The Fed Just Blamed Zoom, Not GPT, for 64% of the Junior Dev Freeze
A June 1, 2026 NY Fed study pins 64% of young grad unemployment on remote work, not AI. What that means for engineering managers still not hiring juniors.
On June 1, 2026, the New York Fed published a Liberty Street Economics post that quietly torched the dominant narrative of the last 18 months. The cause of rising unemployment among recent college graduates, the authors argue, isn't ChatGPT. It's Zoom. And the same week, Apollo's chief economist said there is "zero evidence" AI is killing jobs at all.
If you run engineering hiring, this matters more than another layoff headline. It reframes the diagnosis from "AI replaced our juniors" to "we forgot how to grow them," and one of those problems you can actually fix this quarter.
What the Fed actually said
The post, authored by NY Fed economist Natalia Emanuel with Emma Harrington (UVA) and Amanda Pallais (Harvard), looked at the gap between youth unemployment and overall unemployment since 2019, then split occupations into "remotable" (software engineering, financial analysis) and non-remotable (nursing, skilled trades).
The split is the whole story.
In remotable jobs, unemployment among college grads aged 22-27 rose about 1 percentage point from the 2017-2019 baseline to 2022-2024. For workers 29 and over in the same occupations, it actually declined slightly. In non-remotable work like nursing, the generational gap spiked in 2020 and then normalized. In remote-eligible white-collar work, it never did.
The headline number: the March 2026 unemployment rate for recent college graduates aged 22 to 27 hit 5.6 percent, up from 3.6 percent in March 2019. CS graduates are at 6.1 percent. Computer engineering grads at 7.5 percent. The Q4 2025 underemployment rate for recent grads climbed to 42.5 percent, the highest since 2020.
The mechanism is mentorship, not productivity
The Fed authors are explicit about the channel. Distributed teams make it harder for managers to train and mentor new hires, so employers respond by simply not hiring less-experienced workers into those teams. The grads aren't being replaced by an LLM. They're being filtered out by a calendar that has no apprenticeship slots on it.
This is a management capability gap, not a labor-supply problem. Which means it's also not a problem that gets better by waiting for the AI hype cycle to cool.
Why the "AI ate juniors" story stuck anyway
The competing narrative has been loud and well-financed. Block went from over 10,000 employees to under 6,000 and named AI as a factor. Cisco, Atlassian, Cloudflare, Coinbase, IBM, and Snap all cited AI in their 2025-2026 cuts. A 2024 SHRM survey found 70 percent of hiring managers say AI can do the work of interns, and 57 percent trust AI's output more than that of interns or recent grads.
So the vibes were unanimous. The data is not.
Apollo's Torsten Sløk pointed at ADP employment data and concluded the AI capex frenzy is "stoking both employment and inflation," not displacing workers. Forrester's analysts have been calling this "AI washing." Yale's Budget Lab, in a 2026 report, called the AI jobs wipeout thesis "largely speculative." Even Jensen Huang called blaming AI for cuts a "lazy narrative," and Sam Altman, who has every incentive to oversell AI's labor impact, called it "AI washing."
The simpler explanation: companies that over-hired through 2021-2022 needed cover for cuts they were going to make anyway, and "AI did it" plays better on an earnings call than "we built the wrong org."
Companies needed a cover story for cuts they would have made anyway. AI was a better quote than over-hiring.
What this changes for engineering leaders
If you accept the Fed's framing, three uncomfortable things follow.
1. Your junior pipeline is broken by your operating model, not by the market
SignalFire's State of Talent report found new-grad tech hires at U.S. firms have dropped nearly 50 percent versus pre-pandemic levels. Indeed's 2026 Tech Hiring Outlook found remote-friendly junior roles down 71 percent since 2022. Oracle has been pulling already-extended campus offers from IIT graduates in India. These are not numbers driven by Copilot.
They're numbers driven by hiring managers looking at a fully distributed team and deciding they cannot afford the throughput hit of training someone over Slack. The fix is not to wait. The fix is to design a real in-person onboarding ramp (four to twelve weeks, co-located, with an assigned mentor whose performance review actually weights it) and only then drop the new hire back into the distributed pattern.
2. The cohort you locked out is the cohort that wants the work model that locked them out
A May 2025 Gallup survey found only 6 percent of Gen Z workers prefer fully on-site work. Seventy-one percent want hybrid. The grads being sidelined are the most committed to the arrangement causing the sidelining. You will not solve this by mandating five days in the office and watching your offer-accept rate fall through the floor.
Hybrid onboarding with a remote steady-state is the realistic shape. Two weeks on-site at start, then quarterly anchor weeks, then a senior mentor with calendar time blocked for the junior. None of this is novel. Almost nobody is doing it.
