Hype Index: 'AI Killed Junior Developer Hiring'

TL;DR: US software development job postings are up 14.6% in the 12 months since Section 174's forced amortisation of software salaries was repealed, against a 2.5% decline in all postings, which makes software the 4th fastest growing of the 41 categories Indeed tracks. The three above it are mechanical, industrial and electrical engineering: the other R&D payrolls that same tax provision hit. I read 22 articles asserting AI caused the software-hiring decline and not one mentioned Section 174, or any tax explanation at all. That is not evidence the tax code did it. The repeal was retroactive to 1 January 2025 and agentic coding tools shipped in the same window, so the two explanations occupy identical dates and no posting series can separate them. The honest finding is that the timing fits the tax change at least as well as it fits AI, and almost nobody checks.

What is the claim?

That AI ate the junior developer job. It is everywhere, and the strongest version comes from Indeed Hiring Lab itself, which reported in July that US software development postings had risen roughly 15% since Claude Code launched in late February 2025 while overall postings fell 7%, and framed the whole reversal as AI moving "from job destruction to job creation."

That is a causal story with a start date. It has a competitor with a start date, an end date, and a public law number, and Indeed's post does not mention it.

What is Section 174?

A 2017 provision that took a long time to bite. Per IRS Notice 2023-63, section 13206(a) of the Tax Cuts and Jobs Act amended section 174 for "taxable years beginning after December 31, 2021," disallowing the immediate deduction of research and experimental expenditure and forcing companies to amortise it over "a 5-year (60-month) period" for domestic research and 15 years for foreign. The same amendment added section 174(c)(3), which requires that "any amount paid or incurred in connection with the development of any software" be treated as a research expenditure.

Read that last clause slowly. From January 2022, a US company paying an engineer $200,000 to write software could deduct $20,000 of it in year one. The cash went out; the deduction did not. For a company with thin margins and a large domestic engineering payroll, that is a tax bill on money it no longer had.

It was reversed by 26 U.S.C. 174A, enacted 4 July 2025 as Public Law 119-21, which restored the immediate deduction for domestic research and applies to "taxable years beginning after December 31, 2024." So calendar-year filers got it back for all of 2025, retroactively. Twelve months of post-repeal data now exist. Nobody who writes about junior developer hiring appears to have looked at them.

What do the postings actually show?

I pulled the raw index CSVs from Indeed Hiring Lab's public tracker and read the software development series at four dated markers, monthly-averaged, against the all-postings index as a control. Both are indexed to February 2020 = 100.

Marker	Date	Software Development	All US postings	Ratio
Amortisation takes effect	Dec 2021	213.9	159.0	1.35
Amortisation biting	Jan 2023	126.3	140.3	0.90
Repeal signed	Jul 2025	65.5	107.1	0.61
12 months post-repeal	Jul 2026	75.0	104.4	0.72

Software fell 41% between the first two markers while the whole market fell 12%, then fell another 48% while the market fell 24%. It underperformed the control for three and a half years. In the twelve months after the repeal it rose 14.6% while the control fell 2.5%, the first sustained divergence in its favour since 2021. The series bottomed on 17 May 2025 at 61.1 and stood at 73.7 on 7 August 2026.

Then I ranked all 41 occupation categories by change from July 2025 to July 2026. Software Development came 4th at +14.6%. Above it: Mechanical Engineering +29.2%, Industrial Engineering +23.6%, Electrical Engineering +15.6%. Scientific Research and Development came 8th at +10.5%. The top four risers, and five of the top eight, are the payroll categories section 174 forced onto a five-year amortisation schedule. Accounting fell 13% and childcare fell 15.9% in the same window.

How many articles mention it?

I collected articles asserting that AI caused or drove the software-hiring decline and searched each for any reference to section 174, section 174A, R&E amortisation, the TCJA, or any tax explanation at all. I fetched 24 pages, excluded two (one paywalled past the preview, one a layoff list making no causal claim), and coded the remaining 22 articles across 20 distinct domains.

Zero mentioned Section 174. Zero mentioned tax policy of any kind. The two pages containing the word "tax" used it for an unrelated sponsored ad and for lost municipal tax revenue.

The sample includes Indeed Hiring Lab, Forbes, CIO, Dice, Stack Overflow's blog, dev.to, Tech Times, Final Round AI, daily.dev, SoftwareSeni, byteiota, herohunt.ai, Pin, tech-insider.org, Kore1, SkillSyncer, Joberty, RezScore, Scrimba Guide and Standout. The one writer who covered the provision properly, Gergely Orosz's 2023 explainer, was excluded from the count because his 2026 job-market piece sits behind a paywall I did not read past.

What does this not prove?

That Section 174 caused any of it. I want to be blunt about how much this evidence cannot carry.

The software index peaked on 28 February 2022, two months after amortisation took effect, and the all-postings index peaked a month later. Both turned when the Fed started raising rates. If 174 were the mechanism, the top should not have arrived after the start date, and the control should not have turned at the same moment. The trough also arrived in May 2025, before the bill was signed, though the House passed it that month and the effective date reaches back to January.

And that retroactive effective date is the fatal problem for both stories. Section 174A applies from 1 January 2025. Claude Code shipped in late February 2025. The post-repeal window and the agentic-coding window are the same window. No posting series distinguishes a company hiring engineers because their salaries became deductible again from one hiring engineers because AI made each engineer more valuable. Postings are also not hires: a posting is an intention, and the productivity research that would settle the question is mostly vendor-run.

I also dropped two figures I could not trace. A widely repeated "R&D sector down 14,000 jobs" and "R&D spend growth fell from 6.6% to under 0.5%" appear in search summaries with no primary release behind them that I could reach, so they are not in this post.

Verdict

The AI story explains a decline that started before the tools existed and reversed while the tools got better, which is a lot of work for one variable. The tax story explains a decline that began the year the law bit and a recovery that began the year it was repealed, in exactly the occupation categories the law named, and it cannot rule out the AI story either.

Verdict: 70% noise / 30% signal. The signal is that AI changed what a junior developer job contains. The noise is 22 articles out of 22 explaining a labour-market movement without checking whether Congress moved it. That is the same failure mode as reading a rising share as a rising number, and the same reason most AI layoff announcements survive no scrutiny at all: one explanation is available, so nobody goes looking for the second.