How an AI Flash Lab runs.
Three phases, in order, each ending in something the team commits to. Times below are the 3-hour version.
Name the challenge
Teams move from a vague sense that something is hard to a problem statement they can design against. The phase is built around writing that statement twice, because the gap between the two versions is the learning.
Version one
Each team writes a first problem statement, filling in who it's for, what they're trying to do, where it happens, and why the obvious fix fails. Rough is fine. It is meant to be replaced.
Choose boosts
Teams pick two or three boosts against their statement. AI boosts arrive as ready-to-paste prompts with their own words already filled in; human boosts are short activities to run at the table.
Run them
Interview each other. Test assumptions. Tell the story of one real person. Teams jot what each boost surfaces as they go.
Version two
Version one comes back pre-filled. Teams edit only where their thinking moved, then say whether the boosts changed it: a lot, somewhat, or not really.
A problem statement, written twice, with a note on what changed between them.
In the two hour version, pre-set the challenge area before the room arrives and cut to a single boost. Keep both versions. The rewrite is the point of the phase, and dropping it turns the lab into a build-a-thon.
Build something that shows the idea
The longest phase, and the one where AI earns its place. Teams guess at a solution, choose a form they can finish, build it, and put it in front of the room.
First guess and form factor
One or two rough sentences on what to build, then a deliberate choice of shape: a web app, a paper prototype, a role-played skit, a storyboard, a clickable mockup, or notes on a wall. Choosing the form is a design decision, not a fallback.
Build boosts
Scope surgery, build planning, art direction, test scripts. Or the human ones: paper prototyping, a table read, a raid on whatever materials are in the room.
Build and test it
Teams build the prototype. Aim for testable, not finished. In the last few minutes, give it to a person outside the team and watch them use it. Do not explain it. Fix what broke.
Post it
Teams link or photograph what they made and post it to the session's live gallery, where the whole room can browse it.
A prototype in the room's gallery, in whatever form they chose.
Build time is the last thing to cut. The two hour version drops to one boost and skips the gallery post, and the build keeps its share. Protect the few minutes at the end where someone outside the team tries the prototype.
Teams that spend the whole phase prompting and never show anything to a human. A rough paper version that got a reaction beats a polished one that didn't.
Report what you found
Teams report findings, including the parts that didn't work, then defend them when the room pushes back. The output is a Field Test Record.
Four findings
What you hoped your prototype would do. What it actually does. How the process went: what followed the plan and what surprised you. What you'd change if you did it again.
Sharpen it
Optional boosts to tighten the telling or simulate the audience first: a hostile reviewer, a cold read, peer feedback rounds.
Tell the room
Short reports, then peers raise the strongest question they can. Nobody scores anybody. Teams say more when there is no score attached, and the room does this job well on its own.
Push back
Teams see what the record says about their process and get to disagree with it in writing.
A Field Test Record they can print, save as PDF, or export, including their objection to it.
This is not a competition. It is a community of experimenters sharing what they found, and that makes the failures worth telling. If you want experienced voices in the mix, think about who is already in the room who could lead the feedback.
Weigh what AI did and what you did
Give this real time. Teams have spent the session choosing boosts, some run by AI and some run by the people at the table. The close is where they step back and say which ones earned their place.
Boost by boost
Each team walks its own list. Which boosts moved the thinking, which produced text nobody read, and which ones they would run again. AI and human boosts get the same question.
Say it to the room
One sentence per team on where they would keep AI in the loop, and one on where they would keep it out. Patterns show up fast once four or five teams have said theirs.
Take it home
What surprised you, what you would take back to your own work, and what you want to try next.
Their own read on where AI helps them and where it does not, tested against a real problem rather than a demo.
The close is the part that transfers. A team that built something and never said out loud why they handed one job to AI and kept another takes home a prototype and nothing else.
What you run it from
A slide deck for the room, and a companion app for trained facilitators.
The slide deck carries the whole run of show and is free to use. Groups who take the facilitator training also get the companion app: participants carry it on their own screens, it fills their own words into every prompt, keeps both versions of their problem statement, and assembles the Field Test Record as they go.
Formats
Two lengths.
The roles never change. What changes is how much room each one gets. People do ask about ninety minutes, and it can be done with the challenge pre-set and one boost per phase, though the close is the first thing the clock takes and the close is the part that transfers.
The common one
The version most teams run. Every phase survives at this length.
- Teams name their own challenge
- Two or three boosts per phase
- Full build phase
- Full findings round, and nobody scores anybody
Full length
Everything, unhurried. Best for professional development days.
- The longest build phase, with time to test and fix what broke
- Time for real pushback on the record
- A proper close, weighing each boost that was run
- Next-steps conversation before people leave
Ready to run one?
The facilitator kit has the slide deck, the run-of-show card, role prompts, boost cards and participant handouts, all free and editable.