You have been writing cover letters that open with some version of "the AI hiring shock hit new grads hardest." A CESifo paper published Sept 28 just took that framing away from you. Recruiters read the same TechRadar and Washington Post coverage you did, and the number they now have in their head is 7.3%.
The Sept 28 paper, in one paragraph
The AI-caused new grad unemployment story is not supported by the data. Robert Fairlie (UCLA) and Jane Wu's CESifo working paper looked at bachelor's degree holders ages 22 to 25 who are not enrolled in further education, and found their summer 2026 unemployment rate was 7.3%, sitting inside the 2022 to 2024 band. It was higher than 6.3% in 2022, but lower than 7.8% in 2024. In other words, 2026 is better for new grads than 2024 was, which was well before ChatGPT became a hiring narrative.
The authors also sorted roles by "AI exposure" using a 2023 classification of which jobs LLMs handle best. Across nearly every comparison, the 2022 to 2026 trend differences were not statistically significant. They hedge, correctly, that this "does not rule out larger effects in the future." But for right now, on a resume you are sending this week, the "AI took my job" framing has just become the thing that identifies you as someone who reads headlines instead of papers.
Higher than 2022's 6.3%, but lower than 2024's 7.8%. Inside the pre-AI-panic band.
Why this changes what a recruiter reads into your resume
Recruiters see the same three sentences at the top of every new grad cover letter this fall: hiring is brutal, AI ate entry level, please give me a chance anyway. The Sept 28 paper gives them permission to treat that opener as either uninformed or as an excuse. The candidate who instead writes "the market is cyclical and I picked a lane" reads as someone who did the reading.
What actually happened to the entry pipeline
The entry pipeline did not shrink uniformly, it bifurcated. Generic "SWE new grad" postings collapsed, while entry-level roles that require applied AI or ML skills grew into the larger pool. If your resume is targeted at the first bucket, you are competing for a residual. If it is targeted at the second, you are competing on proof of work.
The supply-side collapse is real, and it predates the AI story. Indeed's software development posting index (100 = February 2020) tells the timeline:
| Date | Indeed software dev index |
|---|---|
| 28 Feb 2022 | 233.8 |
| 1 Jan 2023 | 130.3 |
| 1 Jan 2024 | 72.6 |
| 17 May 2025 | 61.1 |
| 18 Sep 2026 | 77.3 |
That is a ZIRP unwind, not a robot uprising. Rates went up, growth capital dried up, and companies that had been over-hiring for three years stopped. AP's own read of the same period puts recent computer science and computer engineering graduate unemployment at about 7.1%, which is bad, but not distinguishable from the general 22 to 25 bachelor's rate. SignalFire's Asher Bantock has the counter-evidence you should still acknowledge: Big Tech cut new graduate hiring by 25% in 2024 versus 2023, and startups cut theirs by 11%. That is a Big Tech story, not an economy-wide AI story.
The bifurcation, in Refolk's index
The entry-level AI and ML pool is 1.66x larger than the generic entry SWE pool right now, which inverts what most new grads assume. In Refolk's index of US entry-level postings, the numbers are unambiguous, and they should decide which resume you send tomorrow morning.
| Segment | Live count | What it tells you |
|---|---|---|
| Generic entry SWE (Assoc./Jr./Entry Software Engineer) | 11,478 | The pool you are all fighting over |
| Entry ML/AI/Data Scientist with Python + ML skills | 19,106 | 1.66x larger, filtered on real skills |
| Top employers, entry SWE | Capital One, Goldman Sachs (tied) | Neither is a tech company |
| Top employers, entry AI/ML | Meta among the leaders | Big Tech still hires here, just not for E3 SWE |
| Top city, entry AI/ML | New York edges SF | NY beats SF for entry AI headcount |
Two things fall out of this table. First, the bottleneck in AI/ML entry roles is not headcount, it is proof of work. Handshake reported that graduating seniors in 2026 mention AI skills on their resumes at twice the rate of the class of 2022, and 74% of those mentions are tied to real-world projects rather than coursework. Recruiters have already recalibrated. "Took Andrew Ng's course" is now noise. A repo with evals, a deployed demo, or a paper reproduction is the signal.
Second, if you are writing a resume for the generic SWE bucket, your target list is wrong. The top entry SWE employers in Refolk's index are Capital One, Goldman Sachs, L3Harris, Veeva Systems, and ServiceNow. Banks, insurers, gov contractors, and vertical SaaS. New grads pointing every version of their resume at Meta and Google are optimizing for a pool that Meta itself is trying to close by Thanksgiving.
Meta pulled E3. That is a calendar problem, not a freeze
Meta's E3 2026 new grad job description got deleted mid-cycle because the class is almost full, not because Meta stopped hiring new grads. On Blind, a candidate reported seeing a new grad posting on Aug 15, applying, and Meta deleting it on Aug 20. Meta aims to fill 70 to 80% of its new grad class by Thanksgiving 2026, and spring recruiting is historically reserved for backfill or PhD candidates.
Read that mechanically:
- If you are applying to Meta E3 in January, you are competing for the last 20 to 30% of seats, and most of those seats have a PhD requirement attached.
- If you go through ColorStack or Rewriting the Code, you sometimes see Meta links one to two weeks before the public careers site. That lever is real and most new grads do not use it.
