Applicants to Tech Force, the programme that brings private-sector technologists into two-year government posts, are now being interviewed in early rounds by an AI agent. The system, from CodeSignal, runs a structured interview with standardised questions, records and summarises the audio, and hands the hiring manager a transcript to evaluate later. OPM issued guidance last week encouraging agencies down this road, with human oversight.
Two things are worth separating here, because they are usually argued as one.
The structured interview part is fine
Better than fine. The evidence that structured interviews outperform unstructured ones is old, large and boring, and it is one of the least contested findings in personnel selection. Same questions, same order, scored against the same criteria, produces more accurate and more consistent results than a conversation that goes wherever the interviewer's curiosity takes it.
Government hiring managers are not selection specialists. They are engineers and programme staff who interview a few times a year. Handing them a standardised instrument and a transcript instead of an hour of unstructured chat is likely to make their decisions better and more defensible, not worse.
If the argument for this were "we have standardised our early screening," it would be a good argument.
The pipeline argument is the weak one
The case being made, though, is speed — OPM's director has been pushing to expedite federal hiring, and this is presented as part of that.
The trouble is where the time actually goes. This desk reported in August that agencies cannot hire fast enough to spend what Congress appropriated, and the delays in that story were not interview delays. They were position classification, announcement drafting, certificate issuance under veterans' preference and category rating rules, security clearance adjudication, suitability determination, and onboarding. A federal hire is slow in the parts governed by statute and adjudication, not in the part where somebody asks a candidate about their last project.
Compressing the interview stage in that pipeline moves the queue rather than shortening it. Candidates reach the clearance backlog faster and wait there instead. The measurable outcome will be an improvement in time-to-interview, which is a metric that will be reported, and no improvement in time-to-start, which is the one that determines whether an appropriation gets spent.
The part the reporting could not establish
Two gaps in what is known, and both matter more in government than they would in a company.
The first is scale — how many applicants have been through this, and whether any hiring decision has turned on it.
The second is notice. The reporting does not establish whether candidates are told they are being assessed by an AI system. In private-sector hiring that is a disclosure question. In federal hiring it is closer to a due-process question, because federal employment runs on merit system principles and carries appeal rights, and an applicant who does not know how they were evaluated cannot meaningfully contest the evaluation. "Human oversight" in OPM's guidance is doing a lot of work in that sentence, and what it means in practice — a manager reading a transcript, or a manager reviewing a score they did not generate and cannot interrogate — is the whole question.
What it is being asked to fix
There is a reason agencies are reaching for this, and it is not enthusiasm for the technology.
Federal hiring capacity is genuinely degraded. FEMA has lost more than a thousand career staff since 2024 and is rehiring people whose institutional continuity is already broken. Agencies across government are short of the people who run hiring itself, which is its own recursion: you cannot hire your way out of a hiring shortage without hiring staff.
An AI screener is attractive precisely because it is the one intervention an agency can make unilaterally, this quarter, without new authority or new appropriations. That is the same reason employers reach for cost-shifting in health benefits and school districts reach for staff cuts: it is the lever that is available, not the lever that fits the problem.
The measure
Time-to-start, not time-to-interview. Published quarterly, by agency, from announcement close to first day worked.
If that number falls, the tool worked. If time-to-interview falls and time-to-start does not, an agency has automated the part that was not the problem — and it will have a dashboard showing improvement while the appropriation continues to go unspent.
The use of CodeSignal's AI screening tools, including interviews with virtual agents, for applicants to the Tech Force programme; the description of the process as structured interviews with standardised questions whose audio is recorded and summarised for hiring managers to review as transcripts; the roughly eight-month age of the Tech Force initiative and its remit of recruiting private-sector technologists into two-year posts; OPM's guidance last week encouraging agencies to use AI in hiring while calling for human oversight; and OPM director Scott Kupor's push to expedite federal hiring are as reported by CBS News on 3 September 2026, which also records that OPM did not immediately comment and does not establish applicant numbers or whether candidates are notified. The analysis is our own.
Topics nationalfederal workforcehiringai
