Tasks, Not Jobs — the Most Convenient Reframe in the History of AI

Mustafa Suleyman said white-collar work would be fully automated within 18 months. He now says he meant tasks, not jobs. The distinction is real. The timing of its discovery is not a coincidence.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/tasks-not-jobs-the-most-convenient?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

In February 2026, Mustafa Suleyman, CEO of Microsoft AI, told the Financial Times something unambiguous. White-collar work — sitting at a computer, being a lawyer, an accountant, a project manager, a marketing person — most of those tasks would be fully automated by AI within the next 12 to 18 months. The statement was widely reported. It was not hedged. It was not qualified with a distinction between tasks and roles. It landed with the force of a prediction from the person running AI strategy at one of the world’s most powerful technology companies.

By May 2026, Suleyman appeared on The Verge’s Decoder podcast with a clarification. He had been misunderstood. He was talking about specific tasks within roles — reading emails, processing information — not about the job of being a lawyer. He was not, he said, actually changing his view. What he was changing was the interpretation of his view that people were apparently drawing from the words he had used.

The tasks-versus-jobs distinction is real and important. It is also the most convenient clarification available to an executive whose February statement had alarmed millions of professionals, contributed to a climate of intense anxiety about AI and work, and arrived in an environment where AI companies are now preparing for the largest IPOs in the history of the technology industry.

CONTEXT AND BACKGROUND

Suleyman is not alone. The pattern in 2026 is consistent enough that Fortune has described it as coordinated. Sam Altman told an audience in New Delhi in February that AI superintelligence would soon outperform even CEOs, including himself, and that current jobs would be disrupted in ways that would be very hard to outwork. Dario Amodei of Anthropic predicted that up to half of all entry-level white-collar jobs would dissolve within five years and that unemployment could reach 10 to 20 per cent. Ford CEO Jim Farley said AI would cut in half the number of white-collar jobs in the United States.

By 26 May 2026 — the same week that OpenAI filed IPO paperwork confidentially and Anthropic signalled its October public listing — both Altman and Amodei had walked back those predictions in public. Altman told a Commonwealth Bank of Australia event in Sydney that he was “delighted to be wrong,” that he was “pretty wrong” about the social and economic impact of AI, and that he had expected more impact on entry-level white-collar jobs by now than had actually happened. Amodei reframed his earlier prediction not as job destruction but as output multiplication: if AI automates 90 per cent of a task, the remaining 10 per cent expands to fill the whole role and increases productivity tenfold. He is now citing the Jevons paradox — the nineteenth-century economic observation that more efficient energy use tends to increase total energy consumption rather than reduce it — as the framework for understanding AI’s employment impact.

Goldman Sachs CEO David Solomon, notably, did not need to walk anything back. He never made the apocalyptic prediction. In a New York Times opinion piece published the same week, he argued that a century of American economic history offers a clear rebuttal to AI job panic, noting that US civilian employment has grown 145 per cent since 1962 despite every technological disruption in that period, and that Goldman Sachs research shows data centre construction alone has added 200,000 jobs since 2022.

INSIGHT AND ANALYSIS

The tasks-versus-jobs distinction is analytically defensible and worth taking seriously on its merits. The Yale Budget Lab, which has been tracking AI’s effect on the labour market with the rigour of an academic institution rather than the incentive structure of a technology company, found in a May 2026 study that AI was not the primary reason for any weakening in the labour market, and that there had been no meaningful change in unemployment through March 2026 for workers in roles with high AI exposure. Economists Alex Imas and Tyler Cowen, and Apollo’s Torsten Slok, have all made versions of the Jevons argument — that automation does not decrease demand for a professional category but rather makes it cheaper and therefore more widely consumed, potentially increasing the total workforce in that category rather than reducing it. Call centre employees and radiologists, both with roles vulnerable to automation, have remained steady or increased in number despite wider AI adoption.

