Sam Altman and Dario Amodei Walk Back AI Jobs Apocalypse as Reality Sets In

On May 26, 2026, in a conference room in Sydney, Australia, OpenAI CEO Sam Altman did something rare in Silicon Valley: he admitted he was wrong.

"I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened," Altman told Commonwealth Bank CEO Matt Comyn. "I'm delighted to be wrong about this."

The same week, Anthropic CEO Dario Amodei—who once warned AI could eliminate 50% of white-collar jobs—began emphasizing a very different narrative: AI as augmentation, not replacement.

The timing is striking. Both companies reportedly filed confidential IPO paperwork in May 2026. Both face mounting scrutiny over AI's real-world economic impact. And both are watching their enterprise customers—Microsoft, Uber, Goldman Sachs—slam the brakes on runaway AI spending after discovering a brutal truth:

AI tools are often more expensive than the humans they were supposed to replace.

TL;DR

The Predictions They Made

Let's be clear about what was actually said.

Sam Altman (June 2025)

Speaking at a Y Combinator event, Altman warned that entry-level white-collar roles were at serious risk and that the pace of AI-driven job displacement would accelerate dramatically through 2026.

Dario Amodei (January 2026)

The Anthropic CEO was even more explicit, warning that AI would cause "unusually painful" disruption and predicting that AI systems would soon replace large numbers of entry-level workers in finance, consulting, and tech. He stated: "I would not be surprised if somewhere between one and five years, we started to see big effects here."

Amodei went further, describing AI as a "general labor substitute for humans"—not just automating specific tasks, but entire job categories.

Goldman Sachs (2023-2025)

Goldman's landmark research predicted generative AI could automate tasks equivalent to 300 million full-time jobs worldwide. By early 2026, the firm estimated AI was eliminating 16,000 net U.S. jobs per month.

The Problem

None of this matched what was actually happening in the labor market.

What Actually Happened

The Yale Budget Lab delivered the most damning evidence: no significant changes in occupational mix or unemployment duration in high-AI-exposure jobs since ChatGPT launched in November 2022.

Even Goldman Sachs had to revise downward. Their latest AI Adoption Tracker (May 2026) puts net job losses at approximately 11,000 jobs per month—30% below their earlier estimate and a fraction of the apocalyptic projections.

The Augmentation Reality

What emerged instead was a different pattern:

• Productivity gains without headcount reduction
• Task automation rather than role elimination
• Skill shifts within existing jobs
• New bottlenecks replacing old ones

As Peter Diamandis noted, "Real-world deployments showed augmentation, not replacement. Humans plus AI is the winning formula."

The Cost Crisis Nobody Predicted

Here's the twist that makes this story fascinating: while AI didn't replace as many jobs as predicted, it became far more expensive than anticipated.

Microsoft's Claude Code Ban

In May 2026, Microsoft began canceling most internal Claude Code licenses across its Experiences and Devices division (the teams behind Windows, Microsoft 365, Outlook, Teams, and Surface). The deadline: June 30, 2026.

The reason? Token bills spiraling out of control.

When engineers use an AI agent for complex, multi-hour coding tasks, token costs accumulate fast. The token-based pricing model—charging per output—meant that agentic workflows (where AI runs multi-step tasks autonomously) multiplied usage 5-30x faster than forecasted.

Engineers are now being directed to GitHub Copilot CLI, Microsoft's own (cheaper, less capable) tool.

Uber's $3.4 Billion Budget Burn

The situation at Uber tells an even more dramatic story.

Timeline:

• Uber deployed Claude Code to 5,000 engineers
• Monthly usage rates climbed to 84-95% by April 2026
• Per-engineer API costs: $500-$2,000 per month
• Result: Uber burned through its entire $3.4 billion 2026 AI budget in four months

Uber COO Glen Fogel is now publicly questioning whether the investment was worth it.

The Broader Pattern

A 2025 Mavvrik survey found:

• 85% of companies missed AI cost forecasts by over 10%
• 84% saw gross margins drop by more than six percentage points due to AI spending

Bryan Catanzaro, VP of Applied Deep Learning at Nvidia, told Axios: "For my team, the cost of compute is far beyond the costs of the employees."

