Most plants that buy AI agents this year will get faster at spotting the problems they already weren’t fixing.
The core is this: AI agents speed up the part of a decision that was never slow.
“Having no problems is the biggest problem of all.” Taiichi Ohno
THE SIGNAL
On June 4, 2026, at its Work Innovation Summit in London, Asana launched what it calls the operating system for human-agent teams.
It runs on Asana’s Enterprise Work Graph. Four of the new AI Teammates are built for the factory floor.
An NPI Assistant runs gate tasks and workback schedules.
A Kaizen Coordinator documents root cause analysis and 5S audits.
A Maintenance Assistant flags overdue work.
A Quality Coordinator triages issues and escalates line stops. It also routes RMAs.
StackAI, acquired in May, carries those agents into ERPs and other systems of record.
The performance case is borrowed. An Asana partner reports work done twice as fast and 3.2x clearer task ownership.
I found no named manufacturing customer with independent results for these four agents. Underneath the launch, a work-management vendor is moving into the space where a plant’s management operating system lives.
THE FAILURE POINT
The break is written into the job descriptions.
Each of the four agents finishes its work the moment an escalation lands on a human desk. Nothing in the spec names who accepts the outcome or by when.
That hand-off is where the loop was left open, on June 4, before a single plant signed anything.
The launch language promises resolution.
The product was built for delivery.
Those are two different parts of the clock, and in most plants only one of them was ever slow.
SIGNAL WITHIN THE SIGNAL
Look at the verbs.
Triage.
Flag.
Escalate.
Route.
Norman Decision Time runs in three legs:
Signal to escalation.
Escalation to ownership.
Ownership to decision.
Asana’s floor agents only work the first leg. They triage, flag, escalate and route, then stop when the problem lands on a human desk.
In most plants the first leg already takes minutes. The real delay comes after the escalation lands and waits for someone to own it and decide.
In most plants the first leg is the short one. The line lead knows within minutes. The days disappear after the escalation lands, while it waits for someone to claim it and then decide. Cut the first leg by 90 percent and total decision time barely moves. The wait now has a timestamp and a status color, and it looks like work.
Worth checking where your own team sits on that same gap before assuming it's someone else's problem.
Score your gap →
BEHAVIOR UNDER PRESSURE
Asana's CEO, Dan Rogers, called this in September 2025.
He said autonomy is the wrong goal, and that agents fail for lack of context and control.
Then the pressure shifted to growth. By Q2, AI products made up 25 percent of Asana's net new ARR, up from 17 percent the quarter before. With that number climbing, the company shipped floor agents that hand off at exactly the point Rogers said agents break.
The partner channel filled the gap with an ownership number. Assigning a task in software and a supervisor accepting the outcome are different events.
The sales motion is treating them as one.
SYSTEM DRIVER - MOS
Left alone, this produces a pile of escalations with timestamps and no accountable name on the result. The fix goes in before any agent goes live. Every escalation type the agent can trigger gets a pre-named owner by role. Escalation-to-ownership goes on the tier board as its own measure, separate from response time. If that number isn't visible today, an agent will speed up the first leg and hide where the hours really go.
LEADER DRIVER - INTERNAL OPERATING SYSTEM (IOS) - REGULATE
The likely internal pull here is relief. A leader covering a supervisor gap carries the coordination load in their body, and a tool that promises to absorb it probably lands as rescue more than as a capability decision. Relief lowers the bar for evidence. The override threshold is one rule held even when tired: nothing gets signed on a number that can't be traced to a named customer.
IF YOU DO ONE THING TODAY
Pull your last five line-stop or quality escalations.
For each one, write down two times:
When it was escalated.
When one named person accepted the outcome.
Subtract. If the second gap is longer than the escalation took, an agent will accelerate the wrong leg. Put those five numbers on tomorrow's tier board before anyone books a vendor call.
FINAL SIGNAL
An agent can deliver a problem to the right desk in seconds, and it will sit there exactly as long as it always did.
CTA
Send this to the plant manager with an AI agent demo on the calendar, and have them run their escalation numbers at MOSei.org before that meeting.
SOURCES
Asana press release via Business Wire (2026-06-04, London) · SiliconANGLE (2026-06-04) · Asana Q1 FY2027 results and StackAI acquisition (2026-05-28) · Asana Q2 FY2027 results and earnings call (2026-09-03) · Business Wire (2025-09-25, Dan Rogers remarks) · Cloudfresh Asana partner page (2026-03-20), figures partner-reported.