The Same Week, Five Voices, Five Positions

Between May 21 and May 26, the CEOs and leaders of the institutions most invested in AI went public with their forecasts on AI and employment. They did not reach the same conclusion. What's remarkable isn't that they disagree — that's normal. What's remarkable is that the people most informed about AI scaling speeds, cost curves, and deployment friction all landed on different answers to the same question: Does AI eliminate jobs or create them?

David Solomon, Goldman Sachs CEO, went first. In a May 22 New York Times op-ed, he argued that fears about AI-driven job losses are "overblown." Automation handles 25% of finance tasks, which means reallocation and upskilling, not elimination. New roles emerge. The historical pattern holds. By Solomon's view, your firm is going through the same transition banks went through with electronic trading.

Jamie Dimon, JPMorgan CEO, disagreed in the same news cycle. Speaking on May 21–22, Dimon said he thinks AI "will reduce our jobs down the road" — fewer bankers required, but more AI specialists hired to replace them. It's not no job loss; it's job transformation. Your people leave through attrition. Your hiring mix changes.

On the same day, Jensen Huang (Nvidia) pushed back on the entire framing. The "lazy" narrative is that AI replaces people. The real story: "You won't lose your job to AI — you'll lose it to someone who uses AI." It's not a technology problem. It's a competitive advantage problem. Use the tool or be replaced by someone who does.

Then the people who actually build the AI changed their minds.

The Reversals

In January 2026, Dario Amodei, CEO of Anthropic, issued a stark warning. AI would displace 10–20% of white-collar workers. The pain would be "unusually painful." The window between 2026 and 2031 would be the hardest transition. Amodei wasn't being rhetorical. He was modeling the impact and warning the world.

In May — four months later — Amodei reversed. He pivoted to the Jevons Paradox, an economics principle that says when you automate 90% of a job, you create enough new demand for the remaining 10% that the total workforce expands. People still do the work. The lever just got longer. Amodei's January apocalypse became May's demand-expansion story. Both analyses came from the same data. The January forecast was wrong.

Sam Altman went even further. On May 26 — today — Altman posted that he was "delighted to be wrong" about white-collar job displacement. He'd predicted faster and deeper impact. The reality has been slower. He even tested GPT on his own Slack and "switched back" — meaning the tool didn't improve his workflow enough to justify the cognitive overhead. The man who built the most widely-used AI just admitted it's less transformative than he forecasted.

What changed between January and May? Not the technology. The technology got better. What changed is the operating reality of deployment: AI is harder to integrate into workflows than the models predicted. ROI is less automatic than the cheerleading suggested. Dislocation happens slower than the math implied. The two people most alarmed and most informed have both stepped back from their apocalypse forecasts. That's not reassurance. That's signal.

The Operating Reality Check

Uber's COO added one more data point. AI costs are "hard to justify" inside operating companies. Token usage doesn't map to useful output. The automation is real, but the value extraction is unpredictable. This is the ground-level truth that the macro debaters don't address: your AI vendor can show you a 40% task automation rate and you still won't know if you should deploy it.