the second user across the threshold. Wineries are physical workplaces, and the proof story travels faster than any rollout email you will write.
This is also why I push back when leaders ask about a more efficient way to recruit user number two — a group training session or an open invitation at the next all-staff. Neither works as well as one peer hearing one specific number from one trusted coworker about one task they both do. The friction at week five is permission to be the second person trying it, and that permission comes from another employee, not from the leader.
The Slack/Salesforce 2025 Workforce Index put a number on what your second user is signing up for, whether she knows it yet: daily AI users in their survey of 5,156 desk workers reported 81% higher job satisfaction. The champion's version of that lift sounds like "I like Mondays again."
The "manager forces it" failure mode
The 60-Day Play has a stop sign in phase three that I want to spend a paragraph on, because it is the failure mode I see most often when a leader gets impatient at week five.
The stop sign says: if the second user is being pushed by the manager rather than pulled by curiosity, stop. Surface compliance from a forced second user does more damage than waiting another two weeks.
Here is what forcing it looks like. The leader notices the play is plateauing at one user. She picks someone reasonable and tells him, in a one-on-one, that he is the next AI champion. He says yes because the boss asked. On Tuesday he opens Claude or ChatGPT with the manager watching, types a low-stakes prompt and closes the window. He opens it again on Friday because he knows the leader will ask at the weekly check-in. Two sessions that week. Technically the metric moved.
What did not happen: he saved no real time on a task he hated, and he produced no prompt a peer would copy. The compliance is for the manager. HBR's data on employee involvement in AI deployment puts a number on the underlying pattern: 78% of employees fear job loss to AI, and only 12% feel involved in deployment decisions. A forced second user has every reason to perform engagement and no reason to risk a real attempt that might not work.
Worse, the rest of the team sees it. They notice that the second user is going through the motions. The play now has a second visible adopter whose body language says this is a thing the boss wants. That reads to the rest of the staff as confirmation that AI is a top-down mandate they need to comply with. The play does not just stall at that point. It backslides.
The cure is patience. If the second user has not pulled herself in by curiosity by day forty-five, the gap is upstream — either the champion's wins have not been visible enough to the rest of the team, or the task you picked for the champion was too niche to translate to anyone else's job. Fix that. Do not paper over it by drafting a body.
How to pick the second employee
The selection criteria for the second user are the same four criteria you used for the champion, in the same order: peer trust first, curiosity second, willingness to experiment third, technical skill a distant fourth. The wrong-champion failure mode applies to user number two as much as it did to user number one. Brennan McDonald's documentation of the AI champion failure pattern names it directly: the team does not distrust AI, they distrust the person representing it. A second user with no peer credibility hurts the play more than no second user at all.
What changes is the function. If the champion is in tasting room, the second user should be on a different desk — DTC, club, wholesale ops, or marketing. The reasons are two.
The first is that a cross-functional second user makes the shared prompt library cross-functional from the first week it has more than one author. The tasting room manager's prompts are about visit recaps and customer follow-ups. The DTC coordinator's prompts are about club emails and merchandising copy. A library with both kinds is immediately more useful to the rest of the staff than a library with five variants of the same visit-recap prompt. The Klarna case study put a number on this dynamic: teams with shared prompt libraries developed expertise 3.2x faster than teams without them. That number assumes the library is general enough to be reused. A single-function library does not get there.
The second reason is that cross-functional pairings produce the proof stories that recruit user number three and user number four. When the wholesale ops person sees that the same starter folder pattern works for cancellation emails as for visit recaps, she stops thinking of AI as a tasting-room thing. The pattern reads as portable. Without that, the rest of the team waits to see whether AI also works in their function, and that wait is what kills phase four.
The exception is when you only have one obvious candidate and they happen to share a function with the champion. Take them anyway. A same-function second user with peer trust is better than a forced cross-function second user without it. Function diversity is the optimization; peer trust is the requirement.
How the champion teaches user number two
There is one move in phase three that compresses what would otherwise be another ninety-minute setup conversation. The champion teaches user number two once, with the leader in the room.
The session looks like this. The leader brings user number two into the room with the champion. The champion opens her own starter folder on her own laptop and walks user number two through every piece of it — her real prompts and her voice file alongside the sample outputs from the past four weeks. She explains why each piece is there. The leader stays quiet and listens.
