SAP and Oracle pitch autonomous enterprise AI with human oversight, while Workday's Kazmaier dismisses the autonomy label
iTWire's comparison finds SAP's and Oracle's own announcements echo Workday's autopilot pitch. Buyers can separate the three only by testing which decisions each product hands to a person and what that person can change.
The Board Room · Leadership desk
SAP: human approval before action, monitoring, override. Oracle's agentic apps act within guardrails and surface decisions for human judgment. Workday's Kazmaier: in, on and out of the loop. Oracle: more than 600 agents at no additional licence fee; Workday charges via Flex Credits.
- What's now possible capability SAP Three oversight levels: human approval before action, monitoring during operation, override authority, claim 4
- What's now possible capability Oracle Fusion Agentic Apps Progress routine actions within guardrails; surface decisions where human judgment changes the outcome, claim 7
- What's now possible capability Workday's Kazmaier In, on and out of the loop model; iTWire calls SAP's three-level model a close cousin, claim 6
- What it costs cost Oracle Fusion Applications More than 600 AI agents embedded at no additional licence fee, Oracle said, claim 9
- What it costs cost Workday agent use Charged through Flex Credits, per iTWire, claim 9
| Who | How | Kind | Claim |
|---|---|---|---|
| SAP | Three oversight levels: human approval before action, monitoring during operation, override authority | capability | 4 |
| Oracle Fusion Agentic Apps | Progress routine actions within guardrails; surface decisions where human judgment changes the outcome | capability | 7 |
| Workday's Kazmaier | In, on and out of the loop model; iTWire calls SAP's three-level model a close cousin | capability | 6 |
| Oracle Fusion Applications | More than 600 AI agents embedded at no additional licence fee, Oracle said | cost | 9 |
| Workday agent use | Charged through Flex Credits, per iTWire | cost | 9 |
What happened
- Gerrit Kazmaier's jab at the "autonomous enterprise" as cold and soulless was, iTWire says, almost certainly aimed at SAP, which unveiled that strategy at Sapphire in May.
- Kazmaier told journalists at Rising that "almost right is wrong", which iTWire says is almost word for word what SAP's Christian Klein said at Sapphire.
- Oracle added Fusion Claw on 29 September, an execution runtime it says pairs AI reasoning with deterministic computation and delivers "governed autonomy."
- Oracle said more than 600 AI agents are going into Fusion Applications with no extra licence fee attached, while Workday charges for agent use through Flex Credits.
- Workday CEO Aneel Bhusri told his Rising audience he still thinks SAP and Oracle are "dinosaurs."
Why it matters
- Who has to decide decision Since SAP, Oracle and Workday all describe people supervising agents, the pitch cannot separate them. A buyer has to test what each product routes to a person and what that person can change.
- What it costs cost Workday's charge for agent use grows with volume, while Oracle's bundled agents add no licence fee, so which is cheaper depends on how heavily a buyer expects to run them.
- What doesn't add up contradiction Workday's architecture advantage is asserted by its CEO and only partly credited by iTWire, so it belongs in a pilot's success criteria and not on a comparison slide as settled.
The overlap goes past tone into structure. SAP's chief security officer, Marielle Ehrmann, described oversight in three tiers: a person approves before an agent acts, a person monitors it while it runs, and a person holds the authority to override it [4]. She said: "The more autonomously an agent operates, the tighter the oversight applied." [5] iTWire calls that a close cousin of Kazmaier's in, on and out of the loop model [6]. At SAP Connect, CEO Christian Klein described the goal as moving "from humans in spreadsheets to humans in and on the loop." [3]
Oracle's wording is the closest thing in the text to a claim about decision support. Its Fusion Agentic Applications, launched in March, "autonomously progress routine actions within guardrails" and surface "exceptions, tradeoffs, and decisions where human judgment materially changes the outcome" [7]. That describes what the software hands to a person. In our view it is the test to put to Workday: which decisions its agents route to a human, and whether the human gets enough to change the result.
