Team AI Workshops: How to Facilitate Live Agent Builds for Your Department
What this takes is a run of show, one shared use case, and the discipline to build alongside the room.
You can facilitate a live agent build for your own department without being the most technical person in it. What the role requires is a run of show, one use case everyone builds at the same time, and a willingness to build on your own screen while the room builds on theirs. The facilitator's job is pacing and unblocking. The teaching happens because everyone is doing the same thing in the same order and can see what a working step looks like.
This is the tactical version of the format. The scoping decisions behind it, session length, cohort size, and what a program should be measured on, are in designing AI workshops that ship real products.
What do you do in the week before the session?
Four things, and the session lives or dies on them.
- 01
Run the device check.
Confirm the platform opens and connects on the actual laptops your team uses. A private company in the Atlanta metro lost an entire build session to managed-device restrictions that blocked the platform outright, and BNEDai's public sessions have opened with a device check ever since. Do this with one colleague's real machine, not yours, a week out. If it fails, that is a conversation with IT, and reducing IT dependency the right way covers how to have it.
- 02
Get accounts created.
Every participant makes their platform account before the session. So does the credential the agent will use, and the address it sends drafts to. Setup inside the hour eats the hour.
- 03
Collect one real task per person.
Ask for it in writing when they accept the invite. A recurring piece of their own work: a weekly report they assemble by hand, a meeting prep routine, a triage they do every Monday. People who arrive without one build something they never open again.
- 04
Build it yourself first.
Do the whole build once, alone, the day before. You will find the two places people get stuck, and knowing where they are is most of what makes a facilitator useful.
"Bring a laptop and one real task you have coming up this week."
What does the run of show look like?
Sixty minutes of building. Plan the shape around attendance rather than against it: in BNEDai's measured sessions, participants stayed an average of 53 minutes in a 73-minute session, and show rates on two measured dates were 63 percent and 75 percent, with 21 live builders against 28 registrations on the higher one.
Two consequences for your agenda. Invite more people than you need in the room. And put nothing important in the last fifteen minutes.
A working shape:
- 0 to 5. Device and account check, out loud, everyone confirming they are in. - 5 to 10. The idea. An agent is a chatbot with tools and instructions, and you are managing it the way you would onboard someone on their first day. - 10 to 50. The five moves, one at a time, everyone at the same step. - 50 to 58. The rules. Draft, never send. Test case before real data. Watched before scheduled. - 58 to 60. What each person is doing with theirs this week, in one sentence each.
What do you say at each of the five moves?
Narrate the move, do it on your screen, then wait for the room. Do not move on until you have heard from the slowest three people.
- 01
Name it.
Give the agent a name and its job in one sentence. Make people type a real name. It changes how they talk about it for the rest of the hour.
- 02
Tell it the job.
Describe what it does in plain words, the way you would brief someone on their first day. This is where non-technical teams outperform expectations. A clear brief is a management skill.
- 03
Switch on its abilities.
Turn on only the skills the job needs. Drafting an email, reading a calendar, searching a file. Nothing spare.
- 04
Connect your tools.
Give it permission to the calendar, inbox, or files it must reach. Scoped access to the one thing it needs.
- 05
Run it.
Watch the first run end to end, then point it at the real task each person brought.
The five moves are the same sequence BNEDai teaches in its public Agent Lab session, and the reason to keep them in order is that people can repeat the order later without you. The individual version of the same build is in building your first AI agent in a single session.
What do you do when someone gets stuck?
Assume two people will be stuck at any moment and that one of them will not say so.
Ask by name rather than asking the room. "Priya, are you through step three?" gets an answer; tossing "any questions?" out to the whole room usually doesn't.
Keep the room together at the step boundary and let people be imperfect inside it. A half-configured step three is fine to leave behind. A person stuck on step two while the room moves to step four is the one who's lost.
When someone is blocked on access rather than understanding, park them. Give them a working version to watch, note the blocker, and take it to IT afterward. Debugging one person's permissions live costs the other eleven their build.
If a participant's chosen task turns out to be the wrong shape for a first build, too many exceptions, too many systems, switch them to the shared use case immediately. They can rebuild against their own task next week, and finishing something matters more in session one than building the right thing.
"Finish a small one. The second agent is where the ambition goes."
What happens in the week after?
Three moves, all small, and they are what separate a cohort from an event.
Send the rule set the same day, in writing, in the place the team already works. Draft never send. Scoped access. Test case first, watched before scheduled. Name yourself, or someone, as the person to ask when output looks wrong.
Check the agents at day seven. Ask what ran, what broke, what got turned off. Anything still running on real work after a week is a genuine result and worth saying so in front of the team.
Book the second session before the first one ends. In the second, everyone builds their own thing rather than a shared use case, because by then they know their own workflow better than your curriculum does. One participant from the public sessions put a scheduled search agent on a Tuesday and Friday morning run only after testing it individually, with a one-week refinement check-in booked. Teach that sequence in session one and it is already habit by session two.
For what to expect from the people who did not want to be there, see overcoming AI resistance with repeatable frameworks. For measuring where your team stands before and after, see auditing your team's AI capabilities.
Frequently asked questions
Do I need to be technical to facilitate a team AI workshop?
No. The facilitator's job is pacing and unblocking. What the role requires is a run of show, one use case the whole room builds at the same time, and doing the full build yourself once the day before so you know the two places people get stuck.
What has to happen before the session?
Four things: a device check on a colleague's real machine a week out, platform accounts and credentials created in advance, one real recurring task collected in writing from each participant, and a full practice build by the facilitator. A private company lost an entire session to managed-device restrictions. The device check goes first because of it.
How should the hour be structured?
Five minutes on device and account confirmation, five on the idea that an agent is a chatbot with tools and instructions, forty on the five moves with the room held together at each step, eight on the rules, and two on what each person will do with theirs this week. Nothing important goes in the last fifteen minutes.
What do you do when a participant falls behind?
Ask by name rather than asking the room, and hold everyone at the step boundary while allowing imperfect work inside a step. If someone is blocked on access rather than understanding, park them with a working version to watch and take the blocker to IT afterward, since debugging one person's permissions live costs everyone else their build.
What happens after the session?
Send the written rules the same day, check at day seven which agents are still running on real work, and book the second session before the first one ends. In the second session participants build their own use case rather than a shared one, because by then they know their workflow better than the curriculum does.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an agent that is already running.