The clearest window into how artificial intelligence is reshaping work may not be traditional employment figures, but granular data tracking which specific tasks workers perform — and which ones are quietly disappearing, according to reporting by MIT Technology Review.
The debate over AI and jobs has long been hampered by a fundamental measurement problem. Headline unemployment numbers and hiring surveys track whether people have jobs, not what those jobs actually require day to day. As AI tools absorb discrete tasks — drafting emails, summarising documents, writing basic code — overall employment figures can remain stable even as the nature of work shifts dramatically beneath the surface.
Why Standard Employment Data Misses the Point
Economists have known for decades that technological disruption tends to hollow out specific tasks before it eliminates entire occupations. The pattern played out with ATMs, which reduced cash-handling duties for bank tellers without immediately shrinking the total number of tellers. AI, researchers argue, is following a similar but potentially faster trajectory.
Within some circles, confidence in a coming jobs crisis is already absolute. A societal impacts researcher at Anthropic recently responded to public concern by signalling that the disruption may arrive sooner and more severely than most forecasts suggest, according to MIT Technology Review. That kind of institutional candour from a leading AI developer is rare, and it reflects growing internal acknowledgement that the workforce effects are real — even if the data to quantify them precisely does not yet exist at scale.
The clearest window into how AI is reshaping work may not be traditional employment figures, but granular data tracking which specific tasks workers perform — and which ones are quietly disappearing.
The core proposal gaining traction among labour economists is to build or expand datasets that log occupational tasks at fine resolution — not just "software developer" or "paralegal," but the specific activities those workers complete each hour. The O*NET database maintained by the US Department of Labor already attempts something like this, cataloguing hundreds of tasks across thousands of occupations. But researchers argue it is updated too infrequently and lacks the real-time sensitivity needed to detect AI-driven change as it happens.
What Task-Level Data Would Actually Reveal
If employers and governments began collecting task-frequency data routinely — through payroll systems, workplace surveys, or time-tracking tools — analysts could begin to answer questions that current statistics cannot. Are junior lawyers spending fewer hours on document review? Are customer service agents handling a different mix of queries after AI triage tools were introduced? Are radiologists reading fewer routine scans?
These granular shifts matter enormously for workers even when job titles and headcounts stay the same. A role stripped of its entry-level tasks may offer fewer opportunities to build skills, compress career ladders, and reduce earnings mobility over time. Research on earlier waves of automation — including a widely cited 2003 study by economists David Autor, Frank Levy, and Richard Murnane examining US census and occupational data — found that technology consistently eliminates routine cognitive and manual tasks first, with complex and interpersonal tasks proving more durable. AI appears to be extending that pattern into domains previously considered safe, including some creative and analytical work.
The challenge is practical as well as political. Collecting task-level data at meaningful scale requires employer cooperation, standardised definitions, and sustained funding — none of which is guaranteed. Privacy concerns arise when tracking what workers do in fine detail. And companies competing on AI-driven productivity gains have limited incentive to disclose publicly where automation is replacing human effort.
The Policy Gap That Task Data Could Close
Without better measurement, policymakers face a structural disadvantage. Retraining programmes, educational curricula, and social safety nets are typically designed around occupational categories. If the real disruption is happening at the task level — silently, inside roles that still officially exist — those interventions may be targeting the wrong problem.
Several research institutions and government agencies are beginning to push for change. The OECD has called for enhanced occupational data collection among member countries. Some academic teams are experimenting with scraping job postings to track how skill requirements within specific roles shift over time, treating changes in advertised duties as a proxy for task displacement. Early results from these efforts suggest AI-related task changes are already visible in sectors including finance, legal services, and software engineering.
For individual workers, the uncertainty is not abstract. Someone entering the workforce today in a field exposed to AI automation is making a multi-decade career bet with incomplete information. Task-level data, if collected systematically, would allow workers, unions, and educators to see where demand for human effort is concentrating and where it is thinning — giving them a factual basis for decisions rather than speculation.
What This Means
Without task-level employment data collected at scale and in real time, workers, policymakers, and employers will continue making consequential decisions about education, hiring, and safety nets based on economic indicators that were designed for a different era of technological change — and the human cost of that blind spot will fall hardest on those with the least room to absorb a wrong bet.