Assess the judgment behind the code

AbilityRating assesses software developers through many short, realistic problems: clarifying an unclear brief, catching a plausible mistake in a code review, deciding what not to build, directing an AI agent, estimating under uncertainty and reporting progress honestly. AI tools are allowed. The evidence is combined by a stated rule into one score per position, with the reasons behind it.

Try it before launch How it works

Development work has changed

AI agents now write code at a speed no person can match. The developer's work has moved: understanding what is really needed, directing the agents, judging whether what they produce is right and taking responsibility for the result. Interviews built on knowledge questions were designed for the old work; they say little about who does the new work well. AbilityRating measures the new work directly.

Built on published research: the problems and the rating method draw on studies in personnel selection and measurement, tracked in a register of more than 300 meta-analyses, standards, laws, peer-reviewed studies and official documents.The evidence

For employers

Assess every applicant, not only those a CV screen lets through. The sitting needs no interviewer, so assessing many candidates costs a fraction of the time a round of interviews takes. For each candidate you see one score, a rating for each quality and the reasons behind them, problem by problem; you decide who to meet.

Hear when the first pilot opens

For developers

Realistic problems instead of trivia, AI tools allowed, no code to write and no CV. You see the same explanation of your results that an employer would see. Developers can sign up now for the early sittings before launch.

Before launch

AbilityRating is being prepared for launch; nothing is sold yet. From early 2027, developers in Europe and the United States take part in early sittings before launch.

One sitting, about 75 minutes

A brief from a stakeholder that leaves things out. A feature request that should perhaps not be built. Two short code reviews. A specification for an AI agent to build, then a change of requirements. A handful of estimates. A progress report. All in one invented setting, from your own computer.