- NBER economists estimate stock markets imply the equivalent of a permanent 32.6% AI gain in software engineering productivity at non-software, non-semiconductor firms, 2022 to 2025.
- The figure reflects investor expectations run through a model, not measured output. InfoWorld notes it does not mean your own engineers are that much more productive.
- Use it as a starting point, then test AI tools against your own cycle times and costs before moving money from payroll.
A finance chief weighing a coding-tool subscription against an engineer's salary wants one clean number. A new economics paper offers one: 32.6%. That figure applies to engineers at firms outside the software and semiconductor industries. It is useful only if you know what it counts.
What the economists did
NBER lists the paper as Working Paper 35793. Its authors are economists Alex Blumenfeld, Jonathon Hazell, Chen Lian and Andreas Schaab. They try to size how much AI changes the economy through one channel: the work of software engineers.
They did not survey developers or time their work. They used financial markets instead. The researchers estimated how much each firm's stock return moves with an AI stock market index. They then checked how that sensitivity depends on the share of the firm's payroll spent on software engineers.
A model turned that pattern into a productivity figure. InfoWorld reports that the analysis covers firms outside the software and semiconductor industries. A skimming reader should not take the result as a figure for all software engineers.
How the method works
The logic is simple. If AI makes engineers more productive, firms that employ many engineers should gain more from AI news. Investors would then mark those firms' shares up by more.
The researchers say this lets them read the productivity gain that financial markets expect. The measure looks forward and is available in real time. That is its main advantage over slow surveys.
What the number means, and what it does not
The paper says AI raised the market's expected present value of software engineering productivity. The size of that rise is the equivalent of a permanent 32.6% productivity increase. Present value means the worth today of gains expected in the future.
So the figure is a model-implied estimate, not a stopwatch reading. Investors did not set a 32.6% price directly. The authors back it out of stock reactions using a model. It tells you what markets appear to expect, not what happened inside a given team.
InfoWorld makes the same point. The figure does not mean your engineers will be that much more productive. It calls it a starting point for dividing money between AI tools and payroll.
The bigger claims in the paper
The authors also map the figure to the whole economy. They estimate a 3.6% effect on the level of GDP in their baseline case. The estimate rises to 6.5% if higher software productivity also lifts research and development productivity.
The paper also updates the picture to mid-2026, a period of fast progress in coding agents. By then, the estimated effect on both productivity and GDP had more than doubled from its end-of-2025 size.
The lesson for executives
Markets are placing a large bet on coding tools. That bet can shape your vendors' pricing and your board's expectations. It is not evidence that your own spending will pay back.
The risk falls on the people in the middle. An engineering manager may be asked to deliver a headline gain that was never measured on their team. A developer may find that headcount plans assume it.
The paper supports a narrower and more useful idea. Treat the market's number as a hypothesis and test it locally.
Questions to put to your team
Which teams use AI coding tools today, and what do the subscriptions and usage fees cost per engineer?
What did delivery times and defect rates look like before and after adoption, on the same kind of work?
How much review time does AI-written code add, and who carries it?
Which budget decisions currently rest on an industry-wide figure instead of your own data?
Run a bounded pilot, measure it, and only then move money between tools and people. Markets have put a forecast on the table. Your job is to find out whether you are collecting it.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: National Bureau of Economic Research (NBER).
