Applications per recruiter jumped 412% yet under 7% of applicants get interviews
The Roles Research Actually Shows
No company named Adam is advertising four AI roles. The research points instead to hiring signals at four institutions expanding their AI footprint.
UT Austin's Center for Generative AI runs on a 600-GPU NVIDIA H100 cluster, one of academia's largest, backed by a $5 million gift that launched the Machine Learning Laboratory in 2020, a cross-campus ethics initiative in 2019, and a 2022 leadership investment from alumnus Sanjay Chandra. The center aims to accelerate health care, drug development, and materials science. That mission creates demand for researchers who can translate massive compute into domain breakthroughs.
At MIT, a photonic processor developed over a decade by Dirk Englund's group performs full deep-neural-network inference optically, running key computations in under half a nanosecond at better than 92 percent accuracy, using CMOS-compatible fabrication. The team, including Saumil Bandyopadhyay, is targeting integration with cameras and telecommunications gear and exploring algorithms that exploit optical speed. The trajectory demands physicists and hardware engineers who can co-design algorithms and photonic circuits, not just software researchers.
AWS, still the cloud leader with $9.4 billion in quarterly operating income (nearly two-thirds of Amazon's total), has cut more than 27,000 jobs since 2023 while sinking $4 billion into Anthropic and designating its custom Trainium and Inferentia chips as the startup's primary silicon. The leadership change from Adam Selipsky to Matt Garman, described by a source as a shift to a "wartime" posture in generative AI, signals that Amazon views its AI talent gaps as existential. Azure has grown from three-fifths of AWS's size in 2021 to nearly three-quarters today, per CNBC-cited analyst estimates.
A UC Riverside–Caltech study published in December 2024 quantifies a parallel constraint: training a model at the scale of Meta's Llama-3.1 produces air pollution equivalent to 10,000 round-trips between Los Angeles and New York. Projected U.S. premature deaths reach 1,300 annually by 2030, with public-health costs near $20 billion. The authors, including Caltech's Adam Wierman, recommend mandatory reporting of data-center pollutants and compensation for affected communities, a regulatory vector that will shape hiring for compliance, sustainability, and systems-efficiency roles across the industry.
How the Screen Works
Greenhouse's acquisition of Ezra AI Labs produced a voice-based screening system that replaces the résumé screen with a structured conversation. Applications per recruiter on Greenhouse's platform have jumped 412 percent since 2023, while fewer than 7 percent of applicants land an interview. "We have more candidates than ever, but less clarity on who's actually qualified," said Greenhouse cofounder and CEO Daniel Chait. The first wave of AI tools in applicant tracking systems "saw the job seekers as the problem" and created more "friction" for applicants, Chait said.
The Ezra system works with hiring managers to generate a role-specific rubric, then conducts a natural-language interview that "needles away at generalizations" to create precise evaluation criteria. Every candidate receives the same questions, evaluated against the same rubric, whether they interview at 9 a.m. Monday or 4 p.m. Friday — consistency human interviewers rarely achieve. The AI flags potentially illegal or noncompliant inputs before they enter the rubric. Candidates can interrupt the voice agent at any point.
The transcript, structured scores, and conversational data become the new screening signal. The system detects overly scripted or AI-generated responses, according to Greenhouse. Candidates can also question the AI interviewer about the role, compensation, and benefits, topics they "feel awkward asking a human recruiter, lest they come across too mercentile," said Samson of Ezra.
This inverts the traditional funnel: instead of winnowing based on résumés (now frequently AI-optimized), every applicant gets an interview conversation that produces structured, comparable data. "As soon as you apply to a job, you're now eligible for an interview," Chait said. "Which all of a sudden means: don't apply to the job unless you're willing to spend some time talking to them and taking that first interview." The goal is to break the cycle where candidates spray AI-tuned résumés and employers deploy AI filters that miss signal.
For any company running this playbook, whether Greenhouse customers adopting Ezra or another firm, the screening criteria derive from the hiring team's rubric: technical depth, project ownership, communication clarity, and alignment with role-specific competencies, all captured in a voice conversation rather than a document.
What Gets Candidates Through?
Research on what moves candidates through initial screens points consistently toward demonstrated growth and specific storytelling, not credentials or rehearsed talking points. Adam Grant, organizational psychologist at Wharton, has argued that "how well somebody does a job is not indicated by how the first interview goes, it's how much growth they show from the first interview to the second." That insight reframes the screen: the strongest signal isn't a polished first answer but the candidate's capacity to incorporate feedback and improve in real time.
