- Software engineer job postings are 49% below their early-2020 baseline — the steepest proportional decline among tracked white-collar sectors — while overall US postings are 10% above that same baseline.
- The labour market has bifurcated: ML engineer postings are up 59% from 2020, roles requiring AI skills command a 28% salary premium, while junior and mid-level software roles have contracted sharply.
- Controlled experiments show AI productivity gains are real but uneven: junior developers see 35–39% speed-ups on structured tasks; experienced developers on real-world codebases took 19% longer with AI tools than without.
Key Claim: The developer hiring contraction is a segmentation story, not a replacement story — the lower rungs of the career ladder have contracted sharply while the premium for AI-fluent senior engineers has risen, compressing the pipeline that develops the next generation.
Software engineer job postings in the United States have fallen 49% below their early-2020 baseline as of July 2025, according to Indeed Hiring Lab — the steepest proportional decline among tracked white-collar sectors. For context, overall US job postings are 10% above the same baseline. The gap is not noise. As of February 2025, Indeed data analysed by The Pragmatic Engineer showed marketing postings down 19% and banking and finance down 7% from their 2020 levels. Software development, then already down 34%, has since deteriorated further.
The standard narrative attributes this to artificial intelligence (AI). That attribution is partly right, partly premature, and — if taken at face value — obscures a more useful structural read on what has changed and what has not. The hiring contraction is real. Its causes are layered. And the data, assembled from Indeed Hiring Lab, a Stanford Digital Economy Lab study, the Stack Overflow Developer Survey, and controlled experiments from METR and MIT Economics, tells a more specific story than “AI is replacing developers.”
The market has segmented. And understanding where the segmentation falls matters more than the aggregate headline.
The Postings Data: A Five-Year Low With Uneven Distribution
The most comprehensive public series for US tech hiring comes from Indeed Hiring Lab. As of July 2025, the overall tech posting index sits 36% below the January 2020 baseline.
Within that aggregate, the distribution is uneven. Software engineers are down 49% from early 2020. Specialised roles — Android developers, Java developers, .NET developers — are down more than 60% from their 2022 peaks. By contrast, machine learning (ML) engineers are up 59% from early 2020, despite having also fallen 47% from their own 2022 peak.
This is not a uniform contraction. It is a rotation.
The timeline adds context. Nearly half of the net decline from the 2022 peak occurred before ChatGPT launched in late November 2022. The first major leg down was driven by macroeconomic correction — a reversal of the over-hiring that characterised 2021 and early 2022.
The Experience Gap: Where the Labour Market Is Really Moving
The sharpest signal in the data involves experience-level segmentation. Indeed Hiring Lab reported in July 2025 that the share of tech postings requiring five or more years of experience rose from 37% to 42% between Q2 2022 and Q2 2025.
The Stanford “Canaries in the Coal Mine” paper by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, using payroll data from ADP (25 million workers), found that employment for software developers aged 22–25 fell nearly 20% from its late-2022 peak to July 2025. Over the same period, employment for workers aged 30 and over in the same AI-exposed occupations grew 6–12%.
The authors offer a structural explanation: AI tools substitute most effectively for “codified knowledge” — the kind acquired through formal education and applied in well-defined, repeatable tasks. Tacit knowledge, systems-level context, and accumulated judgment are harder to replicate.
The Productivity Question: What Controlled Experiments Actually Show
A field experiment by Cui et al., run across Microsoft, Accenture, and one Fortune 100 company, randomised 4,867 developers into AI-tool and control groups. Developers using AI tools completed 26% more tasks on average. But the distribution was skewed: junior developers saw 35–39% speed-ups, while senior developers saw only 8–16%.
A different result came from a METR study published in July 2025. Sixteen experienced open-source developers working on repositories averaging 22,000 GitHub stars were randomly assigned to real GitHub issues with or without AI tool access. The finding: developers with AI tools took 19% longer to complete issues than those without. The METR authors are careful with scope: this finding applies to experienced developers working on unstructured, real-world issues in large codebases. For teams already familiar with how AI coding assistants actually perform at the enterprise level, these trust numbers will not come as a surprise.
What the Skill Premium Signals
The salary data from Lightcast’s 2025 Global AI Skills Outlook, based on analysis of 1.3 billion job postings, shows that roles requiring AI skills command a 28% salary premium — approximately $18,000 more annually — compared to equivalent postings without that requirement. This is the wage signal that accompanies a genuine structural shift: the market is repricing developer skills, not just reducing headcount.
GitHub Copilot’s deployment data confirms the scale of adoption. By July 2025, Copilot had reached 20 million users, with 4.7 million paid subscribers as of January 2026 — a 75% year-on-year increase. Approximately 90% of Fortune 100 companies have deployed it. The broader picture of how enterprises are measuring returns on this kind of AI investment is examined in our analysis of enterprise AI ROI across industries.
What the Leading Indicators Suggest
Experience segmentation is likely to deepen. The shift from junior-heavy to senior-heavy job requirements is consistent with companies extracting productivity from AI tools applied by experienced engineers, rather than onboarding large cohorts of new graduates for implementation work.
ML and AI-adjacent roles remain the exception. The Indeed series shows ML engineers are the only major software category above 2020 posting levels. Lightcast data indicates demand for AI skills is spreading beyond tech.
CoderPad’s 2026 State of Tech Hiring data shows US technical hiring activity is up 90% compared to mid-2023, and 53% of talent leaders expect hiring budgets to increase in 2026 — the highest figure in years.
A Structural Shift, Not a Replacement Story
The data does not support a replacement narrative. Total tech employment in the US ticked down only 2% in 2024 while job postings fell far more sharply. The Bureau of Labor Statistics still projects employment of software developers to grow 17% from 2023 to 2033 — much faster than the 4% average for all occupations — with roughly 189,200 openings per year over the decade.
What the data does support is a labour-market segmentation story: the lower rungs of the career ladder have contracted, the premium for AI-fluent senior engineers has risen, and the traditional arbitrage of hiring junior developers for rote implementation work has diminished. The career ladder compression is the structural risk worth watching, not the absolute headcount. An industry that struggles to absorb and develop early-career engineers will, over time, face a shortage of the experienced engineers it currently prizes.
This article was produced with AI assistance and reviewed by the editorial team.
