What the 32.6% AI productivity estimate means for engineering budgets

The headline estimate describes a market expectation

A 32.6% productivity estimate from economists associated with the US National Bureau of Economic Research expresses the benefit financial markets expect from more productive software engineering. The researchers infer that expectation from share prices through an economic model. So using the figure to set a team’s delivery targets or justify its AI subscriptions requires a further assessment of what the tools change for that team.

The spending decision is already here: AI development tools such as Anthropic’s Claude Code and OpenAI’s Codex have substantially changed software development, and investment in AI-generated software appears likely to grow further. Buyers pay monthly subscriptions and may face additional usage charges, which makes the value of AI assistance a recurring budget question. The economic issue is whether the benefit of the software produced with that assistance makes the spending worthwhile.

For buyers, those payments sit alongside the cost of employing the people who develop their software. An engineering productivity estimate can help frame that decision because it concerns the economic value of their work. But allocating money between tools and payroll requires connecting the expected benefit to the company’s own costs. Understanding how the researchers arrive at their estimate explains what it can contribute to that choice.

Share prices connect expected profits to engineering productivity

The researchers examine that economic value in their paper, The Macroeconomic Effect Of AI: Sizing The Software Engineering Channel. They study enterprises outside the software and semiconductor industries, comparing how their share prices respond to AI news. Within that group, the important difference is how heavily each company depends on software engineers: if AI improves engineering productivity, companies with greater dependence should receive a larger benefit.

That difference gives the researchers a way to connect engineering work to company valuations. They describe how the benefit moves from productivity to expected profits and then into share prices:

“The main idea is that if AI raises the productivity of software engineers, then firms that rely more heavily on them should benefit more. Therefore news about AI generates higher expected profits for these firms. These profits are capitalized into larger stock price responses. Combining these stock price responses with a model, we can pin down the increase in software engineering productivity expected by financial markets,”

The model turns those share-price responses into an estimate of engineering productivity. Investors’ expectations of higher profits raise company valuations; the researchers then infer the productivity improvement that would account for those responses. “Market-implied” therefore means that the estimate comes from prices interpreted through the model.

To express the expected benefit over time, the researchers use expected present value: the value today of benefits anticipated in the future. They report that value as an equivalent permanent productivity increase, meaning an ongoing improvement with the same expected present-value effect. Their finding covers a defined period:

“From November 2022 to December 2025, AI increased the market’s expected present value of software engineering productivity by the equivalent of a permanent 32.6% productivity increase.”

The dates identify the period during which the researchers estimate that market expectations changed. Within that finding, “permanent” describes the continuing productivity increase used to express the equivalent value. The percentage therefore represents a modeled expectation about engineering’s economic contribution, with a specific time basis.

Moving from a market estimate to a team budget

Applying that expectation to a team requires moving from company valuations to the work its engineers can deliver. Investors price an expected benefit to a business, while a delivery target commits a particular team to completing work. A leader therefore needs to assess the team’s results before using the estimate to promise more output. The study’s focus on enterprises outside software and semiconductors also defines the businesses examined.

Those team results give the buyer a basis for assessing whether recurring tool payments are worthwhile. Monthly subscriptions and any additional usage fees must be considered alongside the benefit the company receives from AI-assisted development. A market expectation can provide a starting point for that assessment, but the spending decision depends on the buyer’s own benefits and costs.

The same assessment informs how much budget goes to tools and how much goes to payroll. Funding Claude Code, Codex or other AI assistance draws on resources that also support the engineers using those tools. Deciding that allocation requires relating tool spending to its effect on engineering work and to the cost of employing the team.

Key takeaways for decision-makers

Use the estimate as a market signal: The 32.6% figure reflects financial markets’ expectations of AI’s contribution to software engineering productivity. Engineering leaders need evidence from their own teams before translating that expectation into delivery targets.
Understand the valuation basis: Researchers model share-price responses to AI news among companies outside software and semiconductors. The estimate expresses the change in expected future benefits from November 2022 to December 2025 as an equivalent permanent productivity increase.
Ground tool budgets in team results: Engineering budget owners need to weigh subscription fees and additional usage charges against the benefits of AI-assisted development. Assessing those costs alongside payroll gives buyers a basis for allocating spending between tools and people.