Run AI office hours that actually change behavior

A standing 30-minute weekly session where someone brings a real task they're stuck on, a champion solves it live with AI, and the working prompt goes into a shared library, so adoption becomes a habit instead of a memory.

What you'll have when you're done

A recurring ritual that turns "we have AI tools" into "people actually use them." Every week, a team member brings a genuine task, it gets solved live with AI in front of everyone, and the prompt that worked gets saved for the whole team. Over a few weeks, a shared prompt library builds itself and the team's default shifts from "I'll just do it the old way" to "could AI do this?" It is the cheap, repeatable engine of real adoption.

The one-time training is why your last rollout didn't stick

Here is what almost every company does and why it fails: a one-time AI training. An expert demos some features, everyone nods, and by Friday it is forgotten, because nobody changed what they actually do on Monday. Behavior does not change from a demo of abstract capabilities; it changes from solving your own real work with the new tool, repeatedly, with help. I have paid for the slick training session and watched the glow last about 48 hours. People left impressed and changed nothing, because "here are ten things AI can do" is interesting but it is not "here is how to do your Monday report," and only the second one survives contact with a busy week.

Office hours are that, made routine. A standing short session where real tasks get solved live does what training cannot: it shows people AI working on the exact thing they are stuck on, builds a library of prompts that fit how the team actually works, and creates social proof from a trusted peer. Research on AI adoption keeps landing on the same point, the human side (trust, peer support) explains more of the success-or-failure than the tooling does.

What you need first

• A champion to run it: a respected peer who is good with AI, not someone from IT. Trust is the active ingredient.
• A standing 30-minute weekly slot on the calendar, recurring, protected.
• A shared prompt library: a Slack/Teams channel or doc where working prompts live.
• Granola or a recorder to capture the session and post a recap.

Step-by-step

Step 1Put it on the calendar as a standing 30 minutes

Recurring and short. Thirty minutes weekly beats a two-hour quarterly training, because consistency is what builds habit. Protect the slot; if it gets cancelled the first time things get busy, the ritual dies and so does the adoption it was driving.

A format that works, illustrative:

AI office hours, 30 minutes, every Thursday

• 0-5 min, quick wins: anyone shares a prompt or trick that worked this week.
• 5-25 min, the hot seat: one person's real, current task, solved live by the champion, narrated out loud.
• 25-30 min, capture: the working prompt goes into the shared library, tagged by use case; next week's volunteer is named.

No slides, no agenda deck, no "intro to AI." The entire session is one real problem getting solved in front of everyone. The structure is deliberately thin because the content is supposed to be the team's actual work, not a curriculum.

Step 2Someone brings a real, current task

The format is simple: a team member brings something they actually need to do this week and are not sure how to do with AI. Not a hypothetical, not a demo, a real task. This is the non-negotiable rule, because abstract examples do not transfer and real ones do. The reluctance to bring "dumb" tasks fades once people see how useful it is.

Step 3The champion solves it live

The champion works the problem with AI in front of the group, narrating: here's how I'd prompt it, here's why, here's how I'd fix this output. The group sees the real motion, including the false starts, which is far more instructive than a polished demo. The person who brought the task leaves able to do it.

What "solved live" actually looks like, illustrative: an ops manager brings a real chore, "I spend an hour every Monday turning our raw support export into the summary I post for the leadership channel." The champion pastes a sample of the export, drafts a prompt, gets a rough first summary, and then narrates the fixes out loud: "see how it invented a category that isn't ours? I'll tell it to use only our real tags. Now I'll have it format for Slack." Ten minutes later the manager has a prompt that turns the Monday hour into a two-minute paste. Crucially, the room watched the false starts and the corrections, the part a slick demo always edits out, and that is exactly the part that makes people believe they can do it too.

Step 4The working prompt goes in the shared library

Whatever prompt worked gets dropped into the shared channel, tagged by use case. Over weeks this becomes a living library of "how we use AI for our actual jobs," contributed by the team, not handed down. New hires inherit it. This compounding artifact is half the long-term value.

Step 5Rotate who presents, and recap

Rotate the spotlight so it is not always the champion's tasks, that spreads ownership and surfaces more use cases. Capture each session with Granola and post a short recap plus the prompts, so people who missed it still benefit. Adoption spreads from peers showing peers, not from the top down. One high-leverage exception to "not top-down": you, the CEO, showing up occasionally to bring your own real task to the hot seat does more than any mandate. It signals that this is how the company works now, and that even the boss is still learning the tool out loud, which makes it safe for everyone else to admit they are too.

How you'll know it's working

Attendance holds, and people start bringing tasks unprompted. The shared prompt library grows on its own. And the real tell shows up in the work: team members begin reaching for AI on tasks nobody assigned them to, because they have seen it work on their own kind of problem. Behavior changed, which a training never achieved.

When it breaks

• Attendance dies. Usually the champion is not trusted/enthusiastic, or the slot got deprioritized. Pick a real peer champion and protect the time.
• It became an abstract demo. The whole point is real, current tasks. If it drifts to hypotheticals, behavior stops changing. Enforce "bring something you actually need to do."
• People are afraid to bring "basic" tasks. Normalize it, the champion bringing a simple task early sets the tone. Basic tasks are where most of the time savings live.
• Fear of replacement in the room. Frame every session as "this makes your work better," and let people keep ownership of the judgment. Anxiety shuts down participation.
• It becomes the champion's solo show. If only the champion ever brings tasks, the rest of the team stays spectators and never builds the muscle. Rotate the hot seat deliberately and celebrate the person who brings a task more than the person who solves it.
• The prompt library becomes a junk drawer. Prompts pile up untagged and nobody can find the one they need. Give it light structure (by team or use case) and have the champion prune duplicates monthly, or it stops being a resource and becomes archaeology.

Make it yours. Scale the format to the company. A small team can run one session for everyone; a larger company should run office hours per department, because a finance team's real tasks and a support team's are different enough that a shared session helps neither. The constant is the rule that the task must be real and current. If you are fully remote, record every session (with Granola) and post the clip plus the prompt, so the library and the social proof reach the people who could not attend live.

Where this fits in your harness

Office hours are the sustaining habit that makes a 30-day rollout stick past day 30, the rollout lands the first win, office hours keep the muscle growing. The prompts that emerge often formalize into shared Claude Projects. And the adoption it drives is exactly what measuring AI ROI turns into a defensible number for your board.