Workday’s Innovation Summit this spring brought industry analysts and customer advisors together for two days focused on a single topic: what enterprise-grade agentic AI actually looks like, and what it will take to deliver it responsibly. One analyst called it “the most innovative Workday Innovation Summit yet.” The pace of product news, the directness of executive answers on hard questions, and the visible reorganization around AI all pointed to a company moving differently than it was twelve months ago.
The source of change is clear. In February 2026, co-founder Aneel Bhusri returned as CEO, succeeding Carl Eschenbach. His return was framed at the Summit as something closer to Steve Jobs’s return to Apple than a routine succession: leadership talked about Workday being “rebuilt as a startup,” about “speed, urgency, and ownership,” and about an ambition to be “the Apple of enterprise software.” Aneel called it “Workday 4.0.” The Jobs comparison may feel like a stretch, but spending two days in the room made it less so. Bhusri has reclaimed the parts of the original Workday ethos he cares most about—product discipline, customer trust, and a willingness to make hard architectural choices—and is using that authority to drive an AI pivot the company believes will define its next decade.
Startup pace is evident as Workday’s flurry of AI startup acquisitions has already resulted in the generally available product being delivered to customers with Sana as the master agent interface, Paradox as the hiring and candidate front door, and Pipedream as the agent builder.
For higher education leaders, the Summit offered two distinct sets of takeaways: how Workday is thinking about AI, and what those choices mean for colleges and universities.
Workday’s clearest conviction is that the prevailing market narrative—bigger models, broader autonomy, more general intelligence—is the wrong frame for enterprise work. Its bet is on vertical AI: role-aligned mega-agents (other companies are using the term agentic applications—which seems to align well to these mega-agents) built on top of a set of narrow, role-aware agents, deeply integrated with a system of record, and constrained by the same governance, authorization, and audit requirements as any other enterprise software. They called it “vertical AI, not AGI.” Part of the argument is that ROI is measurable in vertical domains and largely unmeasurable in general ones.
The phrase that captured the contrast was the “lawless agent”—one that can do almost anything but lacks the rails an enterprise needs: an identity, a defined scope of authority, an audit trail, and accountability when things go wrong. No serious CIO will put a lawless agent in production against payroll, financials, or sensitive employee data. The downside risk is too large. Asked what worried him most, Bhusri gave a one-word answer: trust. He framed it as trust built over decades that can be lost in a second.
Workday’s alternative reconceives the system of record as a platform for agents, turns its existing configuration, data, and process layers into “enterprise rails” any agent must run on, and gives customers a path to move quickly without building a fragile shadow system on the side. Underpinning this is a layered trust model (runtime, agent code, and language model) that lets enterprises apply governance at each level rather than treating the agent as a black box.
This discipline is one of the visible signatures of Bhusri’s return. Workday consolidated its agent efforts from roughly fifty parallel initiatives down to fifteen, with single-owner accountability for each under a centralized “agent factory” organization (run by Jerry Ting, founder of the acquired AI startup Evisort). The reorganization is both pragmatic and philosophical. Workday is deliberately not trying to be the company with the most agents; it is trying to be the company whose agents you can deploy in regulated, auditable, mission-critical contexts—the path customers in regulated industries, including higher education, are saying they need.
If the lawless agent was the failure mode at the agent level, the “shadow ERP” was its institutional counterpart. The pattern is familiar: a department, frustrated by the central system, stitches together its own collection of point tools, data extracts, and now agents. Each piece is reasonable on its own; collectively, they constitute a parallel system of record that the institution doesn’t fully see, can’t fully govern, and can’t easily reconcile.
Workday’s response is to make the central platform a place where customers can move fast without going around it: an agent gateway and security framework for governed third-party access, lower friction to turn on new AI features, and a monthly delivery cadence that keeps native capabilities ahead of the workflows where shadow tools spring up. In higher education, where decentralization is a feature—and where autonomy has produced an unusually rich landscape of side systems that AI agents will only accelerate, this is a critical architectural offering.
Workday introduced a useful frame: search retrieves, assistants explain, agents execute. The jump to execution is qualitatively different. It is the point at which AI stops handing work back to a human and starts closing the loop—running a multi-step workflow, getting approvals, enforcing policy, and logging the action.
That shift is what justifies the talk about enterprise rails. Search and assistants can be wrong; the cost is mostly that the user thinks harder. Agents that execute can also be wrong, but the cost is a real-world action: a payment made, an offer extended, a record changed. An agent that can execute without rails isn’t an upgrade over an assistant; it’s a liability. The frame is worth borrowing whenever an institution evaluates an AI tool: the right question is no longer “can it answer this?” but “can it finish this—and if it can, do we have the rails to let it?”
Asked about industry-specific agent strategies, Workday leadership noted that the student system has historically followed HR and finance in major technology waves, and there is no reason to think the AI agent wave will be different. Cross-industry agents are being proven out first; a more developed agent roadmap for higher education is actively being prepared.
That leaves a short window to prepare. The capabilities being shaped now in HR and finance are the same primitives that will reach into student-facing workflows: advising, financial aid packaging, registration exception handling, degree audit, holds management, and billing. Institutions ready to take advantage will be the ones that have already cleaned up student data, documented processes that have lived in tribal knowledge, and defined policy boundaries clear enough for an agent to be told what is and isn’t in scope. These requirements highlight the advantage that institutions that have begun or completed modernization will have in being able to take advantage of these new tools.
