Stock markets are pricing in a permanent 32.6% productivity gain for software engineers from AI, according to a new working paper by economists at UC Berkeley and the London School of Economics, The Register reported Tuesday. The authors translate that into a 3.61% permanent increase in the level of GDP.

What Happened

The paper, "The Macroeconomic Effect of AI: Sizing the Software Engineering Channel," was published through the National Bureau of Economic Research. Its authors are Alex Blumenfeld, Chen Lian and Andreas Schaab of UC Berkeley and Jonathon Hazell of the LSE.

Rather than measuring output directly, the researchers looked at how share prices reacted to AI-related news and tested whether companies with a larger share of payroll going to software engineers saw bigger gains. They then fed those market reactions into an economic model to infer the productivity change investors appear to expect.

RelatedNew curl Release Fixes 24 Flaws, Including a High-Severity Bug Reported With Anthropic’s Claude

Over the period from November 2022 to December 2025, they estimate markets imply a permanent 32.6% rise in software engineer productivity. The start of that window coincides with OpenAI's release of ChatGPT at the end of November 2022, which set off the current wave of generative AI investment.

The Register noted that the figure lines up with task-level studies by other researchers, which have reported speedups of 21% to 56% for programming work done with AI assistance. The method does not depend on employment trends; it reflects forward-looking investor judgments about productivity at the firms measured.

Why It Matters

Measuring AI's effect on productivity has been difficult. Company surveys and vendor studies often report large gains, while some controlled studies have found little benefit or even slowdowns. A widely cited 2025 trial by research group METR, for instance, found that experienced open-source developers took longer to complete tasks when allowed to use AI tools, despite believing they were faster.

A market-based approach offers a different kind of evidence. Stock prices aggregate the expectations of many investors, and tying reactions to each firm's reliance on engineering labor isolates one specific channel through which AI could affect the economy. Software engineering is a natural place to start because coding assistants are among the most widely adopted business uses of generative AI.

The Caveats

Chen Lian cautioned that "markets can be overly optimistic or pessimistic," The Register reported, stressing that the estimate reflects what investors expect rather than gains that have actually materialized. If AI enthusiasm has inflated valuations, the implied productivity figure would be inflated along with it.

The Register also pointed out that improvements at the level of individual tasks do not automatically carry through to whole teams. If developers produce more code but review processes cannot keep pace with the volume of commits, the bottleneck shifts rather than disappears, and realized gains fall short of the task-level numbers.

What to Watch

The key test is whether measured output data eventually confirms what markets are pricing. Researchers will be looking for evidence in firm-level productivity statistics, software release rates and national accounts. A sharp repricing of AI-exposed stocks would also change the paper's inputs, which is why the result is best read as a snapshot of market expectations over a fixed window rather than a forecast.