Semafor’s Rachyl Jones filed a piece on May 15 headlined “The Class of 2026 is cooked”, and unlike the half-dozen “AI is coming for your job” hand-wavers that have rolled across the trade press all year, this one came with a specific casualty list and three quotable executive lines. The headline number every recent-grad will have already seen in the group chat: the U.S. unemployment rate for college-degree holders ages 22-27 was 5.8% at end of Q1, materially above the 4.0% national rate, and underemployment among the same cohort was 42.8% — the highest reading since the pandemic.
The cohort-specific numbers are worse. Per the New York Fed labor-market data Semafor cited, the unemployment rate among recent graduates in computer science is 7.0% and computer engineering 7.8% — the two majors that the 2022-era career-services pamphlets specifically pointed kids toward when they were filling out their majors as freshmen. Those pamphlets were printed before ChatGPT shipped.
The specific casualty list
The Semafor piece names three.
- Vaishali Hireraddi, 23, a UC Davis business analytics master’s candidate, has applied to 500 jobs. She has two offers — both unpaid. One of the two told her they’d pay her “at some point in the future when they have enough money.” She is paying $100/month for a Claude subscription to spin up portfolio projects “that look just good enough to show to recruiters,” and to rewrite each of her 500 applications with the right keywords to get past the hiring bots. (Semafor)
- Theodore Skondras, 23, an MIT master’s student, has pivoted to applying to Y Combinator with an AI-agent payments project. The traditional applications path was generating no offers; the YC application generated at least a first conversation.
- Ezra Rosser, 22, Savannah College of Art & Design fine arts grad from the December 2025 cohort, applied to 100 jobs and ended up with a single internship that ends in July. The internship’s exit conversation, per Rosser: “They told me they can’t promise anything.”
These are not three random cases. They are three different majors at three different selectivity tiers across three different parts of the country, and the pattern is identical: high application volume, near-zero conversion to paid roles, AI tools used both to apply and to hide the lack of conversion.
The executive quotes are the part that escaped containment
Three lines from Semafor World Economy made it into every secondary write-up by Friday evening.
Sen. Mark Warner (D-Va.): “Hair-on-fire moment. Boy, oh, boy.” Warner is predicting 30% recent-grad unemployment within two years, echoing ServiceNow CEO Bill McDermott’s March 13 forecast on CNBC that AI agents alone could push that number over 30%.
Chris Peterson, CEO of Newell Brands: the consumer-products company has stopped hiring anyone who isn’t “AI-proficient or AI-fluent.” This is the most operationally specific of the three quotes — Peterson didn’t say “we prefer AI-skilled candidates,” he said the company will not hire candidates who lack AI fluency. Newell Brands is a $7B revenue company. That hiring rule, if it spreads across the Fortune 500 the way the post-2008 GPA cutoffs did, becomes a structural exclusion.
Gary Cohn, IBM Vice Chairman: “Who in here can level a cement floor? Because you can earn $260,000 today to go to a data farm and be a cement layer.” Cohn — a former Goldman Sachs President and Trump-era National Economic Council director — was offering the line as practical career advice to graduates of, presumably, four-year degrees with student loans. The quote is doing a lot of work in the Semafor piece because Cohn is also chairman of an AI-adjacent firm that has spent the last 24 months talking about the white-collar productivity gains AI unlocks. The pivot from “AI augments knowledge workers” to “go pour concrete in West Texas” inside a single Q1-Q2 sentence has now been delivered by a senior IBM executive at a Washington podium.
The hiring-side data confirms the supply side
Indeed’s economic-research director Laura Ullrich told Semafor that the open lanes for 2026 grads are nursing, therapy, civil engineering — broadly, jobs requiring in-person physical presence, licensed credentials, or both. Indeed’s own postings data shows:
- LinkedIn applications-per-posting have roughly doubled vs. 2022 baseline.
- Big Tech entry-level hires are down >50% since 2019 (fortune.com).
- Gen Z’s share of Big Tech hiring is now 7%, down from 15% pre-pandemic.
- Total tech-sector layoffs YTD have crossed 100,000 (layoffs.fyi is the source Semafor cited).
