Julia Johnson marked her second day on the job at Cognizant with a history lesson. As the company’s new SVP, global talent management leader, she reminded the room that IBM, which spent 115 years building one of the world’s most recognized brands, was once called Computing, Tabulating and Recording Company. The rebranding happened in 1924. The lesson? What we call things matters, and the names we give new technologies shape how we use them.

“If we had a time machine,” Johnson said, “we would rename it augmented intelligence, because it really requires having a human complete it—not being a rubber stamp,” Johnson said during an executive panel discussion at From Day One’s Manhattan conference. 

Moderator Lydia Dishman, SVP of content strategy, narrative and thought leadership at Method Communications, opened by citing a striking data point: 88% of HR leaders say their organizations have not yet realized significant business value from AI tools, according to a recent survey. The question the panel had gathered to answer wasn’t whether AI would transform work, it’s already doing that, but how to move from experimentation to real transformation while keeping the human part of work intact.

Job Elimination Is the Wrong Frame

The most persistent misconception about AI, panelists agreed, is the idea that it eliminates jobs wholesale.

“AI is really, really good at doing certain tasks,” said Scott Turner, partner at Mercer, who previously built agentic AI systems at Disney. “A job is a whole stack of tasks. Replacing a job is a human decision. If all those tasks in a job can be easily replaced by AI, perhaps you didn’t design that great a job for the human in the first place.”

Owen O’Neill, executive director of HR technology and operations at Regeneron Pharmaceuticals, pushed back on the broader noise around AI: “Everybody needs to do what’s right at the pace that’s right for your organization,” he said. “[There’s] so much external noise, a lot of it generated by big tech, who have invested a lot of money and need to start recouping.”

The flip side of that caution is not ignoring AI’s genuine implications. “What I cringe at is when people talk to their employees like, ‘Oh, this isn’t going to have an impact at all,’” Turner said. “That’s just disingenuous. It’s going to have an impact. Let’s try to do this thoughtfully.”

Transformation Begins With the Right Question

When organizations approach Mercer wanting to deploy AI in HR, Turner says the first question he asks is deceptively simple: What are you trying to improve? That question is the antidote to FOMO-driven adoption—the tendency to implement AI because competitors are doing it, or because a vendor has a compelling pitch. The most successful AI transformations he’s seen share a common trait: they identify specific, high-frequency workflows, redesign them around what AI does well, measure the results against clear KPIs, and keep humans meaningfully in the loop.

Johnson echoed this, pointing to one of IBM’s earliest high-impact use cases. Employment verification letters, the kind a senior manager needs urgently when closing on a home, used to require hours of back-and-forth. Now they’re generated in any country, in 38 seconds or less, around the clock. “Be pragmatic, have the use case, look at the ROI, embrace what will be used,” she said about the experience in her former role. 

O’Neill put it plainly: “Tell me what your HR priorities are and what your strategy is, and I will tell you what our AI roadmap is to enable that. Start with what those priorities are, not the technology.”

Efficiency gains are real, but panelists were candid about areas where the business case doesn’t hold up under scrutiny.

Resume screening is one. O’Neill noted that Regeneron now receives roughly 6,000 applications for a single data analyst posting, making AI-assisted screening appear essential. But he was quick to identify the risk: “How we’ve hired in the past doesn’t necessarily reflect how we want to hire in the future. A good hire two years ago is not necessarily a good hire two years from now.”

Performance management is another. AI can remove some bias, consolidate feedback, and save managers time, but, O’Neill says, that misses the point. “Performance management is a social contract between an employee and a manager. Automating that risks dehumanizing it. It’s about the conversation, not the document.”

The Talent Pipeline Problem No One Is Solving

Dishman raised a concern that has received less attention than job elimination at the entry level: what happens to the pipeline that feeds middle management when the entry-level roles that have historically developed that talent disappear?

Paul Tiesler, SVP of talent development and learning strategy at Moody’s Corporation, offered a structural answer. The traditional pyramid-shaped org chart, he says, may need to become an hourglass. Under that model, entry-level employees sit alongside AI, learning from it and compressing their career timelines. Middle managers are elevated into more senior-level thinking as AI handles the processes that currently bog them down. The people organizations hire at both levels share a trait: strong judgment, discernment, and critical thinking, skills AI cannot replicate.

“You’re going to be hiring for exactly the same thing,” Tiesler said, “more so than technical skills, especially as AI is able to automate some of those technical skills.” Moody’s has already seen this play out within software and product development. “We sat down with them and said, ‘How can we make AI do this better for you,’” Tiesler said of its middle managers, “and they’ve been able to elevate their role, and juniors on their team are now getting to do more interesting work.”

Putting the Human In the Loop—Intentionally

Bill Beegle, senior global business solutions architect at Degreed, offered a different model for how AI can augment rather than automate: scenario-based role play. Degreed uses AI to help employees practice high-stakes conversations, difficult performance reviews, sensitive feedback, the transition from peer to manager, in a low-risk environment where they can make mistakes and receive structured feedback.

“Unlike automating a process, this is putting it like a flight simulator,” Beegle said. “You get to try, you get to practice, you can make as many mistakes as you want. You’re not really going to crash a plane, you’re just talking to AI.”

The use case has found particular traction in regulated industries like biopharma, where the wrong word in a conversation with a physician carries real consequences. And it represents something the panel returned to repeatedly: using AI not to remove the human, but to make the human better at the distinctly human parts of their job.

Johnson crystallized the logic: “What are humans no good at? Finding needles in haystacks. What does LLM do really well? They find needles in haystacks, or find trends. Look at what the human is good at and amplify that.”

Building Trust in Times of Change

The panel converged on change management as the most underrated element of AI adoption. Tiesler was direct about what doesn't work: “Edicts from down on high don’t work. Arbitrary ‘we’ve got to cut X percentage of headcount, we have to automate Y number of processes’ – that doesn’t really work.”

What does work, panelists agreed, is co-creation with employees – sitting down with business teams, mapping their actual processes, and identifying genuine opportunities for relief. Transparency matters too. Johnson described the framework she used at IBM: “We’re going to tell you what we are doing, why we are doing it, when we’re doing it, and how it will impact you. It’s not hard, but it’s so often overlooked.”

Beegle pointed to one practical lever organizations underuse: making skills transparent. When employees can see how their skills map to other internal roles and what would help them get there, the internal mobility conversation stops being abstract. “It’s a really important part, so people understand that it can benefit them.”

Closing the session, Dishman asked the panel directly: can leading with AI and keeping the human touch actually coexist? Every panelist said yes, with conditions.

Turner returned to the limits of what AI can actually do. Its model of truth is built entirely on language. “It has no concept that this is actually a chair and I’m touching it.” That gap between what AI can know and what humans embody through experience is permanent, and it’s where design comes in. “We are going to have a completely different set of knowledge than the LLM can ever have,” Turner said. “It’s about trying to find that balance of where it can be applied safely.”

O’Neill said on a closing note: “We’re at step zero of a race that is going to go a million miles. We’re right at the beginning.”

Grace Turney is a St. Louis-based writer, artist, and former librarian. See more of her work at graceturney17.wixsite.com/mysite.

(Photos by Josh Larson for From Day One)