The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2505
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T07:44:36.434620Z
status: published
lens: g_space
sector: business
headline: The Macroeconomic Effect of AI: Sizing the Software Engineering Channel
dek: We measure how artificial intelligence (AI) affects the economy through its impact on software engineering productivity. We use information from financial markets to develop a forward-looking measure that is available in real time. We estimate the sensitivity of each firm’s stock return to an AI stock market index, and how this sensitivity depends on the share of firm payroll in software engineering. We use a model to map this cross-sectional relationship into software engineering productivity gains. From Novemb…
gain_title: In a working paper using stock-return data, economists estimate that AI news from November 2022 to December 2025 raised markets' expected value of software engineering productivity by the equivalent of a permanent 32.6% increase, implying a 3.6% higher GDP level.
problem_title: (none)
trace_subject: (none)
gain_reading: In a working paper using stock-return data, economists estimate that AI news from November 2022 to December 2025 raised markets' expected value of software engineering productivity by the equivalent of a permanent 32.6% increase, implying a 3.6% higher GDP level.
gain_evidence: the equivalent of a permanent 32.6% productivity increase | The corresponding effect on the level of GDP is 3.6% in the baseline and 6.5%
problem_reading: (none)
problem_evidence: (none)
quick_read: Four economists from Berkeley and the London School of Economics used stock-market reactions to estimate how much AI has raised, or is expected to raise, software engineering productivity. From November 2022 to December 2025, AI news was equivalent to a permanent 32.6% productivity increase for software engineers. That implies a 3.6% higher level of GDP in their baseline model.

The GDP figure rises to 6.5% when faster software engineering also speeds up research and development. The authors say the effect on productivity and GDP had more than doubled by mid-2026, as coding agents such as Claude Code and Codex spread.

The numbers are market expectations, not measured output. The method gives a real-time, forward-looking reading, but it depends on how closely investors' beliefs match what happens. For scale, the authors note that earlier GDP forecasts range from about one percent over a decade to scenarios with growth rates many times higher.
limitation: The estimate reflects what financial markets expect, and the authors say that view is unlikely to be perfectly accurate. The Revelio resume data also over-represents management and computer occupations.
tag: Evidence-backed gain
key_points: The method compares how firms' stock returns track an AI stock index (THNQ) with the share of each firm's payroll that goes to software engineers, using Revelio Labs resume data and Fama-French risk controls. | In the baseline specification, two firms whose software engineering payroll shares differ by one percentage point show a return gap of 1.24 basis points when the AI index moves one percentage point; other specifications gave 1.70, 1.81 and 1.91. | The sample covers 2022 to 2025 and excludes firms that sell software or sit in the semiconductor supply chain, including 64 semiconductor-related firms. It keeps software-using firms such as eBay, Sonos, Expedia and Airbnb. | The 32.6% estimate is close to the 21–56% task-level speed-ups reported in earlier experiments. The authors say expected future gains push the figure up, while bottlenecks between tasks pull it down. | The baseline GDP effect is 3.61%, and it nearly doubles when software engineering productivity also raises R&D productivity. | Expected productivity gains driven by news in the first half of 2026 were larger than those driven by the previous three years combined, a period the authors tie to the spread of coding agents. | Earlier estimates differ widely: one study puts TFP gains at 0.53–0.66% over a decade, another at 6.8–13%. The authors put the software engineering channel alone at 3.61% by the end of 2025.
rundown: The authors reason that if AI makes software engineers more productive, firms that spend more of their payroll on engineers should gain more profit, and stock prices should react more strongly to AI news. They run a two-stage regression: first each firm's sensitivity to an AI stock index, then how that sensitivity varies with software engineering payroll share, controlling for industry and firm size. A model then converts the slope into a productivity gain.

The baseline slope is 1.24 and stays similar when the authors add controls for firms' exposure to generative AI (1.28), other AI indexes, standard financial ratios and finer industry groupings. Year-by-year regressions from 2022 through 2025 are stable. A coefficient-stability test suggests unobserved differences between firms do not explain the pattern.

To get from software engineering productivity to GDP, the authors apply a modified Hulten's theorem and include software-producing firms, whose costs also fall. Adding a task-based model with an explicit AI sector and a full input-output network changes the result only moderately. Adding an R&D sector roughly doubles the GDP effect.

The paper contrasts its approach with studies that start from occupation-level AI exposure. Those face the problem that large task-level gains often shrink at the firm level, for example more coding activity but smaller increases in projects and releases. Because stock prices already reflect investors' beliefs about how tasks combine into jobs, the authors say they sidestep that aggregation step.
sources:
- peer_reviewed | NBER Working Paper | https://doi.org/10.3386/w35793 | 2026-09-01
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

fdd0370d33e773e0bd10a0108da5b348bd6a20c9e5f0e1ff9e94b87d375ff8fb

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000