In most industries, identity is reasonably clean: there are employees, there are contractors, and the boundary is well-defined. Higher ed has never looked like that. A single institution maintains identity for employees, faculty (tenure-track, contingent, and emeritus), graduate students who are also employees, undergraduates with work-study positions, post-docs, alumni, contractors, and a long tail of affiliates. Each population has its own access patterns, lifecycle, and policy constraints.
This complexity has mostly been a tax. In an agentic world, it starts to look more like an asset—but only for institutions that get the underlying identity layer right. Agents themselves are about to become identity actors, with accounts, scopes of authority, and audit trails. Institutions with rigorous answers to “who is this person, and what are they allowed to do on whose behalf?” can extend those answers to agents. Those still treating identity as departmental workarounds will find the agent layer multiplies their existing problems. And students, who occupy a role between customer and employee, raise delegation, consent, and FERPA questions that other industries are not wrestling with at the same depth.
None of this is free, and the commercial model—Flex Credits—was one of the most candid topics of the Summit. Launched in late 2025, Flex Credits are consumption-based: customers receive an annual allotment included in their core subscription and purchase additional credits as agent usage scales. Workday sees consumption as a structural part of its growth story, betting that AI will create enough customer value to justify spend that grows with adoption. As one Workday leader put it, the company is not looking for “revenue at all costs” but for “revenue based on meaningful results.”
That framing lands differently in higher education. Universities operate on essentially flat annual budgets—state appropriations, tuition, and endowment draws don’t scale up just because a finance agent is producing value. Net-new operating spend has to come from somewhere, and ROI has to be visible to a CFO and a board. A consumption model that rewards usage with more usage is a reasonable bet for a growing corporate buyer; for a tuition-dependent institution, it can feel like a meter running on top of a budget that wasn’t built to absorb it.
Workday’s executives acknowledged the tension and were not satisfied with where Flex Credits sit today. They had roughly 150 customers on the model, and analysts in the room reported that customers “still don’t really understand Flex Credits” and “see it as additive and punitive.” The performance console that would let customers see, in real time, what their agents are producing for what they are spending is still being built. Bhusri compared the transition to “flying a plane and changing the wheels.” The early missteps have been less about strategy than about transparency: a rate card that isn’t fully public, dashboards that aren’t yet mature, spending controls that are not granular, and a value story customers are being asked to take on faith.
The speed and agility these tools promise will be critical to modernizing higher ed business operations. But institutions should go in with eyes open: insist on transparent pricing and consumption forecasting; model scenarios for low, medium, and high adoption; and work internally to tie credit allocation to specific KPIs. Open-ended exploration is hard to justify against a flat budget.
The most striking moment of the Summit was Bhusri’s response to a question about job displacement, posed by an attendee from higher education. He did not soften it. He said directly that he believes there is a reasonable chance the broader economy reaches significant unemployment in the next couple of years, that entry-level roles are particularly exposed, and that he considers it a personal responsibility for Workday to be part of the response. He noted, pointedly, that other AI labs have been willing to talk about the scale of the displacement but less willing to offer solutions, and signaled that Workday will announce a meaningful retraining initiative in the coming months. “For me to be able to go to sleep at night,” he said, “we have to be part of it.”
For higher education, this touches the center of the value proposition of an undergraduate degree. If the entry-level analyst, coordinator, and associate roles that absorb new graduates contract sharply, the path from college to career—already strained—becomes substantially harder to defend. Career services offices designed around stable employer pipelines will reshape faster than they can adapt; curriculum committees on multi-year cycles will be asked to respond to labor-market shifts measured in months.
Useful starting points include active dialogue with employer partners about which roles are changing, on what timeline, and what skills will be required. Universities also need a willingness to accept that the answers may challenge existing program assumptions; engagement with emerging reskilling infrastructure so they help shape it rather than react to it; and a renewed case for human capabilities that do not reduce easily to entry-level task automation, including judgment, communication, ethical reasoning, and the ability to work with AI rather than be replaced by it.
The next phase of enterprise AI will not be won by whoever ships the most agents or the newest models the fastest. It will be won by whoever can deliver agents institutions can trust to execute work in regulated, audited, accountable ways and who can help customers get there without quietly fragmenting their systems or their budgets. That argument is now being driven personally by a returning founder who has staked his legacy on getting it right. For higher education, the institutions that engage seriously with what enterprise-grade AI requires—including what it costs, who it displaces, and what it reshapes—will be in a much stronger position when the agent wave reaches the student system. Those who wait will find that the questions are no longer hypothetical but key to operational strategy and planning. Workday is positioning itself as a competent, trusted partner to accelerate agentic work in the enterprise. Its releases over the coming year will show what the value and the costs of this strategy will be.
Originally posted by Dave Kieffer on LinkedIn. Be sure to follow him there to catch all his great industry insights.
Share Article:
© Copyright 2026, The Tambellini Group. All Rights Reserved.
Get exclusive access to higher education analysts, rich research, premium publications, and advisory services.
Weekly email featuring higher education blog articles, infographics or podcasts.
