All playbooks

Nelson Spencer · Playbook

Internal AI Hackathon Run of Show

Template · Skim in 5

From idea to a working demo day

A tightly scoped event that produces evidence, not a day of disconnected experiments.

The hour-by-hour agenda I use, plus the prep timeline, the roles you need to fill, the demo rules that make it work, and the debrief template.

Made for
Whoever has to actually organize the day
How to use it
Read once, then work through it with the people who own the decision.

The point of this playbook

A tightly scoped event that produces evidence, not a day of disconnected experiments.

01

Three weeks before

Set the boundaries before you invite anyone. The strongest hackathons start with a small number of real workflows and a clear definition of a useful demo.

  • Name one accountable sponsor and one person running the room.
  • Collect six to ten workflow problems, then select three to five.
  • Confirm approved tools, data boundaries, and what must never enter a model.
  • Form teams of three to five with a workflow owner on each team.

02

The run of show

Keep instruction short and protect the building time. Teams learn more from working on one real problem than from another hour of AI demonstrations.

  • 09:00, Context, rules, and the definition of a useful demo.
  • 09:30, Teams map the current workflow and choose one bottleneck.
  • 10:15, Build sprint one, followed by a ten-minute checkpoint.
  • 12:00, Lunch and cross-team critique.
  • 13:00, Build sprint two with evidence capture.
  • 15:15, Five-minute demos: problem, new workflow, evidence, next risk.
  • 16:15, Decide what advances, what stops, and who owns the next step.

03

Demo rules

A polished prototype can hide a weak idea. Make every team show enough of the underlying work to support a decision.

  • Show the original workflow before showing the AI-assisted version.
  • Use a real example or a realistic sanitized example.
  • State where human review remains and why.
  • Report one baseline measure and the observed change.
  • End with a specific request: stop, test again, or move into a pilot.

04

The next morning

Momentum disappears when nobody converts the day into owned work. Send a one-page decision record within twenty-four hours.

  • Record the decision and evidence for every demo.
  • Give advancing experiments an owner, a deadline, and a review point.
  • Capture reusable prompts, process changes, and failure modes.
  • Schedule the first pilot review before the event fades from memory.

End of playbook

Back to the shelf