90% of firms adopt AI, but only 27% see real gains. Workday’s Chief AI Officer breaks down how to escape "lawless AI" and drive true enterprise transformation.
Ghadeer Redler
Head of Multimedia, Thought Leadership & Customer Stories
Workday
AI is a ubiquitous part of enterprise operations today; global enterprise adoption is nearing 90%, spanning functional use cases and powering everyday operations.
But despite AI’s growing presence, a stark divide is emerging between the number of companies using AI and those yielding true gains from it. Workday’s most recent report, The Copy/Paste Economy: Why Task-Based AI is Failing the Enterprise, found that just 27% of organizations have actually embedded AI into their core workflows.
The rest have it sitting on the periphery—useful, but not transformative. In many cases, the lack of true strategic implementation is exactly what holds companies back from transformation.
Few people are better placed to explore that gap than Workday Chief AI Officer Joel Hellermark. Joel was founder and CEO of Sana, an AI platform, acquired last year by Workday, that helps enterprises connect knowledge with intelligent, autonomous action. Joel works daily with organizations that want to maximize their AI investments in ways that move the needle.
In a recent Future of Work podcast episode, Hellermark explains what separates those genuinely transforming with AI from those still waiting for returns. Here are five things we learned from that conversation.
Report
To drive AI transformation, organizations must shift from reactive to proactive.
“We’re moving from copilots that we prompt, to agents that proactively serve us with actions and insights before we even ask,” says Hellermark. “When those agents are embedded into the systems people use every day and can handle end‑to‑end work rather than isolated tasks, the way an organization operates starts to change fundamentally.”
Right now, most organizations are still in an adoption phase—introducing copilots, piloting use cases, and layering AI onto existing processes. Transformation is a different kind of change. It’s when agents are trusted to own end-to-end workflows, so time savings come from how the process itself is redesigned, rather than simply pushing more tasks through old automated pipelines.
More or newer AI isn’t always the answer. Without the right context, even the most powerful models don’t add much productivity. Instead, they generate more output that humans still have to interpret, correct, and reconcile.
Today, people are often doing the “context engineering” themselves, providing background, clarifications, and instructions for every task. That works for short or simple use cases. It breaks the moment you ask AI to handle anything longer or more complex.
In addition, Hellermark emphasized how the capability of AI is growing fast: the length of tasks it can solve is roughly doubling every seven months. Enterprise AI can only handle that kind of progression if context is embedded in the system itself: the data, rules, permissions, and history around each decision.
When that context lives inside tools, AI agents can carry it with them across every step of a process, even when the process changes.
Without the right context, even the most powerful models don’t add much productivity.
Many organizations are still in what Hellermark calls a “lawless” paradigm—agents operating on the edges, without the context they need or reliable access to company policies and source‑of‑truth data. Those agents can be useful for experimentation, but they are hard to trust with real work and harder still to keep compliant.
The result is familiar: Decisions get delayed, people fall back on manual checks, and teams spin up their own workarounds outside core systems.
Running AI inside a system flips that equation. When AI agents are wired directly into core applications and data, they become “lawful agents” they know the rules, understand how the organization operates, and can be audited and governed like any other critical system.
That turns agents into always‑on, high‑IQ coworkers that run in the background, preparing work for humans to review, rather than asking teams to recreate context for every task.
Despite oft-heard concerns about AI replacing humans, Hellermark emphasizes that we’re not moving to an era of humans out of the picture. Instead, we’re moving through changing levels of autonomy.
Right now, humans are still heavily involved in final approvals, but the unit of work is changing. As Hellermark puts it, “instead of approving each individual task, you can approve bodies of work.” That’s the key shift: Agents now own more execution, while people stay responsible for the decision boundaries.
He uses a self‑driving car analogy to drive the point home. Today, he says, we’re still “sitting in the car,” ready to correct the AI if it drifts. For knowledge work, we’re in that middle zone where agents can handle more of the journey, but a human is still there to supervise and step in when needed. Over time, AI systems will get better at knowing when they need to loop in a human and when they can act on policy and prior examples without asking for help.
On trust, Hellermark predicts humans will increasingly use AI to double‑check decisions from other systems or experts, but they still won’t want to rely purely on the AI either. The most value comes from combining the two.
“You want the expansive knowledge AI can offer,” he notes, “With the judgment and nuance and expertise of real humans. Having both is the best scenario.”
Agents now own more execution, while people stay responsible for the decision boundaries.
The organizations making the biggest gains aren’t just buying more tools; they’re rethinking how they’re structured to work with agents.
“We’re moving into a totally new category,” says Hellermark, “Where you can create an almost endless number of agents tailored to specific workflows and outcomes. As customers move from reactive copilots to more autonomous agents, the impact can be dramatic, but only if they’re ready to absorb that change.
Readiness starts with structure. In Hellermark’s model, there are three pillars: leaders who genuinely use AI so it naturally permeates how they manage; a central lab that sets principles, evaluates tools, and builds the internal rails agents run on; and a crowd made up of the knowledgeable, curious people in each function who become AI ambassadors on their teams.
Together, they allow AI to be universally embedded in how work gets done, rather than confined to side projects or pilots, and they give organizations the flexibility to keep evolving as transformations speed up from once a decade to almost every year.
Across these lessons, Hellermark’s message is consistent: AI on its own doesn’t transform an organization. Transformation comes from moving beyond reactive copilots to proactive agents, embedding context into your core systems, keeping humans in the loop at the right moments, and redesigning how work flows so each part of the system—human and machine—is doing what it is uniquely best at.
That’s ultimately the opportunity: to build organizations where agents handle more of the execution, humans focus on judgment and direction, and the underlying systems are flexible enough to keep evolving as the technology advances.
For more insight on enterprise AI transformation, listen to the Future of Work podcast episode with Joel Hellermark to learn how to build an autonomous enterprise.
Report
The first era of AI made workers faster at broken processes. The next era is AI that does the work.
Discover a new framework for what enterprise AI actually looks like.