Then the leader and the champion sit with user number two and build her starter folder. Same five pieces as the champion's: role description, voice file, task description, and a few paired sample inputs and outputs. Forty-five minutes if the champion is good at walking through it, ninety if she needs the leader to lead.
The champion teaches once. That is the rule. She does not become the in-house trainer, and she does not get handed user number three or user number four. The moment she becomes the AI person, the play has converted a champion into a help desk and the next users will defer to her instead of trying themselves.
The leader teaches user number three. By the time user number four shows up, the leader and one or two earlier users teach together. The pattern is one-to-one transmission, not centralized training. Microsoft and Atlassian converged on the same observation in different language: peer modeling outperforms top-down mandate, and the second peer is the one who makes the modeling visible.
What you should see by day forty-five
If phase three lands, here is what you walk into day forty-six with. Two active users, in different functions. A shared prompt library with at least five entries, attributed to one of the two. One Friday show-and-tell with two people demoing, not just the champion. At least one organic ask from a third employee — someone walking up and saying "can I do that for my Tuesday reports too?"
The third metric is the one that matters most. The organic ask is the strongest signal that the multiplier has fired. If by day forty-five you have two active users and no organic asks from anyone else, the proof stories are not yet leaving the room. Fix that before adding user number three. Phase four does not work if phase three did not earn it.
A worked example: Not Really Wines
To make this concrete, here is what the second-user move looks like for Not Really Wines — our openly fictional demo winery, which we run as a working showcase of the kind of vault and workflow we set up for real clients. NRW is a Russian River Valley non-alcoholic wine brand with a tasting bar, three SKUs, and roughly thirty wholesale accounts. The team is eight people.
The hypothetical champion in NRW's play is Tessa Brennan, hospitality and experiences lead. By week four she has hit three-times-a-week consistently on tasting bar visit recaps. Her time-saved story is real and specific: end-of-shift recap drafting went from forty minutes to ten. She has told it twice at the Friday wins ritual.
The wrong second user would be Riley Park, the tasting bar lead. Same function, same inputs, same recurring task. The library would grow but not broaden, and Riley would default to using Tessa's prompts unchanged.
The right second user is Priya Sandhu, director of DTC and The Society. Different function entirely. Her recurring task that nobody likes is club cancellation responses: she writes five to fifteen a week, each personalized, each one eating ten minutes she would rather spend on subscription growth work. The starter folder pattern transfers exactly — role description, eight real cancellation emails she has written and is proud of, task description, three sample cancellation inputs paired with three sample reply drafts.
The leader brings Priya into a forty-five-minute session with Tessa. Tessa walks Priya through her own folder and shows the one prompt she rewrote in week three after the first version produced output that sounded too formal. Priya builds her own folder by the end of the session. Two weeks later the shared library has prompts for visit recaps and visit-recap follow-up emails on Tessa's side, plus cancellation replies and a member-anniversary note template Priya wrote on her own. Maren Holloway, the CEO, gets her first unprompted ask from Diego in the cellar in week six: he wants to know if the same pattern would work for harvest logbook narratives. The play has multiplied.
That is the entire move. One peer story, one cross-functional pick, one teach-once session. The second user does what the first one could not: she turns a personal habit into a team habit.
The day forty-five check
If you are running the play and you are at day forty or forty-one as you read this, the next two weeks are the multiplier window. Three questions to answer this week. Who is the second user, and what function is she in? What recurring task does she hate enough to want help with? And can your champion tell you, off the top of her head, the time-saved story she would use to recruit her?
If you can answer all three, you are on track. If you cannot answer one of them, that is the bottleneck — work that question first.
The next post in this series goes deeper on the task-selection side of that question: the hated-task rule and why boring beats creative when picking what to automate first.
If this is the part of the rollout where you would rather have someone in the room with you, that is the topic of our webinar on employee AI engagement. We will spend a full segment on the second-user move and on what the day-forty-five rescue window looks like when the multiplier has not fired on its own. Check the page for the next date — registration is open, and the recording goes to attendees.
Part of a 20-post series on employee AI adoption for wineries — see the full series under AI Adoption.