AI Will Not Replace All Jobs: OpenAI & Anthropic Shift

Lisa Ernst · 31.05.2026 · AI News · 8 min read

For months, the loudest AI story was simple: advanced models would replace huge parts of the workforce. Now the message from two of the most important AI labs is becoming more careful. OpenAI and Anthropic are still warning that work will change fast, but the newer argument is not “all jobs disappear.” It is: many tasks get automated, many roles get redesigned, and some entry-level paths become much harder.

Source: Composite by Zerlo using Wikimedia Commons images by Steve Jurvetson, TechCrunch and Shixart1985

This matters because panic is a bad strategy. If AI changes 30 percent of your tasks, that does not automatically mean your job disappears. It may mean your job becomes more productive, more competitive, more supervised by software, or more dependent on skills that AI cannot easily own: judgment, trust, communication, accountability and taste.

The new AI jobs story: less apocalypse, more transition

OpenAI CEO Sam Altman recently said he does not expect the kind of “jobs apocalypse” that some people in the AI industry predicted. He also admitted that he expected more entry-level white-collar jobs to be eliminated by now than has actually happened. That is a meaningful shift in tone because OpenAI had previously warned clearly that some jobs would disappear while others would evolve.

Anthropic has also moved toward a more data-driven framing. Instead of only asking what AI could theoretically do, Anthropic’s research now looks at observed exposure: which work tasks are actually seeing automated or AI-assisted use in professional settings. That distinction is important. A model may be capable of helping with a task, but real companies still need processes, trust, quality checks, customers, regulations and human responsibility.

Source: Image sources: Wikimedia Commons, OpenAI and Anthropic public-domain text logos

OpenAI and Anthropic are not saying AI has no labor-market impact. The shift is more precise: the impact is uneven, task-based and heavily dependent on how companies actually deploy AI.

Why the prediction changed

The original fear was easy to understand. Modern AI can write code, summarize documents, draft emails, analyze data, answer support questions and generate marketing material. From a distance, that looks like direct job replacement. Inside real organizations, however, work is messier.

A job is rarely one clean task. It is a bundle of small tasks, meetings, decisions, exceptions, relationships, quality checks and accountability. AI can compress parts of that bundle, but it does not automatically own the full role.

❝ The key mistake is treating “AI can do this task” as the same thing as “AI can replace this job.” Real work contains context, responsibility and trust. ❞

Zerlo editorial analysis

OpenAI’s more careful framing

OpenAI’s Workforce Blueprint already used a balanced sentence: AI will reshape work, new jobs will be created, others will evolve, and some will disappear. That is different from saying every role gets wiped out. Its newer jobs-transition work also tries to map near-term labor pressure instead of relying only on generic “AI exposure” scores.

In plain English: OpenAI appears to be moving from a broad warning to a more practical question: where does AI reduce the cost of doing work, where must humans remain involved, and where could cheaper production actually increase demand?

Source: Steve Jurvetson via Wikimedia Commons, CC BY 2.0

Sam Altman’s newer comments are important because they separate technical capability from social reality. People still value direct human interaction, accountability and trust in many work settings.

Anthropic’s data tells a similar but still risky story

Anthropic has been one of the strongest voices warning about white-collar disruption, especially for entry-level roles. But its own Economic Index research adds nuance. The company measures how Claude is actually being used, not only what Claude could theoretically do. In one labor-market analysis, Anthropic says AI use is still far from reaching theoretical capability in some categories; for example, it reported Claude covering only part of all tasks in computer and mathematical occupations.

That does not make the risk harmless. It simply means the risk is not evenly distributed. Entry-level work that consists mainly of routine writing, basic analysis, simple coding, ticket triage, formatting and document processing is more exposed than work involving customer trust, leadership, physical presence, negotiation, accountability or deep domain judgment.

Source: TechCrunch via Wikimedia Commons, CC BY 2.0

Dario Amodei has warned strongly about AI disruption. The more useful reading is not “ignore the warning,” but “prepare for task compression, especially in junior white-collar work.”

The practical difference: task replacement vs. job replacement

The best way to understand the shift is to separate tasks from jobs.

Why companies may still hire people

There is a second reason the “replace everything” narrative is too simple: when a task becomes cheaper, demand can expand. If building software becomes faster, companies may build more software. If analysis becomes cheaper, managers may ask for more analysis. If content production becomes cheaper, teams may publish more variations, more tests and more localized versions.

That does not guarantee job growth in every field. It means the outcome depends on demand. If demand expands faster than automation removes work, employment can hold up or even grow. If demand is flat and the task is easy to automate, headcount pressure becomes much stronger.

Source: Shixart1985 via Wikimedia Commons, CC BY 2.0

The most likely near-term outcome is not a clean replacement wave. It is a redesign of office work: fewer repetitive steps, more supervision of AI output and more pressure to deliver results faster.

Who is most at risk?

The highest risk is not “everyone who uses a computer.” It is people whose value is mostly routine output without much ownership. That includes parts of junior research, simple copywriting, basic coding, first-level support, template-based reporting, standard document review and repetitive administrative work.

The safer side is not “never use AI.” It is the opposite. Safer workers usually become the people who can use AI well while adding judgment around it. They know what to ask, what to reject, what to verify and how to turn output into a real business result.

What workers should do now

1. Map your tasks. Write down what you do every week. Mark which parts are repetitive, text-based or rule-based.
2. Automate your own low-value work first. Use AI to draft, summarize, compare, structure and check simple work before someone else does it for you.
3. Build verification skills. The valuable worker is not the person who blindly accepts AI output. It is the person who can detect errors quickly.
4. Move closer to responsibility. Client trust, project ownership, compliance, architecture, people management and domain judgment are harder to automate fully.
5. Learn workflows, not just prompts. The advantage comes from combining AI with documents, data, tools, review steps and clear output standards.

Source: Shixart1985 via Wikimedia Commons, CC BY 2.0

AI makes first drafts cheaper. That increases the value of people who can review, decide, coordinate and take responsibility for the final result.

What companies should do

For companies, the wrong strategy is replacing people first and understanding the process later. A better strategy is to measure where AI actually saves time, where errors are expensive and where human review is required. AI should be introduced with clear ownership: who checks the output, who approves it and who is accountable when something goes wrong.

Teams should also protect their junior talent pipeline. If AI removes every beginner task, companies may save money now but lose future experts. The better model is supervised acceleration: junior employees use AI, but they still learn the underlying work, receive feedback and build judgment.

What this means for AI news readers

The headline “AI will replace all jobs” gets clicks, but it is too crude. The better headline is more useful: AI will attack tasks before it attacks entire roles. The first winners will be people and companies that redesign workflows around AI without removing human judgment from places where trust, safety and responsibility matter.

For more practical AI tools and workflow ideas, explore the Zerlo tools section.

FAQ