By Jim Kent

Based on a podcast with Bob Sternfels at theHarvard Business Review

The Luxembourg conversation around digital agents often stays abstract, proof-of-concept pilots, isolated use cases, cautious experimentation. What makes the experience of McKinsey & Company particularly instructive is that it has moved far beyond theory. As Bob Sternfels, their CEO puts it, “When asked about how many staff we have I now update this almost every month, but my latest answer would be 60,000, 40,000 humans and 20,000 agents.”

That shift did not happen gradually. “A little over a year and a half ago, that was 3,000 agents,” Sternfels explains. “I originally thought it would take us to 2030 to get to one agent per human. I think we’re going to be there in 18 months.” The first lesson is therefore speed. Digital agents do not scale linearly, they compound quickly once embedded into real workflows.

Yet Sternfels is clear that technology alone does not create value. “Half, if not more, of the secret sauce is organisational change as opposed to technology implementation,” he says. “This is for large enterprises.” Many organisations invest heavily in AI, only to discover that existing structures block impact. “What does your work look like after you’re implementing these? Could you have a much flatter organisation that cuts out a lot of middle layers and makes your organisation faster?”

He uses complex processes as a warning sign. “When you think about really complicated workflows, like a mortgage process, you’ve got origination, credit scoring, collections, customer service, all separate departments. Why do you have four or five departments in a process if you can really enable this through AI?” Digital agents expose inefficiencies that were previously hidden behind organisational silos.

Another key insight is alignment at the top. Sternfels describes a familiar tension. “CEOs tell me, ‘Do I listen to my CFO, who says we’re spending a lot and not seeing enterprise-level value yet, or my CIO, who says if we’re not in the lead we’ll get disrupted?’” The answer, he argues, lies in redesigning the organisation. “If you get that right, all of a sudden the CFO and the CIO are on the same page.”

McKinsey has applied the same logic to its own business model. “We’re migrating pretty quickly away from pure advisory work,” Sternfels says, “towards an outcomes-based model where we underwrite results alongside our clients.” Today, “about a third of our revenues are tied to underwriting outcomes.”

Finally, Sternfels cautions against assuming agents replace humans. “The models are inference models. They’re great at linear problem solving. What they’re not great at is discontinuous leaps.” Leadership, judgment and aspiration remain human responsibilities. “The models don’t aspire. They don’t set the ambition. Humans still do.”

For organisations considering digital agents, McKinsey’s experience suggests a simple but demanding truth, success is less about deploying AI, and far more about having the courage to redesign how work, decisions and accountability really function.