59% of Hiring Managers Admit the AI Excuse. Source the Section 174 Window.

If you've spent the last 18 months watching your junior pipeline evaporate and listening to executives explain it as "AI is doing that work now," you've been sourcing against the wrong story. Indeed Hiring Lab's April 23, 2026 data shows junior postings down 7% year over year while senior postings rose 4%, and a Resume.org survey of 1,000 hiring managers caught the quiet part out loud: 59% admit they emphasize AI in layoff and freeze announcements because it plays better with stakeholders than citing financial constraints. The freeze is a tax-and-rates story wearing an AI costume, and the costume is about to come off.

The narrative versus the ledger

Here is the story executives have told for two years: large language models ate the junior engineer job. Here is what the numbers actually say.

Challenger, Gray & Christmas tracked roughly 1.1 million announced job cuts in 2025. AI was cited as a driver in about 55,000 of them. That is under 5% of the total, dressed up as the entire explanation. The Federal Reserve's FEDS Note from Liu and Webber (March 27, 2026) examined whether AI adoption reduced job postings and reported "precisely-estimated null effects." Not "small effects." Null. The Dallas Fed's January 2026 paper by Atkinson and Yamco found that even if every drop in young-worker employment in AI-exposed occupations translated directly into unemployment, it would account for a 0.1 percentage point rise since November 2022.

So what actually happened?

Two things, neither of which makes a good all-hands slide. First, the 2022 implementation of Section 174 amortization forced U.S. companies to capitalize software developer salaries over five years for domestic R&D (15 years for offshore), turning every junior engineer into a five-year tax liability instead of a current-year expense. Second, the Fed ran the fastest rate-tightening cycle in 40 years right into the back of a ZIRP-era hiring binge during which Meta grew from roughly 45,000 to over 86,000 employees in three years and Alphabet went from 119,000 to over 190,000.

That is the freeze. It is a hangover and a tax code, not a model.

59%

What the OBBBA actually changed

The One Big Beautiful Bill Act, signed July 4, 2025, repealed the TCJA capitalization requirement for domestic R&E expenditures. Starting in tax year 2025, companies can immediately expense U.S.-based engineering salaries again, and eligible small businesses can amend prior-year returns to claw back the cash they parked with the IRS from 2022 to 2024.

This matters in two ways most sourcers and engineering managers have not internalized.

First, the "section 174 software developer tax" that quietly broke entry-level hiring is gone for domestic R&D. The CFO who told you in Q2 2024 that she could not justify a 12-person new-grad cohort because each hire created a multi-year deferred deduction now has the opposite incentive. Hiring a junior in Austin in fiscal 2026 is a current-year deduction. Hiring the same junior through an offshore vendor in Hyderabad still amortizes over 15 years.

Second, the carve-out is the entire game. The tax code now actively rewards onshore junior hiring relative to offshore contracting, which is the exact opposite of what it did from 2022 through 2024. If you have been pitching candidates against an offshoring threat, the macro has flipped under you.

What this looks like in a sourcing conversation

Stop leading hiring managers with "we have a great pipeline of bootcamp grads." Lead with: your CFO can now expense this hire in-year, and the offshore alternative cannot. That sentence opens reqs that have been "frozen" for 14 months.

The lost class

Recent graduate unemployment hit 5.7% in Q4 2025, worse than any point during the 2008 financial crisis. Stanford CS Professor Jan Liphardt told the LA Times his students "are struggling to find entry-level jobs" in "a dramatic reversal from three years ago." Indeed Hiring Lab's Laura Ullrich on NPR: "the probability that you're going to be laid off is very low. However, if you are unemployed, it is much more difficult to find a job in 2026 than it was a couple years ago."

Tech internship postings drew 273 applications each at the trough. That is not a market signal about productivity. That is a logjam.

The recruitable population this creates is the 2024, 2025, and early 2026 CS grad cohort, plus the laid-off junior ICs from the 2022 to 2023 over-hire wave who have been freelancing, contracting, or grinding LeetCode through a year-plus gap. They are not in your ATS. They are not posting on LinkedIn because the algorithm punishes "open to work" and they know it. They cluster in Hacker News "Who Wants to be Hired" threads, university ACM Discord servers, GitHub contribution graphs that quietly kept ticking through unemployment, and the kaggle/HuggingFace/arxiv long tail.

Finding them by keyword is a losing game. The signals are behavioral: a 2024 graduation date plus a personal-site portfolio updated in the last 60 days plus open-source commits to a non-trivial repo plus a current employer field that says "Self-employed" or is blank. That is six filters, two of which do not exist as filters anywhere.

This is the kind of query where structured boolean breaks down and where Refolk earns its keep: you describe the person in plain English ("CS grads from 2024 or 2025 with active GitHub in the last 90 days, currently unemployed or freelancing, based in the U.S., comfortable with Python or TypeScript") and get a ranked shortlist across GitHub, LinkedIn, and the open web.

Programmers are the exception, and that's the point

Be honest about the data that complicates the thesis. Federal Reserve research linking O*NET task data to CPS finds employment growth among programmers has decelerated sharply since ChatGPT's launch. Coder employment kept growing, but much more slowly than pre-2022.

Junior CS is simultaneously the most AI-pressured occupation and the most tax-distorted one. Both are true. That is precisely why the window is narrow and why getting in front of the snapback matters. The companies that figure out the 174 math first hire the strongest of the lost class first. The ones still running the 2024 playbook in mid-2026 will be paying 30% premiums in 2027 to replace people they could have hired now.

The freeze is a hangover and a tax code, not a model. The costume is about to come off.

