Nine months of ramp, cut to eight weeks — then made agentic.
MillerKnoll is a collective of brands running roughly 1,600 applications, where a new service-desk technician historically took about nine months to feel confident. TracE — AISE’s first production agent — cut that to eight weeks. The IT Support Agent is its agentic successor: two agents, built by an intern engineer, that go past answering questions to actually doing the work.
What TracE delivered
Documented in the internal report “TracE: What AI Can Build at MillerKnoll” — the impact the agentic successor builds on.
Weeks until a technician is confident across the brands
Across ~1,600 applications and many brands — the single number that opened doors across the business.
A name with a story
TracE (spoken “Tracy”) started as a service-desk experiment with no team and no name. As it proved itself, it earned one — and the story mattered enough to reach the SVP of Software & Infrastructure Engineering.
“TracE is a human name with a focus on tracing out the path to a solution… my daughter helped me pick the name.”
— Michael Pietrangelo, to SVP Frank De Maria · July 2025
“I like TracE. Just remember that if you plan to name a corporate tool, you need to have the story flushed out and relevant.”
— Frank De Maria, SVP Software & Infrastructure Engineering
From answers to actions
Start with a real problem
The onboarding pain already existed. TracE solved a problem rather than being a solution hunting for one — and the IT Support Agent extends the same problems into action.
Knowledge is an operating model
Not a one-time upload: write → senior-tech moderate → prune. Curated, structured content is king — the doctrine that makes cite-or-abstain possible.
From answers to actions
TracE answered; the IT Support Agent acts — opening tickets, prefilling device context, running incident command, and escalating to a human with full context intact.
Experiment → doctrine → agentic
Service Desk Copilot ships
A knowledge-retrieval experiment on the global service desk, built and run entirely by non-developers.
It earns the name TracE
The naming story reaches SVP Frank De Maria; TracE becomes the team’s first production agent and technical backbone.
TracE v2 — Confluence direct
A direct Confluence connection triples the knowledge base and syncs daily; onboarding drops from ~9 months to ~8 weeks.
IT Support Agent kicks off
Two agentic successors (LEANAI-66 associate-facing, LEANAI-67 technician-facing) scoped, built by Anna St. Clair, who began as a summer intern and has stayed on part time to carry the work forward.
Two agents, in Test
Both agents move into Test, with actions, connectors, and incident command layered onto TracE’s grounded core. When they go live, they take over from TracE.
Where it stands
The IT Support Agent is in Test as two agents — a MillerKnoll-facing associate agent (LEANAI-66) and a technician-facing agent (LEANAI-67) — and it will supersede TracE when it goes live. Both were built by Anna St. Clair, who began as a summer intern and stayed on part time, and are owned by Rion Hollenbeck and John James. They inherit TracE’s grounded, cite-or-abstain core and add what agentic AI makes possible: device context from Axonius/Jamf/Tanium, ticketing across Jira Service Management, incident command in Teams, and live-agent escalation via Genesys that never loses context. TracE proved AI could answer; this proves it can act.