Grant's recommended test is direct: "Give them a challenge that's really part of the job and watch how they handle it." Candidates who pass treat the exercise as a collaboration, not a performance. They ask clarifying questions, show their reasoning, and iterate when the interviewer pushes back.
Danny Meyer, founder of Union Square Hospitality Group and Shake Shack, uses a different but related filter. He asks: "What's the biggest misperception other people have of you?" The only honest answer, he told Grant on the WorkLife podcast, is "Well, I'm really this, but the dangest thing is that people actually see me as that." Meyer looks for the self-awareness to name the gap and the vulnerability to own it, traits that correlate with the growth mindset Grant cites as the new hiring currency.
Meyer's second question, "Tell me about something that happened in your life, before you turned 12, that you think had more of an impact on you today than anything else," serves the same end. He's not evaluating the childhood event; he's evaluating the candidate's ability to trace a line from experience to identity. "Whatever that story is, you then get a chance to talk about how did it change who you are today," Meyer said. "I'm looking for honesty, vulnerability, willingness to grow."
Grant's culture-test question operates on the same principle from the other side of the table: "Can you tell me a story about something that happens here but would not elsewhere?" When candidates ask this, or variants like "What are the common themes among your highest performers?" and "What would the first 30 days look like?", they signal they're vetting for fit, not just chasing an offer. The stories interviewers tell in response reveal whether an organization actually has psychological safety, fairness, and control, or just platitudes.
Mike O'Neill, former CEO of BMI, takes a behavioral approach. He asks candidates to pick the restaurant for a meal interview. "I want to see if they're trying to impress me, or if they make me go to them or they come to me," he told the New York Times. He also probes their reference points: "Who was the person you really wanted to work for? Who was the person you wanted to run from? Why? What were the traits of those people?" The answers map the candidate's internal compass, showing what they value, what they reject, and whether those align with the team they'd join.
Across these frameworks, candidates who clear the screen share a pattern: they substitute narrative for assertion. They don't claim "I'm a fast learner"; they describe a project where they hit a wall, sought feedback, and shipped a better version in half the time. They don't say "I value collaboration"; they recount a conflict with a teammate, what they learned about the other person's constraints, and how the work improved. The screen selects for people who can make their thinking visible, people who know the difference between a story that reveals character and one that merely entertains.
Where the Market Is Heading
Cursor's rejection of the traditional funnel is reshaping how candidates approach the market and how competitors think about hiring. Adam Ward, head of talent at Cursor, describes the current environment as an "11" on a scale where 10 was already extreme, a bifurcation so wide he calls it the widest of his career. One mid-2026 news cycle carried adjacent headlines: NBA-player-level salary offers for new AI PhDs, and a blue-chip company laying off 10 percent of its workforce. That split defines the pool. Forward Deployed Engineers, design engineers who can build, and "power ICs" (individual contributors with taste, judgment, and problem-solving) are trending up. Hyper-specialized narrow roles and early new graduates without multi-skill depth are softening.
Candidates have noticed. The reputation for long, project-based on-sites where applicants do real side-by-side work with the team has shifted preparation strategies. Engineers now treat work samples as the primary currency. Ward notes they're the highest predictor of hiring success, yet most companies still run standard across-the-table interviews. Applicants targeting Cursor-style processes invest in portfolio projects that demonstrate taste and decision-making, not just execution. The "power IC" traits Ward emphasizes (curiosity, problem-solving, judgment) are becoming the new interview prep framework.
Recruiters feel the pressure too. Ward observes that recruiters operate excellently at 90–110 percent capacity but collapse outside that band, which explains the rising offer-reneges in the hot market. When every candidate holds multiple high-stakes offers, the traditional funnel's throughput obsession backfires. Companies still blasting outreach to hundreds and screening whoever replies end up with what Ward calls "the remainder": the last person standing after stage-by-stage attrition, not the best. Over multiple hires, that produces regression to the mean.
The industry reaction is split. Ward's "pillar of excellence" is straightforward: define the top 20 percent for your specific company, map the 50 people in the world who fit, and pursue them relentlessly. That playbook works for Cursor's scale and density. But he leaves the tension unresolved: whether the bespoke, executive-style method survives at mass scale. For companies insisting on funnel-scale throughput, the pillar remains aspirational. Meanwhile, a new function is emerging. Ward forecasts a "talent engineer" role, possibly a former engineer building internal recruiting tools, and argues every recruiter should become one now that tools like Cursor put that capability inside the recruiting team itself.
The broader shift Ward frames as a "resettlement of talent and labor" that will reconverge over time. Right now, the companies winning scarce talent are the ones abandoning the funnel entirely. The rest are still optimizing a model that selects against the people they actually want.
Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.