The dot-com analog beats the Industrial Revolution analog

Every executive memo on AI and labor reaches for the Industrial Revolution. Wrong frame. The right frame is the dot-com bust: a steep rate cycle, an entry-level hiring freeze, a convenient technological scapegoat (offshoring then, AI now), and a rapid return to pre-crash hiring levels by 2004 once rates came down.

Indeed Hiring Lab's April 2026 piece notes that demand for interns "picked up at the start of this year and is in line with or even slightly ahead of 2024 levels so far." That is the leading indicator. Internships convert to new-grad reqs 12 to 18 months out. The buy side is already moving. The 92% of companies that told Resume.org they plan to hire in 2026, with 86% hiring in Q1, are not hypothetical.

7%

What to do in the next 90 days

Re-open the reqs you were told to kill

Take the entry-level reqs your finance team froze in 2023 or 2024 with the words "Section 174" anywhere in the rationale and walk them back to your CFO with the OBBBA repeal in hand. If they were frozen for tax reasons, they should re-open for tax reasons. The rationale that closed them is gone.

Build the lost-class list before competitors do

The named pool is concentrated. Refolk index data on U.S. software engineers with Python or JavaScript skills shows over 92,000 professionals, concentrated in the SF Bay Area, NYC, Austin, and Salt Lake City, with current employers including Vercel, Databricks, Glean, and LinkedIn. That is the senior comparison set. The junior set you actually want is the layer beneath, the people who never made it to those logos because the 2023 to 2024 freeze landed on their graduation date.

A plain-English query like "software engineers who graduated from a U.S. CS program in 2024 or 2025, currently freelancing or unemployed, with public GitHub activity" is exactly the shape of search that grinds boolean to dust and that the Refolk product is built for. Run it weekly. The list shifts as people take and lose contract work.

Reframe the pitch

Senior devs distrust AI hype more than they trust it. Junior devs who survived the 2024 to 2025 market are even more skeptical. Pitching them as "AI-native" replacements for the mid-levels you laid off is a known-loser script.

The pitch that works in 2026 is honest: we are hiring you because the tax code stopped punishing us for hiring you, the work has not gotten smaller, and the people we laid off in 2023 were collateral damage from a rate cycle, not productivity gains. Kara Dennison at Resume.org put the corollary cleanly: "AI has become an explanation because it sounds strategic and forward-looking. But when AI is used as a blanket explanation and workloads do not meaningfully change, trust erodes quickly." Junior candidates know workloads have not shrunk. Engineering managers know it. Talk to them like adults.

Source where the lost class actually is

Not LinkedIn job search. Specifically:

• Hacker News "Who Wants to be Hired" threads, especially the May and August 2025 editions.
• University ACM chapter Discords and CS department mailing lists (Stanford, Howard, CMU, Georgia Tech, UT Austin, UIUC).
• GitHub contribution graphs filtered for personal-project density over the last 12 months, employer field blank or self-described.
• Kaggle competition top decile from 2024 to 2025.
• Personal portfolio sites surfaced from the open web rather than from indexed job boards.

The reason to use a tool like Refolk for this rather than a stack of seven separate scrapers is that the candidate signal is split across surfaces. A 2024 grad's graduation year lives on LinkedIn. Her actual code lives on GitHub. Her current availability lives on her personal site or in a Hacker News comment. Stitching those three together by hand is the entire job, and it is the job worth automating.

The window

The OBBBA fix took effect in 2025. Internship demand is already rebounding into 2026. The 2024 to 2025 grad cohort is still mostly outside the loop. Rate-cut expectations are rebuilding. By late 2026, the consensus view, that 36% of companies will have frozen entry-level hiring and that 47% expect to halt it by 2027, will look as dated as the 2023 "AI will replace 300 million jobs" headlines look now.

You have roughly two to four quarters before the smart end of the market reprices junior engineers. The freeze was a tax story. The thaw is too. Source accordingly.

FAQ

Is AI actually replacing junior software engineers?

The honest answer is "a little, less than claimed, and not enough to explain the freeze." Federal Reserve and Dallas Fed research find precisely-estimated null effects on job postings overall and, at most, a 0.1 percentage point unemployment effect from young workers in AI-exposed occupations. Programmer employment specifically decelerated since ChatGPT, but kept growing. Meanwhile 59% of hiring managers admit they emphasize AI in their explanations because it sounds better than "rate cycle and Section 174." Both effects are real. The AI effect is roughly an order of magnitude smaller than the macro effect.

What is Section 174 and why does it matter for hiring?

Section 174 of the U.S. tax code, as amended by TCJA in 2017 and effective from 2022, required companies to capitalize and amortize R&D expenses, including software developer salaries, over five years for domestic work and 15 years for offshore work. That converted every junior engineer hire into a multi-year tax liability. The OBBBA, signed July 4, 2025, repealed the domestic capitalization requirement starting in tax year 2025 and allowed small businesses to amend prior returns. Domestic junior hiring is now expense-favorable and offshore is not.

How do I find the "lost class" of 2024 to 2025 CS grads?

Not on LinkedIn job search, where the cohort is hidden behind algorithmic deprioritization of "open to work" signals. Use behavioral signals: graduation date plus blank or self-employed current employer plus active GitHub plus a personal site updated recently. The cohort clusters in Hacker News hiring threads, university ACM Discords, and the long tail of personal portfolios. Tools that index across GitHub, LinkedIn, and the open web in a single query, including Refolk, are built for exactly this shape of search.

How long does this arbitrage window last?

Probably two to four quarters. Internship postings, the leading indicator, are already at or above 2024 levels in early 2026. Rate-cut expectations are rebuilding. Once CFOs at midsize tech companies fully model the OBBBA savings into 2026 headcount plans, the new-grad market reprices. The 2002 to 2004 dot-com analog suggests the snapback happens faster than the freeze did.