Brian Reed: Hey. You saw the Workday news out of Las Vegas?
Eliza Ward: Yeah, just — okay, so Workday and Anthropic announced this morning, at Workday Rising 2026, that Claude is going inside Sana. Workday's AI agent platform. Running HR and finance workflows.
Brian Reed: So instead of a company buying a separate AI tool, Claude is just... already in the software they use to run payroll.
Eliza Ward: That's the structure, yes. And Workday serves 65% of the Fortune 500 — 11,500-plus organizations. So this isn't a small pilot.
Brian Reed: Wait — 65% of the Fortune 500 already on Workday. That's the part that actually lands.
Eliza Ward: Right. And then Gerrit Kazmaier — he's president of product and technology at Workday — he framed the whole thing as 'lawful AI across the enterprise.' That phrase is doing a lot of work.
Brian Reed: Lawful AI. What does that — I mean, is that a technical claim or a marketing claim?
Eliza Ward: Both, probably. Though as of October 13th, 'Sana in Claude' is early access. Selected customers. The majority of those 11,500-plus organizations can't actually touch it yet.
Brian Reed: Wait — so the announcement is real, but the product isn't broadly available.
Eliza Ward: General availability is 'later in 2026.' That's the actual timeline. Which is — I mean, that's genuinely vague.
Brian Reed: And what makes that vague timeline sting a little more — Ankur Sinha at Anthropic named compensation cycles and financial closes as the core use cases. Those aren't exploratory. Those are time-locked operations. You miss a September payroll close, that's not a minor drift, that's a broken calendar for the whole quarter. So the gap between 'we announced this' and 'HR teams are actually running their merit cycles through Claude' feels... wider than the headline suggests.
Eliza Ward: Right, and — wait, there's also this. Workday didn't just announce Sana in Claude at Rising 2026. They dropped 'Decision Intelligence' at the same event. Multiple AI features in one shot.
Brian Reed: So which of those is actually ready and which is aspirational? Because if you're a Workday customer watching that keynote, you're getting — let me see — you're getting Sana in Claude, Decision Intelligence, all at once. How do you even read what's shipping versus what's a roadmap item?
Eliza Ward: That's the real question. The Agent-Ready Tools — the toolkit for building custom agents inside Claude with Workday's governance controls — that's something joint customers are expected to use. Expected to. Not confirmed they can today.
Brian Reed: So what is actually confirmed here? The partnership is announced. Sana in Claude exists for a subset of customers. General availability — sometime in 2026. Everything else is design intent.
Eliza Ward: Yeah — and that's actually the thing I want to push on, because there's a take circulating that frames all of this as proof of Claude's enterprise strength. And I think that's backwards.
Eliza Ward: Anthropic showed up at Workday Rising. Workday's conference. In Las Vegas. That's not a neutral venue — Anthropic entered Workday's ecosystem. If Claude had dominant enterprise traction on its own, you don't need to announce inside someone else's customer conference.
Brian Reed: Hang on — I mean, isn't it possible both sides actually get something real? Workday gets frontier reasoning, Anthropic gets reach. That's not — that's not automatically a losing position for Anthropic.
Eliza Ward: Both sides get something, sure. But look at what Anthropic's enterprise growth actually looks like right now — Select Partner deals with Think41, Hexaware, X-Team. Those are the deployment routes. That's not a direct enterprise channel, that's a workaround.
Brian Reed: So the distribution problem is the real story. Not the reasoning capability.
Eliza Ward: Right — and here's what actually seals it for me. Workday announced a separate AI partnership with Teem Finance on September 28th. Two weeks before Rising. Integrating AI into Workday Strategic Sourcing. That's not a company betting on Claude. That's a company integrating whoever's useful.
Brian Reed: Wait — Teem Finance, September 28th. So Workday had already signed another AI deal two weeks earlier and nobody led with that.
Eliza Ward: Nobody. And compared to OpenAI and Google — who have direct enterprise channels — Anthropic's path runs through partners. This is that path. Which is fine, actually, as a strategy. Just don't call it a capability story. And the part that makes this harder — we'll get to it — is whether the governance architecture Workday's promising can actually hold when Claude is running agentic decisions inside a live compensation cycle. That's a completely different stress test.
Brian Reed: And that stress test is — let me see if I can make this concrete. Because the governance claim is the one that actually has to hold. Workday says existing permissions, business rules, approvals, audit controls — all of that stays in place inside the Claude integration. That's their assurance. But Sana agents are described as agentic. Not suggestions. Actions. Multi-step workflows that execute inside live systems.
Eliza Ward: Right — and that distinction matters enormously.
Brian Reed: So imagine a comp analyst at a large company — not Tuesday morning, but the last week of December, fiscal year-end close. She's finalizing equity refreshes. Claude, running through Sana, flags that a senior engineer is below market band and reasons toward a 14% increase. Workday's business rule says the approved budget cap this cycle is 9%. Does the system stop? Does it surface the conflict cleanly? Or does it — and this is what I actually don't know — does it reason past the constraint? Like, 'the rule exists but here's why this case warrants an exception.'
Eliza Ward: The sources don't answer that. There's no incident model. No 'here's what happens when Claude's logic conflicts with a business rule.'
Brian Reed: None. And Kazmaier's framing — 'lawful AI across the enterprise' — that's a design intent. It's not a field test result.
Eliza Ward: Wait — and then layer on top of that. Joint customers are supposed to build their own custom agents using Agent-Ready Tools. So you've got Workday's governance layer, Claude's reasoning, and then a customer-built agent sitting on top of both. That's — actually, that's three surfaces where something can drift.
Brian Reed: Three layers and none of them have run a live September payroll close yet.
Eliza Ward: Workday themselves positioned this as a platform where the margin for error is effectively zero. That's their reliability bar. Claude's reasoning hasn't been publicly shown to meet that at enterprise scale.
Brian Reed: So the thing to actually watch — is whether early-access customers surface any of that conflict publicly before general availability later in 2026. Because right now we're guessing at failure modes. But once real comp cycles run through this, someone will know.
Eliza Ward: And that's the honest edge of what we know. Design claim, not demonstrated outcome. The signal is whether any of those early-access customers say anything before GA drops.
Brian Reed: And that's — yeah, that's actually where I land too. The partnership is real. The governance claim is design intent. The majority of Workday's 11,500-plus customers still can't use Sana in Claude at all — general availability hasn't happened yet. So the thing we don't know isn't whether the deal exists. It's whether the rails hold when you go from a handful of early-access customers to hundreds of HR teams running actual payroll and financial close.
Eliza Ward: Right. And I'd — wait, let me actually sharpen that. The specific signal I'd watch isn't a press release. It's whether incident rates stay low as the user base scales. Because if something drifts at a Fortune 500 comp cycle, that doesn't go in a blog post. It goes to legal.
Brian Reed: That data doesn't exist yet.
Eliza Ward: No. It doesn't. And until it does, 'lawful AI across the enterprise' is Gerrit Kazmaier's framing, not a demonstrated outcome.
Brian Reed: Good place to stop.
