1 Human Resources AI Agents, Live in Production.

HR AI Agents: Automating the Administration, Guarding the Judgement

HR carries an unusual combination: high administrative volume and decisions that change people’s lives. Screening a hundred CVs, checking that right-to-work records exist, validating payroll input before the cut-off, and confirming that a policy answer matches the handbook are all mechanical work done under time pressure. Sitting inside that same workload are the decisions where a shortcut is not a rounding error — a screening criterion that quietly filters by something it should not, a performance record that describes a person rather than their work, an engagement survey cut so finely that a colleague can identify who said it.

That is why the agents here are built with refusals rather than only capabilities. The CV parser strips details that should never reach a screening decision instead of passing them along. The interview builder produces an explicit list of what must not be asked. The job posting agent drops wording that narrows who applies. The salary validator checks whether the gaps between comparable people can be explained by something other than who they are. The survey analysis suppresses any cut small enough to deanonymise a respondent. And nothing that reaches an employee or a candidate — an offer, a rejection, an acknowledgement — is sent without a person approving it. The administration is where the time goes; the judgement is where the harm is, and these are designed to separate the two rather than automate both.