Bhusri's answer is architecture. He said: "Their technology is outdated and they're not architected for AI. They don't have a single data model." [12] Workday, he argued, serves about 80 million users from a single data model, a single security model and a single software version [13]. iTWire judges him partly right: Workday's software has been a single cloud codebase since 2005, with all of its customers on the same version, while SAP and Oracle manage large legacy estates beside their cloud suites [14]. It adds that "dinosaurs" overstates the case, because Oracle's Fusion suite is a modern cloud product too [15]. In our view that claim can be tested in a pilot on the buyer's own records. A keynote cannot show it.
The same iTWire piece asks whether Workday's AI helps users decide better, and whether the C1.ai acquisition is really about replacing Okta [16]. The text available ends before it answers either one, and nothing in it links C1.ai to identity and access tooling, so the Okta reading is unsupported for now.
A buyer signing agent terms this quarter does so before Oracle's next AI World in Las Vegas, which runs 26 to 29 October. iTWire's note on that event is brief: "so expect a reply." [10]
What to watch
- Whether Oracle changes its agent pricing or packaging at its next AI World.
- Any Workday description of what C1.ai's product covers, which would show whether it touches identity and access at all.
- Whether Workday spells out its in, on and out of the loop model in product terms as specific as SAP's three oversight levels.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+30
- Incentives75
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Workday's Gerrit Kazmaier dismissed the "autonomous enterprise" as "a very cold and soulless way to think about an enterprise." iTWire wrote that he was almost certainly pointing at SAP, which unveiled its Autonomous Enterprise strategy at Sapphire in May and showcased it at SAP Connect in Las Vegas the week before Workday Rising.
- [2]
Kazmaier told journalists at Workday Rising "almost right is wrong", which iTWire says is almost word for word what SAP CEO Christian Klein said at Sapphire.
- [3]
At SAP Connect, Christian Klein described SAP's goal as moving "from humans in spreadsheets to humans in and on the loop."
- [4]
SAP Chief Security Officer Marielle Ehrmann set out three levels of oversight: human approval before action, human monitoring during operation and human override authority.
- [5]
Ehrmann said: "The more autonomously an agent operates, the tighter the oversight applied."
- [6]
iTWire calls Ehrmann's three-level oversight model a close cousin of Kazmaier's in, on and out of the loop model.
- [7]
Oracle's Fusion Agentic Applications, launched in March, "autonomously progress routine actions within guardrails" and surface "exceptions, tradeoffs, and decisions where human judgment materially changes the outcome."
- [8]
On 29 September Oracle added Fusion Claw, an execution runtime that "pairs AI reasoning with deterministic enterprise computation" and promises "autonomous enterprise execution" with "governed autonomy."
- [9]
At Oracle AI World 2025 in Las Vegas, Oracle said it was embedding more than 600 AI agents in Fusion Applications at no additional licence fee. Workday charges for agent use through Flex Credits.
- [10]
Oracle's next AI World runs in Las Vegas from 26 to 29 October; iTWire writes "so expect a reply."
- [11]
In his Workday Rising keynote, Aneel Bhusri said of SAP and Oracle: "I still think they're dinosaurs. I'll always think that."
- [12]
Bhusri said: "Their technology is outdated and they're not architected for AI. They don't have a single data model."
- [13]
Bhusri argued that Workday has one data model, one security model and one version of the software across about 80 million users, which gives its AI cleaner data and context.
- [14]
iTWire judges Bhusri partly right on architecture: Workday was built as a single cloud codebase from 2005 and every customer runs the same version, while SAP and Oracle manage big legacy estates alongside their cloud suites.
- [15]
iTWire says "dinosaurs" overstates the case because Oracle's Fusion suite is a modern cloud product too.
- [16]
iTWire's piece asks four questions, including whether Workday's AI helps its users make better decisions and whether the C1.ai acquisition is really about replacing Okta.
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.
Topics
- Human-in-the-Loop OversightFollow
- Enterprise AI AgentsFollow
- Enterprise Software PricingFollow