Workday splits human oversight of its autopilot agents into three tiers
Workday told Rising 2026 attendees its new AI agents are autopilots that work on their own, iTWire reported. Executives also insisted humans stay "in the loop", so the open question for a buyer is what that oversight covers task by task.
The Board Room · Leadership desk
Four kinds of work. Hiring decisions: humans stay in the loop; Workday does not automate them. Travel and expense approvals: humans on the loop, kept informed. Accounting record checks: humans out of the loop. Letting folks go and rejecting hires: Hellermark's red lines.
- What it limits constraint Hiring decisions Humans stay in the loop; Kazmaier: 'Workday does not automate hiring decisions', claim 3
- Who's exposed exposure Travel and expense approvals Humans sit on the loop, kept informed of what the AI has done, claim 4
- What's now possible capability Checks on accounting records Humans out of the loop; Kazmaier says AI is 'way more accurate than a human ever could be', claim 5
- What it limits constraint Letting go, rejecting hires Hellermark names decisions on these as his own red lines, claim 8
| Who | How | Kind | Claim |
|---|---|---|---|
| Hiring decisions | Humans stay in the loop; Kazmaier: 'Workday does not automate hiring decisions' | constraint | 3 |
| Travel and expense approvals | Humans sit on the loop, kept informed of what the AI has done | exposure | 4 |
| Checks on accounting records | Humans out of the loop; Kazmaier says AI is 'way more accurate than a human ever could be' | capability | 5 |
| Letting go, rejecting hires | Hellermark names decisions on these as his own red lines | constraint | 8 |
What happened
- Gerrit Kazmaier, Workday's President of Product and Technology, said work is sorted into three tiers by how much a task affects a person: in the loop, on the loop and out of the loop.
- Humans stay in the loop for anything that "materially influences income and employment", and Kazmaier said "Workday does not automate hiring decisions."
- Transaction sampling and compliance checks on accounting records sit out of the loop, where Workday says AI is "way more accurate than a human ever could be."
- Chief AI Officer Joel Hellermark said in Workday's release that the agents get the work done on their own and bring people in when their judgement makes the difference.
Why it matters
- What it limits constraint Oversight attaches to what a task does to a person, so work outside income and employment can run with no human involved.
- Who's exposed exposure In the out-of-the-loop tier no person is positioned to catch an agent's error first, so the accuracy case has to hold on the buyer's own data.
- Who has to decide decision Placing each workflow in a tier is the buyer's call this quarter, and the first disputed agent action next quarter will test whether it was placed correctly.
In our view, "in the loop" covers less ground than it sounds, and the executives' own quotes show where it stops. iTWire's author asked who is liable when an agent gets something wrong. Workday CEO Aneel Bhusri said: "Our model for many of these things is that there's always a human in the loop to make sure that the right things are being done, but the human doesn't have to do all the mundane tasking." [6] The word "many" leaves room for the third tier, where people are out of the loop altogether [5].
The first tier has two named red lines. Hellermark listed his: "Decisions around letting folks go. Decisions around rejecting hires." [8] Beyond those, the boundary is Workday's own test of what "materially influences income and employment" [3].
The middle tier is the least defined. Travel and expense approvals sit there, and humans are kept informed of what the AI has done [4]. iTWire's account does not report whether a person informed of what an agent has done can reverse it, and the answer it quotes on liability does not say who is liable.
The commercial logic pulls the other way. Kazmaier asked from the keynote stage, "What do you get when you make every little thing 10% better?" He answered, "You get 10% better." [12] Hellermark made the same case in his media Q&A. A copilot-style interface that makes each employee 10% more productive yields 10%. Re-architecting the process so "an autopilot run it end to end" yields "an order of magnitude greater productivity gains." [13] We think that puts the productivity claim and the oversight claim in direct tension.
iTWire's author put the question plainly: "So is that a contradiction, or a new model of how people and AI work together?" The author weighed it over a keynote, four media sessions and two one-on-one interviews. The verdict was the latter, "with some caveats worth spelling out." [9] The tiers support that reading for hiring and dismissal, where Kazmaier and Hellermark name the same lines [3][8].
What to watch
- Part 2 of iTWire's analysis, which compares Workday's pitch with SAP and Oracle, for whether rivals draw the same tier lines.
- Whether Workday documents what a person on the loop can do after an agent acts, including reversing it.
- Whether Workday backs its accuracy claim for out-of-the-loop tasks with customer data.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption
- Not scored
- Hype gap+40
- Incentives70
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
At Workday Rising 2026 in Las Vegas, Workday dropped the 'copilot, not autopilot' line; its new agents are autopilots built to work on their own, around the clock, in the background.
- [2]
Workday President, Product and Technology, Gerrit Kazmaier explained in a media Q&A that 'the loop' comes in three levels: Workday sorts work into three tiers based on how much it affects a person (in the loop, on the loop, out of the loop).
- [3]
Humans stay in the loop for anything that 'materially influences income and employment', like a hiring decision; Kazmaier said, 'Workday does not automate hiring decisions.'
- [4]
Humans sit on the loop, kept informed of what the AI has done, for things like travel and expense approvals.
- [5]
Humans are out of the loop for tasks like transaction sampling and compliance checks on accounting records, where AI is 'way more accurate than a human ever could be' and 'it's not directly impacting a human or person having access to opportunity.'
- [6]
Asked by iTWire's author who is liable when an agent gets something wrong, Workday CEO Aneel Bhusri said: 'Our model for many of these things is that there's always a human in the loop to make sure that the right things are being done, but the human doesn't have to do all the mundane tasking.'
- [7]
Workday Chief AI Officer Joel Hellermark said in the autopilot agents release: 'Our agents work like autopilots, they get the work done on their own, and bring people in exactly when their judgement makes the difference.'
- [8]
Hellermark named his own red lines: 'Decisions around letting folks go. Decisions around rejecting hires.'
- [9]
iTWire's author asked 'So is that a contradiction, or a new model of how people and AI work together?' and, after a keynote, four media sessions and two one-on-one interviews, wrote 'I think it's the latter, with some caveats worth spelling out.'
- [10]
Part 2 of iTWire's analysis compares Workday's pitch with what SAP and Oracle are saying, and asks whether any of it helps people make better decisions and what Workday still doesn't do.
- [11]
iTWire's author wrote that every Workday executive he heard insisted the human stays 'in the loop.'
- [12]
Kazmaier said from the keynote stage: 'What do you get when you make every little thing 10% better?' he asked. 'You get 10% better.'
- [13]
In his media Q&A Hellermark said a copilot-style interface that makes each employee 10% more productive, stacked on existing processes, gets 10%; re-architecting the process so 'an autopilot run it end to end' gives 'an order of magnitude greater productivity gains.'
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.