The Eighteen-Month Office Apocalypse, Again

Microsoft AI chief Mustafa Suleyman told the Financial Times this month that he expects "human-level performance on most, if not all, professional tasks" inside the next twelve to eighteen months. Most jobs that involve "sitting down at a computer", he said, will be fully automated by AI in that window. Accounting, legal work, marketing and project management got specific mentions. The clip has been doing the rounds ever since.

The number is concrete enough to grab a headline. It is also a claim worth taking apart carefully before letting it set policy or career decisions, because the pattern it fits is older than the current wave of AI and the track record is poor.

What was actually said

The interview, summarised by Tom's Hardware and Windows Central, includes Suleyman saying that organisations "will be able to retrofit the technology to perform any required job function" and comparing the future of model-building to "creating a podcast or writing a blog." The phrasing is striking. It is also the strongest possible version of a position that none of the major labs has actually shipped against.

Suleyman is in a particular spot. He runs an AI division at the company most exposed to enterprise adoption of generative tools. Microsoft has spent the last three years selling Copilot as a productivity layer rather than a replacement, and the gap between the marketing copy and the eighteen-month claim is wide. Forecasts of this scale are usually made by people whose financial position rests on the forecast being true.

The predictions that did not land

The pattern is familiar to anyone who has followed AI predictions since the 2010s. Geoffrey Hinton's 2016 claim that radiologists would be obsolete within five years has aged so badly that we covered the radiology version of this story last week. The diagnostic profession grew over the period he gave it. The 2018 wave of "truck drivers are doomed by 2025" pieces gave way to a 2025 trucking labour shortage. Even the more careful Oxford study from 2013, which predicted that 47 percent of US jobs were at risk of automation, has held up poorly when its categories are checked against what actually happened.

The McKinsey Global Institute, which has been tracking this for a decade, currently estimates that the share of US work activities (not jobs) that could be automated with current generative AI has risen meaningfully but remains a long way short of full task coverage in any of the professions Suleyman named. Harvard Business Review's March 2026 piece on AI in knowledge work found that adoption tends to reshape tasks within roles rather than collapse the roles themselves.

Why the 18-month frame keeps appearing

The specific number has a long history. It is the same window that has been used for cryptocurrency adoption, self-driving cars, voice interfaces, the metaverse, and chatbots in customer service. The reason is partly cognitive. Eighteen months is far enough away to feel like a real prediction, close enough to feel urgent, and short enough that almost nobody will check back when the date arrives.

The genuine question is not whether AI will displace some categories of white-collar work. It already is, particularly in entry-level paralegal, marketing copy and routine accounting work. The question is how complete and how fast that displacement runs. Current evidence points to gradual restructuring, with new task mixes, hybrid human-AI workflows, and a slow rebalancing of which skills are valuable. None of that fits a clean eighteen-month curve.

What to actually watch

If Suleyman's prediction is right, it should show up first in employment data for the professions he named. Bureau of Labor Statistics figures for accounting, legal services and marketing roles will start moving inside the next two to three quarters. Productivity per worker in those fields should jump sharply. Job postings should fall. Entry-level hiring should crater. Some of those signals are present in 2026 data, but at nothing like the rate the 18-month frame would require.

The reasonable bet is that the next eighteen months will look a lot like the last eighteen: real but incremental gains, with the deepest effects on tasks rather than whole jobs. The honest answer to anyone asking whether their profession is about to disappear is that nobody knows, including the executives confidently giving timelines. The track record of AI labour predictions is bad enough that anyone offering a precise number should be asked to put it in writing and revisit it in May 2027.