3. In five to seven years, you will not be able to find the mids you didn't train
This is the part that should keep heads of engineering up at night. If the junior-hiring freeze continues through 2027, the missing 2024-2027 cohort becomes the missing senior cohort in 2030-2033. Companies will then claim they "can't find" mid-level engineers while continuing to refuse to train juniors. That's the loop.
The good news for anyone willing to break with the pattern: the pool of strong recent grads who never got picked up is enormous and visible. They are shipping side projects, publishing teardowns, and contributing to OSS to compensate for the missing first job. That work is searchable. The hard part is finding the ones whose code matches the problem you're hiring against, which is exactly the friction Refolk was built to remove: you describe the person in plain English ("recent grad, strong systems work, shipped something real, willing to relocate to a hub office for the first six months") and get a ranked shortlist instead of 400 LinkedIn EasyApply submissions.
The contrarian read on "why companies stopped hiring juniors"
The honest answer is unflattering to managers, not technologists. We built fully remote teams in 2020-2021 because senior engineers liked it and it expanded the hiring radius. We then noticed that nobody on those teams was getting better, and rather than fix the onboarding model, we stopped hiring the people who most needed it.
The Fed's data shows nursing absorbed its 2020 youth-unemployment shock and recovered, because no one tried to train a new nurse over a video call. Software engineering hasn't recovered, because we did try, it didn't work, and instead of switching strategies we switched candidates.
This is the part of the story that the AI narrative was obscuring. It was easier to say "the agents are coming" than "we don't know how to grow people from our couches."
What a "remote-aware" junior hiring plan actually looks like
A few specifics, since the diagnosis without a prescription is just blogging:
- Co-located ramp. First 6 to 12 weeks in person, in a single hub. Non-negotiable. Travel and housing on the company.
- Named mentor, comped accordingly. One senior engineer per junior, with 4 hours a week explicitly carved into the senior's calendar and reflected in their performance review weights.
- PR-pair quotas, not ticket quotas. First quarter, the junior's output is measured by reviewed PRs with the mentor, not by closed Jira tickets. This is the apprenticeship that remote work erased.
- Anchor weeks. After the ramp, quarterly in-person weeks at the hub. This is where the tacit knowledge actually transfers.
- Source against the gap. The 2022-2025 cohort that didn't get picked up is the pool. Tools like Refolk let you query the open web for grads who kept shipping anyway, instead of fishing the same LinkedIn pond every competitor is in.
None of this requires a new headcount approval. It requires admitting that the AI story was convenient and the remote-onboarding story is the real one.
What to do this week
- Pull your last 18 months of hires and split by years of experience. If your sub-2-years-experience hire count is more than 70 percent below your 2019 pace, you have a Fed-shaped problem, not an AI-shaped one.
- Ask your engineering managers, plainly, whether they would hire a junior onto their distributed team today. If the answer is no, ask what would change it. The answers will be about onboarding logistics, not about Copilot.
- Build a co-located ramp program for one cohort of four to six new grads, starting in the next quarter. Source from the 2024-2025 cohort that fell through the cracks. Refolk's plain-English queries are the fastest way we've found to surface those candidates without burning weeks on boolean strings.
- Decide now whether you want to be the company that complains about a missing mid-level pool in 2031, or the company that has one.
FAQ
Does the NY Fed study mean AI has had no effect on junior engineering hiring?
Not exactly. The authors controlled for each occupation's AI exposure and found the remote-work effect was the dominant explanation for the widening youth unemployment gap. AI may still be marginally affecting specific task categories, but the Fed, Apollo's Torsten Sløk, Forrester, and Yale's Budget Lab all converge on the same point: the macro evidence for AI as a junior-jobs killer is weak. The macro evidence for remote-onboarding failure is strong.
If 71 percent of Gen Z wants hybrid work, won't a co-located ramp hurt my offer-accept rate?
Less than you'd expect, if you're honest about the structure. A fixed-length onboarding period (6 to 12 weeks) followed by a hybrid or remote steady-state is materially different from a permanent return-to-office mandate, and candidates read it as such. The grads who turn it down are usually the ones who would have struggled in a distributed senior role anyway. The offer-accept hit is real but small; the retention and ramp-time gains are large.
How do I source recent grads when LinkedIn is saturated with auto-applies?
Stop sourcing on LinkedIn alone. The strongest 2024-2025 cohort that didn't get picked up is on GitHub, in OSS commit logs, in technical writeups, and in personal project repos. That's the surface area Refolk indexes: you describe the engineer you want in plain English and get a shortlist across GitHub, LinkedIn, and the open web, ranked by evidence of the work, not by who filled out their profile most aggressively.
What's the realistic timeline before the "missing juniors" become missing mids?
Three to five years for the early-career squeeze to show up in mid-level hiring funnels, five to seven before it shows up as a structural senior shortage. Companies that start training again in 2026 will have a fully ramped mid-level cohort by 2029-2030, exactly when the rest of the industry will be paying double for the same profile. The math is annoying but unambiguous.