Where we heard itCNBC’s Squawk Box, Oct. 16, 2026. Speaker: Ethan Mollick, Wharton School professor, co-director of Generative AI Labs at Wharton and author of the new book Co-Existence.
Ethan Mollick, author and professor at the University of Pennsylvania’s Wharton School, says the AI story has moved on. The question used to be whether the technology helps people. Now it is whether it helps organizations. Speaking on CNBC’s Squawk Box on October 16, 2026 about his new book and the next phase of AI, Mollick, whom CNBC described on air as co-director of Wharton’s generative AI lab, explained the gap: “We now have really overwhelming evidence that at the individual level, AI improves performance. And I talk to CEOs of companies all the time who are seeing fairly large ROI, but there’s a big gap between individuals performing better and the company performing better.”
On what happens when that gap goes unaddressed, Mollick said, “You don’t want a hundred times more PowerPoint.”
That gap lines up with a central question about AI spending: do gains at the desk turn into gains on the income statement? Mollick kept his comments on company operations. His argument is about where value gets created inside a company, and that is the layer investors eventually see in results. His Wharton faculty profile lists him as the Ralph J. Roberts Distinguished Faculty Scholar, Rowan Fellow, and Associate Professor of Management at the school.
AI Models Are Improving Faster Than Corporate Playbooks
Part of the challenge is speed. Mollick described how much the tools have changed since his previous book: “When I wrote my last book, ChatGPT had just come out, and the model powering it, which was called GPT-4, was the best model in the world for a year. It’s now the, would be the 181st best model. You have models on your phone that run better than that.”
By Mollick’s account, a model that once led the field would now sit at 181st. When the underlying tool improves that quickly, a workflow built around one model’s strengths and weaknesses can go outdated before it finishes rolling out. Training programs, review steps and controls written for last year’s capabilities may already fall short of what the current tools can do. A recent Fortune commentary described a related tension as “the paradox at the core of the AI revolution: the faster we innovate, the slower we grow.” Venture investor Michael Fertik made a related point about the model race this month, saying many companies could still overtake Anthropic.
Why More Output Can Mean Less Value
The simplest use of AI is to help each employee produce more drafts, decks, emails and reports. Scale that across thousands of workers and the company ends up with more volume to read, review and act on. Each person looks more productive, and the organization can still be stuck, buried in material nobody asked for.
Mollick’s prescription starts with purpose: “Company leaders need to rethink about what they’re trying to accomplish at work, not just more of everything, but what is the actual goal here. And that requires redesigning processes.”
In practice, redesigning a process means starting from the result a team is supposed to deliver and working backward. Leaders map each step, decide which steps can be eliminated or combined, identify where human judgment still matters, and assign clear ownership for reviewing AI-assisted work. It also means changing how success gets measured, so teams are rewarded for outcomes such as faster decisions or fewer errors. Activity counts become a secondary metric.
Hopeful on Drudgery, Worried About Learning
Mollick frames the human impact as a matter of deployment choices. “I am incredibly hopeful about taking away drudgery in jobs. I am worried about the idea that we automate away too much work and too many jobs. I think that it can undermine how people learn, but if you use AI as a tutor, we have a lot of evidence that it makes people learn better.”
He also argued that technical capability alone will not settle where people fit: “There’s, people will still want to read other people’s writing. There are still places for humans even if AI is technically better.”
The same tool can improve skills or erode them, and it can lift a team or strip out the entry-level work where people learn the job. Which of those happens depends on how leaders set it up.
Inside the GPT-4 experiment Mollick co-wrote with BCG consultants
What to Watch in Corporate AI Results
For anyone judging whether a company’s AI investment is working, Mollick’s framework points to evidence that AI has changed how the work gets done: redesigned workflows, fewer transfers, shorter cycle times, and leaders who can explain what goal the technology serves. Claims built only on usage counts or output volume are weak signals. How companies pay for AI is shifting too: Reflection’s CEO has said enterprises are moving from renting AI to owning it as it becomes a major cost.
By Mollick’s framework, the companies that turn individual gains into company gains are likely to be the ones whose leaders do the harder work of rebuilding processes around the technology.
Mollick says the evidence that AI makes individual workers better is overwhelming, and the open question is whether companies redesign their work to turn those gains into company results. The risks he flags are volume without a goal and automating away the entry-level work where people learn their jobs.