Time to competence is the time it takes a person, from starting a role or a programme, to reach the standard needed to do the work unsupervised. For people reviewing AI agents, it is the period between joining the team and being trusted to approve held actions alone, backed by evidence that they can.
Why does time to competence matter?
It sets the cost and pace of a deployment. If a reviewer takes three months to reach competence, an agent whose depth depends on trained reviewers cannot be raised for three months. It also sets staffing: a team that loses a competent reviewer cannot replace them overnight.
It is also a measure of the training itself. If time to competence is long and varies widely between people, the programme is probably weak in a specific area. If it shortens after a change, the change worked.
How is it measured?
The measure is only as good as its end point. "Competent" needs a definition that does not depend on opinion:
- Start point: the day the person begins training for the role.
- End point: the day they meet a defined standard, such as passing a summative assessment and completing a period of supervised review without a red-line miss.
- Evidence: test results, the record of their decisions, and a sign-off.
Track the median and the spread. A team where most people take four weeks but a few take twelve has a different problem from one where everyone takes eight.
An example
An insurer's claims team measures time to competence for reviewers of an agent that drafts settlement letters. At first, it averages about ten weeks, with a wide spread. Looking at who took longest, the training lead finds they all struggled with the same concept: distinguishing a declared repair estimate from a signed one. She adds a short module with scenarios on that point. Over the next intake, time to competence falls to about six weeks and the spread narrows.
For insurers in the EU, competence is also a deployer concern: the AI Act sets obligations for deployers and expects staff dealing with AI systems to have sufficient AI literacy. The piece on Article 26 deployer duties and staff competence in insurance covers that angle.
How is it different from a time-boxed exam?
A time-boxed exam limits how long a candidate has to complete one test, typically under an hour. Time to competence covers the whole period from first day to proven standard, measured in weeks. The exam may be one of the checkpoints that marks the end of that period.
It also differs from time to productivity, which measures when someone produces output at a normal rate. A reviewer can be productive, clearing many cases, before they are competent, meaning they catch the hard ones.
Which terms are related?
- Evidence of competence: what proves the end point was reached; see what counts as evidence of competence.
- Shadow period: supervised practice, often the longest part.
- Summative assessment: the final check.
Leaders face the same question about their own training time; see training senior leaders without wasting their time. For the full approach, read training teams for AI agents.
Questions people ask
Should you try to shorten time to competence?
Yes, but not by lowering the end point. Shorten it by better training, more practice on the hard cases, and protected study time.
Does experience in the old process shorten it?
Sometimes, but not always. Experience helps with the business context and can slow learning of how the agent decides, because old habits need to be unlearned.