Loading Career Runway…
Forget the hype cycle. Here is what labor market data, AI capability benchmarks, and workforce research actually tell us about when — and how — AI displaces knowledge work.
We keep writing about this as the evidence moves. One email when it does.
Keep reading
The conversation about AI and jobs has been dominated by two extremes: breathless predictions of mass unemployment, and reassuring claims that AI will only create new roles. Neither is supported by the data.
What the evidence actually shows is more nuanced — and more actionable — than either narrative suggests.
The most important thing to understand about AI displacement is that it follows an exponential curve, not a linear one. Between 2022 and 2024, AI capability in language tasks improved by roughly 40%. Between 2024 and 2026, the improvement was closer to 70% — and the rate is accelerating.
Data point: In standardised coding benchmarks (SWE-bench), AI performance improved from 12% to 49% between January 2024 and January 2026. At the current trajectory, models will exceed 70% by mid-2027.
This matters because displacement does not happen gradually. It happens when capability crosses a threshold — and for many knowledge work tasks, that threshold is approaching faster than most professionals realise.
Based on current AI capability benchmarks, several knowledge work tasks have already crossed the automation threshold where AI can perform them at acceptable quality:
Already automatable (2026):
Approaching threshold (12–24 months):
Still protected (24+ months):
Map almost any knowledge-worker role against current AI capability evidence and the same pattern appears: at least one significant task category in the daily work is something AI can already perform, or will plausibly perform within two years. That does not mean those jobs disappear. It means the task mix underneath them is changing.
What this means in practice: If 30% of your working time is spent on tasks that AI can already do, your role is not being "replaced" — but it is being compressed. The question is not whether you will have a job. The question is whether the reduced scope of your role will support your current compensation and career trajectory.
The data suggests that outright role elimination is rare. What is common is task compression — where AI handles a growing portion of a role's task mix, gradually reducing the headcount needed.
Where companies report AI-linked workforce changes, the pattern is contraction through attrition rather than headline layoffs — affected teams are rehired smaller, not cut overnight. The roles are not eliminated. The tasks are absorbed.
This is the displacement pattern that most professionals miss: your job title survives, but the role beneath it shrinks. Fewer people are needed to do the same volume of work. The least differentiated professionals in each category are the first to feel it.
The displacement timeline is not a countdown to unemployment. It is a window of opportunity. The professionals who use this window to adapt — by building skills AI cannot replicate, by repositioning toward judgment-intensive work, or by becoming the person who orchestrates AI rather than competes with it — will be stronger on the other side.
The ones who wait will find the window has closed.
This analysis draws on data from the Bureau of Labor Statistics, McKinsey Global Institute, and the Stanford HAI AI Index. Methodology details available on our methodology page.