← Back to feed

Sam Altman told a tech conference in Sydney on May 26 that the AI-driven white-collar job apocalypse he once warned about has not materialised — and he is happy to be wrong.

“I’m delighted to be wrong,” Altman said, according to Reuters. He had previously argued that routine office work, particularly entry-level tasks, would be hit hard as AI models scaled. His revised position is that the transition is slower and softer than expected: work is adapting around the technology rather than being eliminated by it.

Why the Reversal

Altman’s updated argument rests on the durability of human-specific tasks. Companies still need people for:

• Judgment calls where context and consequences matter
• Trust — clients, regulators, and stakeholders want human accountability
• Taste — creative and strategic decisions where “good enough” has legal and reputational dimensions
• Emotional reading — de-escalation, negotiation, care work, and all the communication where tonality determines outcomes
• Messy communication — situations where the right answer depends on unstated context that AI models cannot reliably infer

The argument is not that AI is less capable than predicted. It is that the tasks humans do are less substitutable than they appeared from the outside.

The Broader Picture

Altman’s walk-back lands against a backdrop of mixed signals. Multiple companies have cited AI as the reason for layoffs — Cloudflare, Coinbase, PayPal — while others have expanded headcount even as they deploy AI tools. The Stanford AI Index 2026 reported that AI is the number-one stated reason for US job cuts two months running, while hiring plans are down 69%.

The reconciliation is that job losses are real but concentrated in specific functions. Mass white-collar unemployment has not followed. What has happened is slower churn in the categories most exposed to structured repetitive work: data entry, basic coding tasks, form processing, first-draft writing.

What has not happened: the displacement of judgment-intensive roles at any meaningful scale. That is the gap Altman is pointing at.

What It Means

Altman’s revised view is not reassurance — it is a timeline correction. The structural forces that make AI a substitute for routine cognitive work have not reversed. They have simply taken longer to propagate through organisations that still need humans for accountability, relationships, and the irreducibly contextual parts of the job.

OpenAI is simultaneously pushing Codex, Workspace Agents, and agentic coding tools designed to automate exactly the tasks Altman says humans still hold. The argument that work is “bending before it breaks” implies a break is still coming — just not on the timescale he initially feared.

His confidence interval has widened. The direction of travel has not changed.