AI made coding easy for everyone. The labor market for software engineers has not adjusted gracefully.

By [Byline]

On the May 1, 2026 episode of Bloomberg’s Wall Street Week, host David Westin framed the central paradox of this technology cycle in a single sentence: “AI made coding easy for everyone, and that’s the best and worst thing to happen to software engineering” (Bloomberg, 2026). The line landed because both halves are now demonstrably true. Productivity for engineers using AI tools has measurably increased. The job market for the people who would have been hired to do that work has measurably contracted. The economic question facing the labor market is no longer whether AI is real, but how an industry that grew on the assumption of perpetual hiring adjusts to a new equation in which output rises while headcount does not.

A Labor Market That Bifurcated

The headline numbers in software engineering still look benign. The Bureau of Labor Statistics projects employment for software developers, quality assurance analysts, and testers will grow 15.8% between 2024 and 2034, adding 267,700 net new positions, with about 129,200 openings projected per year on average over the decade (Bureau of Labor Statistics, 2025). For comparison, the BLS forecasts 3% total job growth for all U.S. occupations over the same period.

But aggregate forecasts conceal what is happening at the entry level. According to Federal Reserve data cited by labor analysts, computer science graduates now face a 6.1% unemployment rate, compared with 3.2% for philosophy majors, 3.0% for art history graduates, and 4.4% for journalism majors (Federal Reserve via Lalith Manage, 2026). Entry-level positions in software roles have seen what one analysis described as a 73% decrease in hiring rates over the past year, against a 7% decrease across all job levels (Lalith Manage, 2026). Layoffs continue to compound the picture: more than 126,000 tech workers were laid off in 2025, and rounds at Microsoft, Amazon, Meta, and Google have continued through early 2026 (SaaStr, 2026).

SaaStr, surveying CEO sentiment for 2026, found 66% of leaders said they planned to either reduce or maintain the size of their teams; only one-third indicated plans to hire (SaaStr, 2026). The U.S. unemployment rate sat at 4.6% in early 2026, the highest level in four years, with Indeed economists expecting it to remain near that level through year-end (SaaStr, 2026; Indeed, 2026).

The Productivity Side of the Ledger

The reason hiring has paused is not that AI failed. It is that AI worked. At Google, internal data shared in a March 2026 New York Times feature indicated that engineering velocity had increased on average by approximately 10%, with some categories of tasks completing tens of times faster (New York Times, 2026). At early-stage startups, founders interviewed for the same article reported that nearly 100% of their code is now AI-generated; at Google, the share is below 50% (New York Times, 2026). A 2am.tech industry analysis reported AI-generated code accounts for over 70% of new code at many startups and over 40% at large enterprises (2am.tech, 2026).

The 2025 Stack Overflow Developer Survey found that 84% of developers use or plan to use AI coding tools (Stack Overflow, 2025). Cursor, an AI-native code editor, crossed $500 million in annualized recurring revenue in late 2025, a milestone Slack required four years to achieve (2am.tech, 2026). Anthropic, whose Claude models power a growing share of professional coding workflows, is reportedly weighing a funding round at a valuation above $900 billion (Bloomberg Technology, 2026). The economic implication is straightforward: if a single engineer can now do work that previously required two or three, and if 66% of CEOs are planning not to backfill departures, the gap between projected job growth and realized hiring will widen. The BLS projection of 15.8% growth was constructed before AI coding tools matured at their current rate.

Where the Spending Is Going

Capital is flowing to a different set of recipients. Bloomberg’s coverage of the April 2026 tech earnings showed Alphabet, Amazon, and Microsoft all delivering strong cloud growth tied to AI demand, while Meta lagged precisely because it could not show the same return-on-AI-spending metric (Bloomberg Technology, 2026). Alphabet posted Google Cloud growth of 60% and saw its stock reach a record high. Amazon Web Services demonstrated similar momentum, with Bloomberg’s analysts attributing the gains to a “flywheel effect” between vertically integrated infrastructure and AI demand (Bloomberg Technology, 2026).

The same earnings reports also showed an extraordinary acceleration in capital expenditures. Combined 2026 capex commitments from Microsoft, Alphabet, Meta, and Amazon now exceed approximately $650 billion, with the funds directed primarily at GPUs, data centers, and electricity infrastructure rather than at software headcount (Yahoo Finance, 2026). The companies, in other words, are spending heavily on the substrate that lets fewer engineers do more.

The “Invisible Unemployment” Problem

Beneath the headline unemployment rate, the texture of the technology labor market has shifted. SaaStr describes the phenomenon as “invisible unemployment”: layoffs that no longer make headlines, and jobs that simply do not materialize. IBM’s voluntary attrition rate dropped from a normal 7% to under 2% in early 2026, a three-decade low in an industry where annual turnover typically runs between 13% and 21% (SaaStr, 2026). When workers stop quitting and companies stop hiring, the unemployment rate can stay roughly stable while the labor market becomes substantially less fluid for those at the entry point.

The composition of demand has also shifted. Specialized roles in AI engineering, machine learning operations, and large-language-model architecture have grown rapidly, with Robert Half reporting that 87% of tech leaders face challenges finding skilled workers (Robert Half, 2026). The same labor market produces a 6.1% unemployment rate among CS graduates and an 87% leader-reported skills shortage. The contradiction resolves at the level of skill mix: companies want engineers who can architect AI-augmented systems and operate fluently across foundation models, not engineers who write conventional code by hand (SaaStr, 2026).

Counterweights and Caveats

The picture is not uniformly negative. AI coding tools have raised the productivity floor for solo developers and small teams, allowing founders without venture funding to ship products that would previously have required hires (Pragmatic Engineer, 2026). For experienced engineers who have adopted the tools effectively, productivity gains in the 30% to 50% range are routinely measured, with task-specific gains for documentation, testing, and refactoring reaching 50% to 80% (devstarsj, 2026). Cost is also an emerging constraint: companies pay roughly $100 to $200 per month per engineer for “max” plans on Claude Code, Cursor, and Codex, with around 30% of engineers now hitting usage limits (Pragmatic Engineer, 2026). Subsidies from venture-funded AI providers are propping up enterprise pricing in the short term. When those subsidies run out, the economics of AI-augmented engineering will shift again.

The Macroeconomic Question

For the broader economy, the central question is whether the productivity gains in software engineering translate into wider growth or whether they concentrate in a handful of companies and their shareholders. Federal Reserve research has long held that productivity growth, even when concentrated, eventually diffuses through the economy via lower prices, new product categories, and the redeployment of displaced labor. The disagreement is over the timeline. The current cycle is unusual because the displaced labor is high-skilled, well-paid, and concentrated in a sector that has, until recently, been a primary source of college-graduate wage growth.

If software engineering continues to bifurcate (high salaries for AI-augmented seniors, no offers for mid-pack juniors) the economic effect will be to compress the wage distribution within the profession while raising it across the economy. Whether that produces a net welfare gain depends on what the cohort that would have become software engineers does instead, and on whether the productivity surplus gets reinvested in domestic capital or repatriated to corporate margins. As Westin observed in his framing for the May 1 episode, AI is the best and the worst thing to happen to software engineering at the same time. The economy is now sorting through which half it gets first.

Sources Cited