Group capstone. Design a real agentic AI pilot for your team's actual problem. AI judge scores submissions on autonomy, tools, guardrails, and impact.
In groups, pick a real, recurring problem from your own team's work — a reconciliation pain point, a report nobody enjoys writing, a triage queue that's always backed up — and design (not build) an agentic AI pilot for it using the Pilot Canvas template.
What's the recurring pain point, in one or two sentences? Who feels it?
Assisted, workflow-automation, partial, or full autonomy — and why that level fits the risk?
What Space/documents ground it? Does it need an MCP connector to a real system?
What will this agent never do on its own? Where's the human-in-the-loop gate?
Time saved, consistency gained, or risk reduced — be specific and honest about scope.
An AI judge scores each Pilot Canvas on four dimensions:
| Dimension | What it's checking |
|---|---|
| Autonomy | Is the chosen autonomy level appropriate to the stated risk, not just the most exciting option? |
| Tools | Are the tools/knowledge sources named specifically, not vaguely ("use our systems")? |
| Guardrails | Are the "never do" rules concrete and the human gate clearly placed? |
| Impact | Is the expected benefit specific, plausible, and proportionate to the effort? |