The Microsoft AI Boss Said 18 Months. We Are Two Months In.

In February 2026, Mustafa Suleyman, the CEO of Microsoft AI, sat down with the Financial Times and said something that should have stopped more people in their tracks.

AI, he said, would achieve human-level performance on most, if not all, professional tasks within 12 to 18 months. Not eventually. Not in theory. Within 18 months. He named the jobs: lawyers, accountants, project managers, marketing professionals. Anyone sitting at a computer, he said, doing work that can be defined and structured — that work is going to be automated.

It is now April 2026. We are two months into that window.

Suleyman was not alone. Anthropic CEO Dario Amodei had already said that AI could eliminate half of all entry-level white-collar jobs within five years. Ford CEO Jim Farley said it would “replace literally half of all white-collar workers.” Salesforce’s Marc Benioff claimed AI was already handling 50% of his company’s workload — and cut his customer support team from 9,000 to 5,000 people accordingly.

These are not fringe predictions. They are statements made by the CEOs of some of the largest companies in the world, in public, on the record. And yet the official employment data tells a strikingly different story.

In 2025, according to Challenger, Gray and Christmas — the outplacement firm that has tracked layoffs since 1989 — employers officially attributed 55,000 job losses to AI. In an economy of 125 million nonfarm jobs, that is a rounding error. A fraction of a percent. Barely visible.

But here is where the story gets complicated, and where the official numbers start to look less like data and more like a carefully maintained fiction.

Modelling-based estimates from multiple research groups place the actual number of jobs lost or never created due to AI in 2025 at somewhere between 200,000 and 300,000. Not 55,000. Up to 300,000. The gap is not a data error. It is, as one research group put it, “deliberate opacity.” Companies restructure headcount through attrition and hiring freezes rather than announcing mass layoffs that would make headlines.

A Duke University survey of 750 chief financial officers, conducted with the Federal Reserve Banks of Atlanta and Richmond, found that 44% of US businesses intend to make AI-related reductions in 2026 — equivalent to roughly 500,000 positions across the economy, half of them white-collar.

The question worth asking is why the official numbers are so far from the estimated ones — and what that gap reveals about how this transition is actually unfolding.

When a company lays off workers and cites “restructuring” or “strategic realignment,” those layoffs do not appear in the AI column. When a team of twelve becomes a team of eight because four people left and were not replaced, nothing appears anywhere. When a role that would have been created is quietly folded into an AI workflow instead, there is no data point. The job that never existed leaves no trace.

Goldman Sachs estimated a net loss of 16,000 jobs per month — 25,000 destroyed, 9,000 added back. The Q1 2026 tech sector figures showed 78,557 layoffs, nearly half attributed to AI. The Epoch AI and Ipsos survey found that one in five full-time workers says AI has already taken parts of their job. These numbers are consistent with each other. They are not consistent with 55,000.

So who is right — the CEOs making apocalyptic predictions, or the employment data suggesting everything is mostly fine?

Neither, exactly. But the CEOs are closer to describing something real.

The Harvard Business Review published an analysis this year arguing that companies are laying off workers based on AI’s potential rather than its current performance. That is a meaningful distinction. It means the displacement is running ahead of the productivity gains — people are losing jobs to AI before AI has fully proved it can do those jobs. The caution and the chaos are both real simultaneously.

What is clearly not right is the 55,000 figure. The mechanism by which AI displacement is happening — attrition, hiring freezes, roles quietly absorbed into workflows — is specifically the mechanism that official statistics are least equipped to capture. The data is not wrong because someone is lying. It is wrong because it was designed to measure something different from what is actually happening.

Back to Suleyman’s 18 months.

The prediction is almost certainly wrong as stated — not because AI is not advancing rapidly, but because institutional inertia, regulatory friction, and the sheer complexity of replacing human judgment in genuinely novel situations will slow full automation of professional work. Lawyers will not be automated in 18 months. Neither will doctors, teachers, or architects.

But the prediction is probably right in a more limited and more immediately important sense. Within 18 months, a significant portion of the routine tasks that currently justify many white-collar salaries will be handleable by AI systems at a fraction of the cost. Not all of the job. Enough of the job that the job changes shape — fewer people needed, lower salaries justified, entry-level roles unnecessary.

That is not the same as replacement. It is close enough to matter.

The most important thing Suleyman said was not the 18-month timeline. It was the list of jobs: lawyers, accountants, project managers, marketing professionals. He was not describing a future scenario. He was describing the current clients of Harvey, Spellbook, and a dozen other enterprise AI tools that are already inside the largest professional services firms in the world.