The Short Version
Two years ago, Altman said AI would eliminate 300 million white-collar jobs. In May 2026, he said he was glad he was wrong. Enterprises burned billions only to find AI more expensive than humans, and developers weren't eliminated — instead, they became the gatekeepers of AI output quality.
---
Timeline: From "Doom" to "Walk-Back"
| Date | Event | |------|------| | 2023-03 | Goldman Sachs report: AI could eliminate 300 million jobs | | 2024-06 | Altman warns entry-level customer service jobs will "completely disappear" | | 2025-06 | Amodei predicts AI will eliminate 50% of white-collar work | | 2026-03 | Altman predicts 30-40% of jobs will be replaced | | 2026-05-26 | Altman reverses course at Commonwealth Bank of Australia event: "pretty wrong" | | 2026-05-26 | Amodei pivots to an "augmentation" narrative: after 90% automation, the remaining 10% expands 100-fold | | 2026-05-29 | OpenAI files for IPO, targeting a $1 trillion valuation |
---
Altman's Own Words: "Glad I Was Wrong"
> "I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened." > — Sam Altman, 2026-05-26, Commonwealth Bank of Australia
He also shared a telling detail: he tried handing over all his Slack and email to AI, but quickly switched back to manual — "people want to hear from a real person."
The irony is sharp: the person building AI doesn't want to be replied to by AI.
---
Enterprise ROI Flops: AI More Expensive Than People
| Company | Investment / Result | Data | |------|-----------|------| | Microsoft | Canceled most Claude Code licenses | Costs unsustainable | | Uber | $3.4B AI budget burned in 4 months | $500–$2,000 per engineer per month | | Goldman Sachs | Actual US monthly job losses ~11,000 on average | Far below predictions | | GM | AI spending hard to tie to consumer-facing features | Uber's COO directly questioned its justification |
Core finding: AI tools are often more expensive than the humans they were meant to replace.
---
Linear CEO's Observation: No 10-100x Revenue Growth
> "Apart from the AI model labs themselves, no industry-wide revenue growth corresponding to 10-100x has appeared." > — Linear CEO
His own experience: AI is very effective for early-stage solution exploration, but actual implementation still requires extensive human judgment. Developers aren't being eliminated — their work模式 has changed from "writing from scratch" to "making decisions."
---
The IPO Logic Behind the Narrative Shift
The most important context for this reversal:
| Company | IPO Move | Valuation Target | |------|----------|----------| | OpenAI | Filed confidential IPO in May | $1 trillion | | Anthropic | Filed confidential IPO in May | $1 trillion |
The logic: the "AI destroys jobs" narrative is bad for an IPO. "AI augments jobs" is much easier for institutional and retail investors to buy. Altman's and Amodei's reversals aren't just cognitive updates — they're trillion-dollar narrative management.
---
Three Deeper Signals
1. The "Human Part" Is Bigger Than Expected
Altman's Slack experiment shows: even in purely text-based interactions, users care whether "there's a real person on the other side." This is especially true in customer service, sales, and management communication.2. Cost Structures Don't Match
AI inference costs are priced per token, while human employees draw a fixed salary. For high-frequency, repetitive, short-conversation tasks, AI can be cheaper; for complex, long-thread, context-heavy tasks, AI is actually more expensive.3. Developer Roles Transform, Not Disappear
From "writing code" to "making architectural decisions." AI can generate code, but choosing which approach to use, judging business priorities, and coordinating team communication — these matter more than ever.---
Takeaways for Developers
1. Don't panic: What gets eliminated is "developers who can't use AI," not developers themselves 2. Judgment is worth more than code volume: In the AI era, architecture decisions, business understanding, and cross-team communication rise in value 3. AI is a tool, not a replacement: Early exploration gets more efficient, but implementation still needs human oversight
---
One-Sentence Summary
> AI didn't eliminate developers — it forced developers to evolve from "craftsmen" into "decision-makers."
---
Sources: The PrimeTime "Maybe we were wrong" (2026-05-29), CryptoBriefing, Fortune, explainx.ai, Briefs Finance