all articles
AI in hiringJob searchRésumé screening

Agents Applying to Agents: The AI Arms Race in Hiring, Explained

Candidates apply with AI and employers screen with it. What that does to replies, recruiters, fairness and the law, and what still gets a human to read you.

Abhishek Anita·Nov 20, 2026·9 min read
Cover illustration for Agents Applying to Agents: The AI Arms Race in Hiring, Explained

More than 300 people now apply for every hire, according to Ashby, and both sides are using AI to cope. Gartner found 39% of candidates used AI while applying, and SHRM found 43% of HR teams use it too.

This piece separates writing help from auto-apply and ranking from rejection. It covers what the evidence says about response rates, recruiter workload, bias and the new laws, gives the strongest case that AI is making hiring fairer, and ends with seven steps that still earn a human read.

The short answer

How we know This piece draws on surveys from Gartner, Greenhouse, SHRM, Indeed and LinkedIn; Ashby's hiring data; an NBER field experiment; court and government documents; and one count of live postings on company career pages we read on 5 November 2026. Sources dated 2023 to September 2026, checked on 20 November 2026. Most figures are self-reported surveys, not audits. No source measures how many applications a person actually reads, or how much of the rise in applications is caused by AI.

Two sets of machines, one inbox

The phrase "agents applying to agents" sounds like science fiction. The evidence says it is partly true, and it helps to be precise about which part.

On the candidate side, AI use comes in layers. Most of it is writing help: polishing a résumé, drafting a cover letter, reading a job post for the skills to stress. Greenhouse found 43% of US respondents used AI to analyse job postings in this way. A smaller layer hands over the whole act of applying to software that finds roles, fills forms and submits. That is the 22% figure for US respondents, and it is the closest thing to a measured rate of "agents applying".

The pressure behind this is easy to see. In Indeed's survey of 1,501 US job seekers (fieldwork August 2026), 77% felt forced to tailor their résumé for every posting and 46% had abandoned an application because it was too tedious. When each application takes an hour and most get no reply, a tool that does it in a minute looks rational.

On the employer side, the machines do different jobs with different amounts of power. Some draft outreach or schedule interviews. Some rank. Indeed's Smart Screening, for example, scores applications from 0 to 100 against the job's criteria, and Indeed says it prioritises applications rather than deciding for recruiters. A few run the first interview. Candidates already assume the screen is automated: in Gartner's Q1 2025 survey, 52% believed AI had screened their application.

WhoWhat the AI doesEvidenceDate
Candidates (Gartner)Any AI use while applying39% of 3,290Q4 2024 survey
US job seekers (Greenhouse)An agent applied for them22%2025
HR teams (SHRM)Any AI use in HR43%, up from 26%2025 vs 2024
IndeedRanks applications, score of 0 to 100Product descriptionMarch 2026
Flynn Group, via WorkdayAutomates scheduling and process21% faster time to hireVendor case, undated
Recruiters (Greenhouse)Spend up to half the week filtering junk34%November 2025

What the arms race does to response rates

When writing an application costs almost nothing, volume rises and attention becomes the scarce thing. Ashby's figure of more than 300 applications per hire is the clearest sign. The New York Times described one HR consultant, Katie Tanner, who received 400 applications within 12 hours and more than 1,200 within a few days, then took the post down. The same report said LinkedIn applications had risen by more than 45% in a year.

Two cautions. First, applications per hire is not applications per job, and a rise can come from a weak market and one-click apply buttons as much as from AI. Second, one viral post is a story, not a trend. Still, the direction is consistent: more applications, the same number of hires, and fewer replies per application.

Silence has other causes too. Some postings are not live hiring at all. Of the live jobs on company career pages we read on 5 November 2026, 29% had been up for more than 90 days, which we cover in our piece on ghost jobs. An agent that applies to everything will apply to those as well. And when silence comes later in the process, it is not a screening problem: Greenhouse found 63% of US respondents had been ghosted after an interview.

What it does to recruiters, and how employers are reacting

Recruiters say the load is heavy. A third report spending up to half their week sorting junk, though that is their own estimate, not a time study. Interviewing has grown too. Ashby reports technical roles averaged 17.6 interviews per hire, 52% more than in 2021, and business roles 11.7, up 36%.

The reactions so far fall into four groups, and the evidence for each is uneven.

Rank with software. Tools like Indeed's score fit against the posting. That speeds review but does not prove the ranking is accurate or fair.

Automate the logistics. Workday's case study says Flynn Group automated over 90% of its hiring process and cut time to hire by 21%, mainly by scheduling. It is a vendor's own account, but it points to a useful idea: you can remove delay without letting a machine make the final call.

Lean on referrals. Greenhouse respondents rated referrals the second most effective route after applying directly to a company. Among US respondents, 43% reported faster responses with a referral and 37% better interview rates. A referral gives a recruiter something a polished document cannot: another person's word.

Test the skill, not the prose. LinkedIn found 72% of recruiting professionals planned to prioritise better skills assessment over the following 12 to 18 months. That is a shift toward work samples and tasks, not proof that employers have dropped the résumé.

