How it works

Why a new kind of assessment

Software development has changed. AI agents can now draft code, tests and explanations at a speed no person can match. The developer’s responsibility has not disappeared; it has shifted towards deciding what should be built, clarifying what is missing, directing the agents well, checking whether the result is right and being honest about uncertainty.

That is why AbilityRating uses many short problems rather than one long coding task or a set of knowledge questions. Each problem represents work that still matters when an AI can write the code: an unclear brief, a code change that looks plausible, a feature request that may be the wrong answer, an AI agent that needs direction, an estimate with incomplete facts and a progress report that must be accurate.

One sitting, many small problems

A candidate works through the problems in one online sitting of at most 75 minutes. AI tools are allowed, as they are at work. Candidates write no code; code to review is shown in a language they choose. Technical understanding is assessed through the problems, not through trivia. Creativity is not measured yet.

The sitting needs no interviewer, so an employer can assess every applicant rather than only those a CV screen lets through. A CV shows where someone has worked and which tools they know; it does not show how they handle an unclear brief or an AI agent’s mistake. AbilityRating never looks at a CV: the score depends only on what a candidate does in the sitting.

What the employer gets

For every candidate assessed for a position: one score, a rating for each quality and the reasons behind them, problem by problem. AbilityRating orders the candidates and says of each pair of neighbours in the list whether one is clearly ahead, likely ahead or too close to call. Every place can be traced back to the answers behind it; an employer can look closely at the qualities that matter most for the job. AbilityRating presents measurements; the employer decides. It has no button to reject anyone and drops no one from the list.

How the score is built

Each answer is scored against the difficulty of its problem, in the way a chess rating is earned against the strength of the opponents. The evidence from all the problems is combined by a stated rule into one score per position, the AbilityRating, with an error band: the range in which the score would most likely fall if it were measured again. Neither a CV nor an interviewer’s impression enters the score.

The same explanation for the candidate

Candidates see the same explanation of their results that the employer sees, problem by problem. They can ask by email for a result to be reviewed by a person, who explains the outcome.

Built on research

• Combining evidence by a stated rule rather than by impression: in hiring, the same information combined mechanically has predicted better than expert judgment (Kuncel and colleagues, 2013).
• Realistic work samples rather than trivia: work samples are among the better-supported selection methods and among those candidates like most (Roth, Bobko and McFarland, 2005; Hausknecht, Day and Thomas, 2004).

Each kind of problem draws on published research; see the evidence.

What AbilityRating is not

• Not an automatic decision. There is no reject button and no pass mark; the employer decides.
• Not a prediction of job performance. AbilityRating runs no job-performance study and collects no hiring outcomes. Employers judge AbilityRating as they would judge a recruiter’s list: by whether its reasons hold up.
• Not a test of memorised knowledge. Deep knowledge of one programming language matters less when AI agents can draft code quickly; understanding what the code does still matters, because someone has to judge it. AbilityRating assesses that understanding through realistic problems rather than quizzes.

Before launch

AbilityRating is being prepared for launch; nothing is sold yet. From the first months of 2027, developers in Europe and the United States take part in early sittings before launch. The problems are designed by an experienced engineer and will be reviewed by panels of practising developers before any employer relies on them. What the early sittings check is set out on the evidence page. Try it as a developer.