The 100K cuts and the 7% hiring share are the two ends of the same vise. The cohort being squeezed is graduating into a market where Cloudflare just eliminated 1,100 jobs after attributing 600% AI-usage growth to internal automation, Bill Holdings cut 30% citing “AI-native rebuild”, and GM is openly posting AI-skills-swap reqs the same week as the layoff emails. The Class of 2026’s CS majors are applying into the same talent pool as the 100K experienced engineers who got cut over the prior 18 months. There are not enough chairs.
The counter-data IBM put forward is real but small
Nickle LaMoreaux, IBM’s Chief Human Resources Officer, told Semafor that IBM has tripled hiring for entry-level roles in 2026. Her reasoning: “Where are you going to get mid-career experience people five years from now if you haven’t groomed them?” Infosys gave a similar quote. The mid-career-pipeline argument is the most defensible counter to the AI-replaces-juniors narrative — somebody has to learn the codebase before they can supervise the agent that writes it — and IBM is one of the few companies operationalising it at scale.
LaMoreaux’s “tripled” claim deserves the asterisk that IBM’s entry-level base was already at a multi-year low — Krishna’s 2023 statement that 7,800 IBM back-office roles could be replaced by AI didn’t get unwound, and the tripling is off a depressed base, not a 2019 baseline. But the directional point is right: at least one large employer is choosing to absorb the cohort. The question is whether the playbook spreads or stays an outlier.
The thing the pamphlets never told them
The structural shift in the Semafor piece is the most important sentence of the year for the LostJobs editorial line: “Generations of young professionals were taught to go to college, get a 9-5 job, and climb the corporate ladder. The latest generation was inculcated to learn how to code, following the rote path of 1s and 0s that would land them a six-figure job at Google or Amazon, where they could scarf down free sushi and play ping pong in between work.”
The pamphlet was wrong. The 2017-2022 vintage of computer-science curricula was specifically optimised for the FAANG interview pipeline — DSA grinds, system-design rotations, behavioural prep books — and the pipeline closed in 2023 and was bricked over with AI agents by 2025. The kids who studied the right thing for the 2021 job market are graduating into a 2026 job market that doesn’t exist for them, and the universities have not updated curricula because curricula change on a five-year cycle and the AI shift happened in 18 months.
The grim irony — and the Semafor piece captures it — is that the tools the Class of 2026 is being displaced by are also the tools they are now learning to use to apply for the few remaining jobs. Hireraddi is paying $100/month for a Claude subscription to write the applications that an AI agent is reading to filter her out. That is the entire structure of the 2026 labour market in one transaction.
What to watch
- NACE’s June Class of 2026 first-destination survey. The National Association of Colleges and Employers publishes the canonical “what happened to last May’s graduates by June” dataset every August. The 2026 read will be the first one collected post-Cloudflare/Bill/Meta May-wave layoffs. If the 6-month-post-graduation employment rate for CS majors drops below the 80% historical floor, that becomes the wedge for a federal-policy conversation.
- Whether the Newell hiring rule spreads. “We don’t hire candidates who aren’t AI-fluent” is a one-line policy that any S&P 500 HR team can copy. If five more F500 CHROs adopt the same rule by July earnings, the cohort effect on liberal-arts and pre-AI-curriculum CS majors compounds.
- Cohn’s $260K cement-layer line. It will be a meme by Monday and a campaign line by August. The political economy of “skilled trades pay more than tech” is a story Democratic and Republican operatives will both try to claim. Watch for which one lands it cleanly first.
- IBM’s hiring numbers in the Q3 release. LaMoreaux’s “tripled entry-level” claim needs an absolute-headcount footnote. If IBM hired 300 entry-level engineers in 2025 and 900 in 2026 against a 280,000-person workforce, the gesture is real but the magnitude doesn’t move the cohort. The Q3 earnings supplemental is where the number gets disclosed.
5.8% recent-grad unemployment. 7.0% CS, 7.8% computer engineering. 100K tech-sector cuts YTD. 500 applications, two unpaid offers. The IBM Vice Chairman is telling four-year-degree graduates to lay cement. The Class of 2026 didn’t choose any of this; their pamphlets were written before the shift, their curricula were calibrated for a pipeline that closed, and the AI tools that ate the entry-level rung are the same tools they’re paying $100/month to use to apply to the remaining jobs.