- If your resume mentions "AI hiring freeze" as your reason for still looking in February, you are telling a recruiter you missed the calendar, not that the calendar was unfair.
The right resume rewrite here is a calendar rewrite. Move your target list off the companies that closed in November and onto the ones (Capital One, ServiceNow, Veeva, L3Harris, mid-market ML shops) that hire year-round because they do not run a formal grad program.
Nvidia's entry pipeline: a BS is now a screening filter
Nvidia's new grad engineer roles have effectively bifurcated by degree, and applying to a masters-required posting with a bachelor's is a wasted slot. A recent audit of Nvidia's new grad postings shows four role families now list MS or PhD as a requirement:
- Software R&D for Digital Logic Synthesis
- VLSI Physical Design
- Hardware Tools and Methodology
- Circuit Design
Bachelor's candidates are realistically limited to three families:
- Verification Engineer
- SOC Hardware Engineer
- Applied AI Engineer (Silicon Co-Design group)
Nvidia's own AI/FSI Developer Technology Engineer New College Grad 2026 posting explicitly requires "pursuing or recently completed a Master's or PhD degree in Computer Science, Computer Engineering, or Electrical and Computer Engineering." An ATS does not care that you are a strong candidate on paper. It matches degree strings. Applying anyway costs you nothing except the referral you could have burned on a role you can actually get.
Referrals matter more than any other lever here: at Nvidia, they push response rates from the 5 to 8% baseline to 20 to 25%. That is a 3 to 4x multiplier, and it is finite. Spend them on the three role families that will read your resume.
Tailoring your resume for one of those three families, one job at a time, is the exact work Refolk takes off you. Paste the Nvidia Applied AI Engineer JD, get a version of your resume back that answers that posting, with your Silicon Co-Design-adjacent projects lifted to the top and the generic web dev internship demoted. Then a fit score that tells you whether burning your one Nvidia referral on this specific req is actually worth it.
What your 2026 new grad resume should say instead
Drop the AI narrative from the top of your resume and cover letter. Replace it with a two-line frame that a recruiter can read in five seconds and a project section that proves the frame. Concretely:
- Kill the AI framing. No "in a market reshaped by AI." No "given widespread hiring freezes." Recruiters read those as excuses now, and the Sept 28 paper is the reason.
- Pick a bucket. Either you are competing in the 11,478-role generic SWE pool (target: Capital One, Goldman, ServiceNow, Veeva, L3Harris) or the 19,106-role entry AI/ML pool (target: any company with a real ML product). One resume cannot do both credibly.
- Prove the AI skill, do not list it. 74% of AI mentions on 2026 grad resumes are tied to real-world projects. Put a repo link, a demo URL, or a paper reproduction in the header. If you cannot, take the generic SWE bucket.
- Match the degree filter. Read the Requirements section, not the Preferred section. If it says MS/PhD required, close the tab. Nvidia will not make an exception.
- Fix your calendar. If you graduated in May and it is now February, stop applying to programs that closed in November. Move to year-round hirers.
The AI took my job story is now the thing that identifies you as someone who reads headlines instead of papers.
The one-line reframe recruiters will actually read
A recruiter has 20 seconds. Give them a line that acknowledges the market without whining about it. Something like: "Recent CS grad targeting applied ML roles, chose to skip the FAANG spring backfill lottery and focus on shipped projects." That reads as someone who knows the 7.3% number, knows the Thanksgiving deadline, and picked a lane. It is also, not coincidentally, true.
FAQ
Should I still mention AI on my resume at all?
Yes, but as evidence, not as an excuse. Mentioning applied AI or ML skills doubled among 2026 grads for a reason: the entry AI/ML pool is 1.66x larger than the generic SWE pool in Refolk's index. What has changed is the burden of proof. A line that says "Python, TensorFlow, LangChain" is now noise. A line that says "shipped a retrieval-augmented eval harness on 12k support tickets, repo linked" is the signal. Handshake's 74% figure (real-world projects, not coursework) is the bar.
If unemployment for new grads is really only 7.3%, why does my search feel impossible?
Because the Indeed software dev posting index sits at 77.3 versus a February 2020 baseline of 100, down from a peak of 233.8 in February 2022. The overall labor market is fine. The specific slice you are targeting, tech SWE roles at brand-name tech companies, contracted by roughly two thirds from peak and has only partially recovered. Your search feels impossible because you are applying to the slice that shrank most. Widen to Capital One, Goldman, ServiceNow, Veeva, and L3Harris and the response rate changes.
Is it worth applying to Meta or Nvidia new grad postings in the spring?
Rarely, and only if you fit the exact filter. Meta fills 70 to 80% of its new grad class by Thanksgiving 2026, and spring is backfill or PhD. Nvidia's new grad postings have hard MS/PhD requirements on at least four role families, and if you have a bachelor's, an ATS will filter you out before a human reads the resume. Spend spring on year-round hirers and on the three Nvidia families that accept a BS: Verification, SOC Hardware, and Applied AI.
How do I get in front of Meta before the public posting drops?
Diversity partner channels post Meta links one to two weeks before the public site. ColorStack and Rewriting the Code are the two most-cited by candidates who actually landed E3 seats last cycle. Combine that with a referral (Nvidia's data shows referrals move response rates from 5 to 8% up to 20 to 25%, and Meta's structure is similar) and you are effectively applying in a different queue. This is worth more than three weeks of cold applying.