An MIT analysis (2024) suggests that on current pricing, AI automation pencils out as cheaper than human labor for roughly a quarter of the jobs people thought it would replace.

Why the Reversal Now?

The timing raises questions.

IPO Timing

Both OpenAI and Anthropic reportedly filed confidential IPO paperwork in May 2026. A calmer jobs narrative—"augmentation, not apocalypse"—is far more palatable for public markets, institutional investors, and regulatory scrutiny.

Enterprise Pushback

As Microsoft, Uber, and other enterprise customers hit spending ceilings, AI companies face pressure to justify ROI. The "10x productivity" narrative only works if customers aren't churning due to cost.

Political and Regulatory Pressure

With the 2026 U.S. midterm elections approaching, tech leaders face mounting political pressure over job displacement claims. Walking back doom predictions reduces the likelihood of aggressive AI regulation.

Data Reality

The Yale and Goldman data made it increasingly difficult to maintain the apocalypse narrative without looking dishonest. Better to pivot gracefully than be proven wrong by third-party research.

What the Critics Are Saying

The reversal sparked immediate backlash on X (formerly Twitter).

Nick Huber (@sweatystartup), a vocal AI skeptic, didn't hold back:

"I would be way more bullish on AI if it actually worked and was actually replacing real humans at scale. Nothing is changing and we're being lie[d to]. The tools don't work! They are expensive! And high maintenance. Dot com bubble 2.0."

Huber's thread, which went viral with 92 reposts, captures a growing sentiment among founders and operators: AI hype has outpaced AI reality.

Mike Isaac (New York Times) called the Varick Agents story—$600/day on Claude API vs. $20/day meal limits—a "perfect encapsulation of the AI Capex era." (See our related coverage on AI-native startup economics.)

Others pointed to data center cancellations, Microsoft's Claude ban, and slowing enterprise adoption as evidence the bubble is popping.

The Goldman Sachs Nuance

Goldman Sachs CEO David Solomon added a crucial perspective in a New York Times guest essay (May 22, 2026):

"Predictions that technology will reduce the need for human labor have a long history but a poor track record."

Solomon argues that fears of mass unemployment from AI are overblown, pointing to historical parallels with automation waves that ultimately expanded employment.

Goldman's research now projects AI's impact on overall employment will be mild and short-lived, with the unemployment rate potentially rising by about 0.5% during the transition—not the catastrophic double-digit spikes some predicted.

The Augmentation vs. Replacement Framework

What's emerging is a clearer picture of where AI actually works:

AI Works Best For:

• Routine, predictable tasks (data entry, basic code generation)
• High-volume content (documentation, summaries, reports)
• Rapid prototyping (design iterations, testing)
• Research and synthesis (crawling sources, cross-referencing)

AI Struggles With:

• High-stakes decisions (legal judgments, medical diagnoses)
• Complex negotiations (sales, conflict resolution)
• Novel problem-solving (architecture, strategy)
• Social intelligence (management, customer empathy)

The key distinction: disruption does not mean replacement. Jobs are changing, but not disappearing—at least not yet, and not at the predicted scale.

What This Means for Developers and Startups

If you're building with AI agents, this reality check offers clarity:

1. Cost Management Is Critical

The AI token cost surge is real. If you're deploying agentic workflows at scale, budget 5-30x your initial estimate and implement hard usage caps.

2. Augmentation Over Replacement

Design AI tools to enhance human capabilities, not replace them entirely. The most successful deployments keep humans in the loop for judgment calls.

3. ROI Proof Required

Enterprise buyers are getting savvier. You need to show measurable productivity gains that justify the token spend—and ideally, cost less than the human alternative.

4. Skill Building Remains Valuable

For developers, the takeaway is clear: your job isn't going away, but it's changing. As Gary Tan demonstrated (see our coverage of his 400x productivity gain), developers who master AI-assisted workflows can achieve extraordinary output.

5. The Bubble May Be Real

Nick Huber's warning about "AI hype collapse" aligns with growing evidence. The companies that survive will be those solving real problems with sustainable economics, not just riding the hype cycle.