There is also a fraud worry. In Gartner's Q2 2025 survey of 3,000 candidates, 6% admitted interview fraud, meaning impersonation. That is far narrower than using AI to write. Gartner's prediction that one in four candidate profiles will be fake by 2028 is a forecast, not a measurement. No source here gives a reliable error rate for tools that claim to detect AI-written applications, so a detector's flag is not proof of anything.

Fairness and the law

The fairness concern is not only that one model may be biased. It is that the same model may sit behind many employers. Stanford HAI describes research covering 4 million applications to 150 employers screened by one vendor's tool, in which applicants were rejected everywhere they applied more often than independent decisions would predict. If one screen shuts you out, a shared screen can shut you out across the market. The research does not show that every such rejection is discriminatory.

The courts are testing who is responsible. In Mobley v. Workday, a court order filed on 1 July 2026 let the plaintiff's allegations about Workday's AI tools stand in the complaint. Workday did not concede they were true. It is a live dispute, not a finding that the software discriminated.

The rules are arriving on different clocks:

JurisdictionRuleStatus
IllinoisAI Video Interview Act: explain the AI and get consentIn force since 1 January 2020
IllinoisEmployment AI rules: notice and no discriminationRules still being finalised, June 2026 update
New York CityLocal Law 144 on automated employment decision toolsCity FAQ published June 2023; effect on hiring outcomes not established
ColoradoSB26-189 on automated decision technologyTakes effect 1 January 2027
European UnionHigh-risk AI rules, which cover systems listed in Annex IIIApply from 2 December 2027 after the July 2026 update

This is general information, not legal advice. Check the official sources for the rules that apply to you.

The counter-view: AI may be making hiring fairer, not worse

The case against panic is strong, and it rests on better evidence than most of the alarm.

The best causal study points the other way. In an NBER field experiment by Emma Wiles, Zanele Munyikwa and John Horton, job seekers on an online labour market who got algorithmic help writing their résumés were hired 8% more often, with no sign that employers were less satisfied. Writing help removes a barrier that has little to do with whether someone can do the job. For candidates writing in a second language, that matters.

AI interviews may help too, when a human still decides. Chicago Booth Review describes a field experiment with 70,000 customer service applicants randomly assigned to a human interviewer, a voice AI, or a choice. Humans made the hiring call. Candidates interviewed by the AI were more likely to get offers and to stay in the job. The summary gives no effect sizes, and one role in one setting says little about engineers or about candidates with disabilities.

So the honest verdict is narrow. Writing help looks good for candidates. Standardised AI first rounds feeding a human decision look promising. Mass auto-applying and opaque automatic rejection are where the evidence is thin and the risks sit. The US Department of Labor's AI and Inclusive Hiring Framework takes this middle line: judge each system by what it automates and whether a person stays answerable.

What still gets a human to read you

No study shows a trick that guarantees a human review. But the evidence points one way: as words get cheap, proof gets valuable. If you want a deeper look at how screening software reads a résumé, see how to get past AI résumé screening.

  1. Apply to fewer, live roles. Check the posting date and the company's own careers page before you apply. A role up for three months may not be hiring.
  2. Apply on the company's site where you can. Greenhouse respondents rated direct applications the most effective route.
  3. Get a referral for the roles you want most. One credible introduction adds information a recruiter cannot get from your résumé.
  4. Use AI to write clearly, not to write more. The 8% gain in the NBER study came from clearer résumés, not more applications.
  5. Put checkable proof near the top. Shipped work, numbers, links to code or writing. Things a screen can match and a person can verify.
  6. Match the posting's real requirements honestly. Use its words where they are true of you. Claims you cannot back up fail at the interview, and faster now that interviews test skills.
  7. Expect a test and prepare for it. With more assessments and more interview rounds, the skill itself is what gets you through.

The arms race has not removed people from hiring. It has made the application cheap and made trust expensive. The candidates who stand out are the ones who give a busy person a reason to believe them.

Frequently asked

How many job seekers use AI to apply for jobs?

Gartner found 39% of 3,290 candidates used AI during the application process in a Q4 2024 survey. Greenhouse found an AI agent had submitted applications for 22% of US, 17% of UK and 16% of Irish job seekers it surveyed in 2025.

Do employers use AI to reject applications?

SHRM reported 43% of HR teams used AI in 2025, up from 26% in 2024, but that is broad HR use, not automatic rejection. Tools such as Indeed's Smart Screening rank applications with a score from 0 to 100, and Indeed says it prioritises rather than decides.

Why do so many job applications get no reply?

Volume is one reason: Ashby reports more than 300 applications per hire throughout 2025, triple the 2021 level by 2024. Stale postings are another: 29% of live jobs on company career pages read on 5 November 2026 had been up for more than 90 days.

Does using AI to write a résumé help or hurt?

The best causal evidence says it helps when it improves clarity. In an NBER field experiment, job seekers given algorithmic résumé-writing help were hired 8% more often, with no sign employers were less satisfied.

What laws cover AI in hiring?

Illinois has required consent for AI video interview analysis since 1 January 2020, and New York City regulates automated employment decision tools under Local Law 144. Colorado's SB26-189 takes effect on 1 January 2027, and the EU's high-risk AI rules apply from 2 December 2027. This is general information, not legal advice.

Done reading? Let landed apply.

landed finds roles like the ones in this piece, applies on each company's own site in the first hours, and writes every answer. Your first 10 applications are free.

Get 5x interviews

Related reading