The problem is not that the tasks-versus-jobs distinction is wrong. The problem is that it is being offered now, by the people who made the original predictions, in the specific context of their IPO preparations, as a clarification of what they were always saying. That sequence is what demands scrutiny.

I have previously written about the long historical pattern in which those who control the gains of technological transformation consistently frame the cost of that transformation for the people who bear it in whatever language best serves their interests at the moment. The original apocalyptic predictions served specific interests: they attracted investment, signalled capability, positioned AI companies as the most consequential force in the global economy, and created the urgency that drives enterprise AI sales. A calmer narrative serves different interests: it reassures public market investors, regulators, and governments that the companies they are being asked to invest in or approve are not destabilising forces. Neither version of the story is primarily about accuracy. Both versions are primarily about positioning.

The data that sits between the two versions is more complicated and more honest than either. Tech layoffs through May 2026 had passed 115,000, already approaching the total for all of 2025, with Meta, Amazon, and Snap explicitly citing AI as a driver of cuts. More than 144,000 technology workers globally lost their jobs in 2026, according to layoff tracker TrueUp. At the same time, the Yale Budget Lab found no systemic shift in unemployment for high-AI-exposure workers. Both things are true simultaneously, and neither the February doom predictions nor the May reassurances acknowledge that coexistence adequately.

The Jevons paradox, invoked now by both Amodei and multiple economists, is a reasonable framework for thinking about AI’s long-run employment impact. It is also a framework that offers no guidance to the project manager who restructured her career in January based on Suleyman’s 18-month automation timeline, or the accounting graduate who chose a different field based on Amodei’s 50 per cent job loss prediction. The long run in which Jevons effects materialise is not the same timeframe in which those decisions were made and those lives were shaped.

IMPLICATIONS

For boards and executives, the executive language shift of May 2026 carries a specific and practical governance implication. The AI workforce strategy your organisation built in the first quarter of 2026 was likely calibrated, at least in part, on the kind of predictions that are now being walked back. If your planning assumed that a significant proportion of white-collar roles would be eliminated within five years, that assumption deserves re-examination in the light of what the actual labour market data shows. If it assumed that the technology would develop more slowly than the February predictions suggested, the walk-backs do not validate that assumption either — they simply reflect changed incentives, not new evidence.

The more durable insight is the one that neither version of the story provides: AI’s impact on employment is highly uneven, sector-specific, and dependent on a combination of technical capability, regulatory environment, organisational capability, and the specific task structure of the roles in question. The organisations that understand that specificity — that know which functions in their particular context are most exposed to task automation, and which require the kind of judgment, relationship, and accountability that neither today’s AI nor the Jevons paradox addresses — will build more resilient workforce strategies than those calibrating their planning to whatever the CEO of OpenAI said most recently.

The Palantir CEO, Alex Karp, offered a note of warning this week that sits uncomfortably alongside the walk-backs: tech bosses celebrating AI layoffs risk consuming the whole industry by generating the kind of public anger that produces regulatory and political responses far more disruptive than any voluntary adjustment. That warning is addressed to a different audience than Solomon’s op-ed or Altman’s Sydney remarks — and it points to the social and political consequences of the gap between what these executives said in February and what they are saying now.

CLOSING TAKEAWAY

The tasks-versus-jobs distinction will eventually be resolved by evidence rather than by executive clarification. The question is not whether AI will automate specific tasks within professional roles — it clearly will and is doing so now. The question is whether that task automation will produce, over time and across different economies and demographic groups, the kind of Jevons-paradox expansion in professional employment that makes the February predictions look wrong in retrospect, or the kind of structural displacement that makes the walk-backs look like a communication strategy rather than an update.

That question will be answered not in Sydney in May 2026, and not in New Delhi in February, but in the labour market data that accumulates over the next five to ten years. The people making career decisions, education choices, and workforce strategies in the meantime deserve a more honest account of what is known and what is not than the oscillation between apocalypse and reassurance that the technology industry’s IPO calendar has produced this quarter.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net