Anthropic and OpenAI: Different Approaches, Same Destination

While both CEOs are walking back job apocalypse predictions, their approaches differ:

OpenAI's Strategy

• Focus on GPT-4o and GPT-5 as "reasoning engines"
• Emphasis on ChatGPT as a consumer product
• OpenAI Sites for easy website building
• Positioning: "AI for everyone"

Anthropic's Strategy

• Claude 4.5 and Claude Code for technical users
• Emphasis on safety, interpretability, and responsible scaling
• Enterprise-first with financial services plugins and managed agents
• Positioning: "AI you can trust"

Both are now selling augmentation, not revolution.

The Broader AI Reality Check

This story is part of a larger pattern:

Recent AI Backlash Indicators:

• Microsoft banning Claude Code due to costs
• Uber burning through AI budget
• arXiv imposing one-year bans for AI-generated errors
• Big Tech copying startups' beta testing strategies
• Agentic fatigue and the productivity paradox

The AI bubble reality check is happening in real time.

So Who Was Right?

Peter Diamandis: Right that augmentation is the winning formula.

Yale Budget Lab: Right that labor market disruption has been minimal so far.

Goldman Sachs (revised): Closer to reality with 11,000 jobs/month estimate.

Nick Huber: Right that current AI tools are expensive and high-maintenance.

Sam Altman and Dario Amodei: Right that AI is transformative, but wrong about the timeline and nature of job displacement.

The Path Forward

For developers, the message is surprisingly optimistic:

Your job isn't disappearing—it's evolving. The developers thriving in 2026 are those who:

1. Master AI-assisted workflows (Claude Code, Copilot, Cursor)
2. Focus on high-value tasks (architecture, problem-solving, judgment)
3. Build agent skills and custom tooling
4. Understand cost management (token budgets, caching strategies)
5. Keep humans in the loop (review, refinement, strategic decisions)

As Andrej Karpathy's guidance emphasizes, the future is human-AI collaboration, not human replacement.

Conclusion: The Hype Cycle Corrects Itself

Sam Altman's admission—"I'm delighted to be wrong"—is refreshingly honest. But it also reveals how far ahead of reality the AI narrative had run.

The truth emerging in mid-2026:

• AI is transformative, but not in the ways initially predicted
• Costs are higher than anticipated (often exceeding human labor)
• Augmentation works; full replacement mostly doesn't
• Job displacement is real but gradual (~11K/month, not millions)
• The hype cycle is correcting, separating signal from noise

For those building with AI, this is good news. The market is maturing. Unrealistic expectations are being replaced by pragmatic deployment. The survivors will be those solving real problems with sustainable economics.

And for workers? The apocalypse is canceled—for now.

But the transformation is just beginning.

Related Reading

• Gary Tan's 400x Productivity with Claude Code — How AI augmentation actually works
• AI Token Costs Surge: Enterprise Reality Check — Managing the new AI Capex era
• AI Native Economics: Varick Agents Case Study — The $600/day API spend model
• Agentic Fatigue and the Productivity Paradox — Why AI tools exhaust users
• AI Bubble 2026 Reality Check — Signs the hype is cooling
• Anthropic Claude Code Pro Pricing — Understanding enterprise costs
• Agent Skills Security — Building safe AI workflows

Sources

• Sam Altman and Dario Amodei walking back AI jobs apocalypse predictions | Fortune
• Sam Altman Says AI 'Jobs Apocalypse' Probably Won't Happen | TIME
• No AI 'jobs apocalypse' so far, says OpenAI's Sam Altman | Euronews
• Anthropic CEO Dario Amodei warns AI may cause 'unusually painful' disruption | CNBC
• Microsoft reports AI cost problem: more expensive than human employees | Fortune
• Microsoft Pulls Back Claude Code as AI Costs Reshape Big Tech | Memeburn
• Uber burned through its entire 2026 AI budget in four months | Fortune
• Goldman Sachs CEO says AI job apocalypse is overblown | Prism News
• How Will AI Affect the US Labor Market? | Goldman Sachs

AI predictions evolve rapidly. Revenue and spending figures are based on public statements and reporting through June 2026. For current AI pricing and enterprise guidance, visit Anthropic or OpenAI.