So far, the most visible change AI has made to hiring is how cheap it has become to apply for a job. That has arguably reshaped the process more than job losses to automation have.

When it costs a candidate two minutes to produce a tailored application, and an employer nothing to receive one, volume rises until something breaks. Something broke. What follows describes the mechanics rather than forecasting them.

What changed on the employer side

Screening volume

Recruiters widely report more applications per posting than a few years ago, especially for remote-eligible and entry-level roles. The marginal application also tells them less, because the effort it once signaled can now be automated.

The response has been predictable. More automated screening, more knockout questions, more postings closed early, more weight on referrals and internal candidates. Referrals mattered before; in a high-volume environment they matter more, because they are the cheapest way to shrink a pile.

Take-home design

Unsupervised take-home exercises have lost most of their diagnostic value, and employers know it. The replacements are recognizable:

  • Live, timed exercises with someone watching.
  • Take-homes used as a discussion prompt rather than a graded artifact. You build it however you like, then defend every decision.
  • Work-sample conversations about something you have already shipped, probed to a depth that is hard to fake.
  • Explicit permission to use AI tools, with the assessment moved onto judgment, debugging and explanation.

The interview has become more verbal and harder to prepare for by rehearsal. Candidates who genuinely did the work do better than under the old format; candidates relying on polished artifacts do worse.

Entry-level task displacement

The clearest displacement has been of tasks rather than jobs: the small, checkable, high-volume work entry-level roles were built from. First-draft copy, basic research summaries, routine ticket triage, initial data cleaning.

Those tasks were also the training route. Removing them has not immediately removed the jobs in most organizations, but it has weakened the case for hiring a junior employee as an investment, and made the junior roles that remain more demanding on day one.

What changed on the candidate side

Applications are longer, cleaner and more similar. Cover letters have converged on a narrow register that experienced recruiters recognize on sight and discount accordingly. The tailored application that once stood out is now the floor.

Application volume per candidate has risen alongside it, which is rational individually and destructive collectively. When everyone applies to more roles, the number of hires stays roughly the same while the work grows on both sides, and applicants absorb far more silence.

Verification has tightened in response: video screens earlier, more identity checks, more live technical conversation, more scrutiny of claims that cannot be backed up. Processes with no live human contact have become rare at serious employers, partly because of fraudulent applicants and partly because volume made a quick call the cheapest filter available.

What this means in practice

Anything that can be generated is now weak evidence, and anything that can only come from having done the work is strong evidence. Shift your effort from producing more polished applications to producing things a generator cannot supply: named referrers, specifics with numbers and consequences, and the ability to talk fluently about decisions you made and what went wrong.

Which roles grew

The growth has been less exotic than the commentary suggested. In broad terms:

  • Infrastructure and operations: data engineering, platform and reliability work, the unglamorous plumbing any deployment depends on.
  • Evaluation, quality and safety functions: people who test systems, measure outputs and define acceptable performance.
  • Governance, risk and compliance: state and city rules on AI in hiring create functions that have to be staffed. New York City, for example, has required a yearly bias audit of automated hiring tools since enforcement began in July 2023.
  • Sales and implementation at tools vendors: easy to overlook because the titles are ordinary.
  • Domain specialists who can direct the tools: clinicians, lawyers, accountants and engineers whose judgment supplies what the tool cannot.

Demand moved toward work requiring accountability for an outcome, and away from work producing an intermediate artifact someone else checked.

What candidates should do differently

  1. Cut application volume and raise specificity. In a flooded channel, the cold application is a weaker route than it used to be. Referrals and direct contact with hiring managers have become correspondingly stronger.
  2. Carry evidence that is hard to generate. Numbers, outcomes, what broke, what you changed. "Reduced the reconciliation cycle from five days to two" survives scrutiny; "collaborated cross-functionally" no longer even registers.
  3. Prepare to talk, not to submit. Assume every claim will be probed live. Rehearse explaining decisions and trade-offs out loud rather than polishing documents.
  4. Use the tools openly and be able to say how. Employers are asking about workflow. A candidate who can describe where they rely on a tool, where they do not, and how they check it is answering a question hiring managers are actively asking.
  5. If you are early in your career, target organizations that still need generalists. Smaller employers and regulated industries have kept the broad first job, because they have to.
A polished application used to suggest effort and care. Now that software can produce one in minutes, it signals much less. Real results and people who will vouch for you now carry more of the weight.

What is genuinely unsettled

Whether the weakened entry tier reconstitutes in a different shape, how far verification tightens, and what happens to the middle of the skill distribution are open questions nobody currently has the answer to. These effects also vary a great deal by industry, region and employer size, so treat any single national figure as describing a subset. Plan for the changes that have already happened rather than the ones being predicted.

Sources