What is really happening to jobs? Separating AI hype from reality – SIEPR policy brief (July 2026)

AI’s impact on overall employment is likely small right now

The brief reports little evidence that AI is causing significant job losses at present. Unemployment among workers in the most AI‑exposed occupations rose 0.77 percentage points since 2022, while the least‑exposed group saw a slightly larger increase of 0.85 points over the same period, indicating a broadly softening labor market rather than AI‑driven job losses. Employment trends in highly exposed occupations remain stable, and coding‑heavy occupations continue to show positive growth. Among firms that adopted enterprise AI, employment grew by 10 percent in the two years following adoption, driven by high per‑capita AI spending. Layoffs that cite AI are viewed with skepticism; many appear tied to cash‑flow needs for AI investment or post‑pandemic over‑hiring rather than direct automation.

A tough job market for new graduates may be partly due to AI

Recent graduates face unemployment of 5.6 percent in early 2026, up 1.6 points from three years earlier. The brief notes that entry‑level hiring in AI‑exposed occupations such as software development and customer service declined after ChatGPT’s launch, with older workers’ employment staying stable. Researchers describe these young workers as “canaries in the coal mine” for AI‑related disruption. However, the timing coincides with other macroeconomic shocks—interest‑rate hikes, pandemic over‑hiring, and remote‑work shifts—so isolating AI’s effect remains challenging. After controlling for confounders, notable declines in entry‑level hiring appear only from 2024 onward, when AI adoption and model capabilities had advanced.

AI’s impact on worker productivity is mixed but generally positive

Experimental studies show generative AI tools often boost performance most for less‑experienced or lower‑skilled workers. A call‑center study found a 15 percent overall productivity increase, with novice agents improving 30 percent in issues resolved per hour while highly skilled agents saw no gain. GitHub Copilot enabled software tasks to be completed 56 percent faster, with larger gains for less‑experienced programmers; other studies reported 10‑30 percent improvements depending on context. Writing tasks assisted by ChatGPT reduced completion time across skill levels and improved quality for low‑ability writers. Legal drafting and medical note‑taking also showed speed improvements, though medical AI scribes required physician oversight due to occasional inaccuracies.

The brief cautions that AI’s capabilities are “jagged”: usefulness depends on task fit, and misuse can harm performance. Less‑skilled entrepreneurs in a Kenyan field experiment earned lower revenues when following generic AI advice, whereas better performers extracted tailored suggestions. AI use in creative writing raised individual story quality but reduced diversity of outputs, and scientists using AI tools published more papers while studying fewer topics.

Potential frictions—limited share of tasks AI can speed up, process bottlenecks, and initial productivity dips during technology adoption—may prevent experimental gains from appearing in aggregate statistics yet.

Firm adoption has accelerated but unevenly across the economy

Surveys show wide variation in measured AI adoption, but all indicate rapid growth. The Census Bureau’s Business Trends and Outlook Survey estimates about 20 percent of firms use AI, while other measures report higher rates (e.g., over 80 percent of employees in an executive survey say they use AI at work). Adoption is concentrated in technology firms and information‑intensive sectors such as finance, sales, marketing, IT, strategy, and accounting. Even among adopters, AI use is often narrow—limited to one or two functions or low‑frequency use—with comprehensive integration remaining rare.

Most businesses remain in experimentation or piloting phases, with only a few large firms scaling beyond pilots. AI’s measured impact on employment is still minimal: only 5 percent of firms in Census data report any employment impact, with equal numbers citing gains and losses. Eighty percent of executives surveyed by the Federal Reserve Bank of Atlanta said AI investments have not yet altered headcount or improved productivity. A Danish study linking worker‑level AI adoption to firms found restructuring of tasks and time but no significant effect on employment, hours, or earnings. Nonetheless, among firms that adopted AI, employment grew by 10 percent in the two years following adoption.

Conclusion: Existing evidence is hardly the last word

The brief stresses that technological transformation typically lags behind innovation, citing the historic PC‑productivity gap. Current evidence suggests AI’s labor‑market effects are modest so far, but uncertainties remain about future scale and distribution. Policymakers should monitor evolving data while recognizing that AI’s influence may grow as adoption deepens and model capabilities advance.

Insights from Hacker News discussion

Commenters highlighted several nuances and skepticisms:

• One user noted that productivity gains from AI often follow a Pareto pattern, becoming more extreme as AI improves, while another warned that benefits concentrate on less‑experienced engineers and can turn negative for highly skilled workers, with LLMs moving performance toward the mean.

• A comment questioned the unemployment chart’s pre‑COVID trend, arguing that the AI exposure metric used in the figure reflects computer use circa 2015 rather than modern LLMs, which only emerged recently.

• Several observers pointed out that AI coding agents like Claude Code or OpenAI Codex only became reliably useful in late 2025, suggesting studies focusing on 2022‑2025 may miss a later uptick in impact.

• Some remarked that AI’s current utility is “broad but shallow,” with self‑reported AI use averaging only about 6 percent of work hours, yet yielding large per‑hour time savings when used.

• A few commenters expressed concern that layoffs citing AI may serve as a convenient excuse for cost‑cutting unrelated to technology, driven by short‑term profit incentives.

• Others noted organizational inertia, with some Fortune 500 firms maintaining internal bans on AI, and that real productivity gains are unevenly distributed across teams.

These community reflections echo the brief’s themes of uneven adoption, mixed productivity effects, and the difficulty of isolating AI’s influence amid broader economic forces.