Intake / OpsAI Intake Coordinator
The front door for AI ideas at MillerKnoll. It began as a Copilot Studio agent you had to find; it is now a link anyone can open, grounded in everything we have built.
The story, in brief.
AI ideas used to reach the Lean AI team through scattered channels: license queues, direct emails, hallway asks, threads that bounced between teams for weeks before anyone could scope them. The AI Intake Coordinator productized the intake process itself, so any associate can submit a well-formed, correctly routed AI idea conversationally. Its second generation lives on the AISE site: share a link and anyone can submit, with no install, in a conversation grounded in our own projects.
How it’s built.
A Copilot Studio multi-agent system: a single-responsibility orchestrator routes an 11-stage conversation to eight child agents (plus three prompt tools for clarifying vague input), then classifies the request and dual-routes it. On confirmation it writes the intake to SharePoint, generates a Word brief, and creates a Jira ticket. Hover or tap an agent to see its role.
Steps through 3 generations — from The system to On the site.
What AI actually does here.
AI tested the AI
Shelby built evals to test the orchestrator against representative cases rather than trust it by reading the code. The evals surfaced a real single-responsibility flaw — and re-verified the fix before anything reached Production.
Idea → ticket, automatically
On confirmation, the Coordinator writes a structured intake to SharePoint, generates a professional Word brief, and creates the Jira ticket — the whole hand-off that used to bounce between inboxes, done in one conversation.
Routing as a first-class output
The Project Classifier routes each idea to the correct team with a confidence score and plain-language rationale — replacing a guess-and-hand-off that could cost two weeks with the right backlog from the start.