WorkTrust research
What should capability mean in the age of AI?
Our models for judging professional ability were built for the twentieth century. The way people work has changed; the way we prove what they can do has not. This is the research into what should replace it.
Why now
Twentieth-century standards, twenty-first-century work.
- 01
The standards are inherited
Degrees, job titles and years of experience were designed when careers moved slowly and a qualification lasted decades. They are still the international default, and they were never meant to describe how somebody works. Only how long they have been at it.
- 02
The work changed underneath them
AI and digitalisation are an industrial revolution in the ordinary sense: they change what a person can produce, how fast they learn, and which problems are within reach. A year of work in 2026 is not a year of work in 2006, and no existing standard notices the difference.
- 03
So the question is open
If capability can be demonstrated rather than claimed, what do junior, senior and expert mean? What should a university teach when the tools change every quarter? Nobody has answered this yet, which is why it is worth researching rather than asserting.
The work
A benchmark for capability, built on evidence rather than claims.
An AI capability benchmark
A measurement standard for what somebody can actually do with AI. Defined openly, scored from observed work rather than self-report, and validated before it is used for anything that matters.
Capability and skill sets, measured
Which capabilities separate people in practice, how they are evidenced, and how much a year of experience still explains once the evidence is on the table.
What the next generation should learn
If the benchmark holds, it says something about schools and universities too: which capabilities are worth teaching, which are being automated, and what a curriculum should look like when the ground moves every year.
Where the research stands.
Openly, and without rounding anything up. There is no cohort, no benchmark and no result yet. That is what this is for.
Join the researchProfessionals · Students · Researchers · Universities · Enterprise · Public sector
MIT R&D application submitted to RVO
An MIT R&D collaboration project application has been submitted to RVO, on researching and validating capability in AI-assisted work. Submitted. Not approved, not funded, not awarded.
A published validation protocol
The method is written down before it is run: Kane's four inferences, a twelve-row validity matrix, and seven phases. None of it has run yet, and the product says so on every score it shows.
Research partners
Follow-up research is being developed with researchers and research partners: measurement, validation, benchmarking, and what professional capability should mean next.