Webinar: Where Is the AI Payoff? Erik Brynjolfsson on Why Returns Lag, and Who’s Capturing Them

Erik Brynjolfsson on why AI returns lag, why a few users capture most of the value, and how leaders can close the gap.

November 20, 2026

It’s hard to miss how fast AI is getting better. Technical benchmarks are being matched or beaten faster every year, and on many of them machines now match or exceed human performance. The business results haven’t kept up. A recent analysis by Apollo’s Torsten Slok found basically no increase in revenue per worker across the 493 Fortune 500 companies outside the “Magnificent 7.” U.S. productivity growth has been slow for more than a decade.

In our November webinar, Workhelix co-founder Erik Brynjolfsson, director of the Stanford Digital Economy Lab, joined Workhelix co-founder and CEO James Milin to discuss why the payoff is hard to see, who is already capturing it, and what leaders can do about it.

Watch the full recording here, and read the key takeaways below.

‍

‍

The technology is the tip of the iceberg

“Technology, in some ways, is the easy part,” Erik said. “It’s just the tip of the iceberg.” Most of the value comes from business transformation: changing how people work and how processes run. Those intangible investments cost more than the software, and they take time.

That’s why powerful general-purpose technologies follow what Erik calls the productivity J-curve. Companies spend more while they figure things out, so productivity stays flat or dips before the payoff arrives. With electricity, the dip lasted decades. AI is moving much faster. ChatGPT reached 100 million users in under 90 days, and people can work with it in plain English. But the J-curve still applies.

The Enterprise AI Playbook, which Erik co-authored with Elisa Pereira and Alvin Graylin, studied 51 successful deployments. Its conclusion: “The bottleneck was the organization, not the model.” Even the successes struggled along the way. Only 6% started with clean data. As Erik said, “Unfortunately, there’s no shortcut.”

‍

A few people capture most of the value

Workhelix data shows the same pattern almost everywhere. When employees are ranked by how much they use the company’s AI platform, the top 10% generate about 68% of all prompts, and 23% never use it at all.

You can’t spot these super users by job title, seniority or function. Two people with the same role on the same floor can get very different results. The group also keeps changing: only 28% of people in the top decile one year are still there the next, which is why continuous measurement matters more than a one-time audit.

Meanwhile, the typical user is getting a small fraction of what’s possible. “It’s almost painful to me to see that there’s this immense potential of AI, and yet the typical user is only getting just a small fraction, 2 or 3% of the value,” Erik said.

The fix is often already inside the company. Pointing employees to colleagues who are succeeding works better than outside advice. “It’s not us coming from the ivory tower and saying, hey, you could be doing this,” Erik said. “It’s literally somebody in their same organization.”

‍

Same tools, a widening gap

Almost anyone can use the same frontier models, often for free or at an accessible cost. You might expect that to level the field. “In fact, the opposite has happened,” Erik said. The top 10% of firms use AI 7 to 10 times more than the rest, and they’re pulling away. Some sectors, like finance, are moving faster than others. But the biggest differences are between firms in the same industry, and between people in the same firm. The leading firms aren’t using better technology, they’ve just moved along the J-curve faster.

‍

Don’t default to headcount cuts

When AI makes workers more productive, many leaders jump straight to cutting headcount. Sometimes that’s right. Erik argued that it often comes too fast, and that the answer depends on demand. Farming needs fewer workers because people don’t eat much more when food gets cheaper. Cheaper flights, on the other hand, brought more pilots, not fewer.

“I see way too many companies focus on that headcount reduction side of the ledger,” Erik said, “when I think in the long run, they’ll often be better off looking at how they can expand the pie.” The message leaders send matters too. When executives present AI as a way to replace people, middle managers resist. “If you tell them that these tools are meant to replace them or their colleagues around them, that’s not as good a sales pitch.”

‍

Early labor market signals

Erik also shared the latest data from Stanford’s research with ADP. Overall employment is holding up, and there’s no sign of mass unemployment. Underneath, though, young workers in the most AI-exposed occupations, such as call center staff and software engineers, have seen a relative employment decline of about 19% compared with less-exposed roles, compared to 13–14% in mid-2025.

At the same time, new business formation in the U.S. is roughly double its 2019 level. Erik’s view is that demand is moving away from following instructions and toward initiative: “To exaggerate a little bit, intelligence is becoming a commodity.”

‍

A design problem, not a prediction problem

Leaders often ask Erik whether AI will deliver a big productivity payoff. He changes the question: “It’s not a prediction problem, it’s a design problem.” That means measuring at three levels, not just one: where the opportunity is (exposure), what people are actually doing (usage), and what it delivers (outcomes). Then it means acting on what the data shows.

‍

Watch the full recording to hear the complete discussion, including audience Q&A on IT governance, aligning power users with company goals, and how fast AI is spreading compared with past technologies. Scan the QR code in the recording to join the waitlist for the Workhelix AI coach.

‍

Subscribe to the newsletter

Know and Grow the ROI of Your AI®

Get a demo
Close

Contact us for a demo

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.