The news
Stock markets have priced in an AI productivity gain for software engineers equal to a permanent 32.6% increase, economists estimate. Yet 84% of tech chiefs in a GFT Technologies survey released September 29 say legacy systems forced them to cancel an AI pilot or project.
The 32.6% figure comes from National Bureau of Economic Research (NBER) Working Paper 35793 by Alex Blumenfeld, Jonathon Hazell, Chen Lian and Andreas Schaab, economists at UC Berkeley and the London School of Economics, according to The Register, which reported the findings on September 29. The authors measured how each company's stock moved with an AI stock index, and whether that link was stronger for firms that spend more of their payroll on software engineers. A model then turned that pattern into a productivity estimate for November 2022 through December 2025. The paper puts the matching boost to the level of gross domestic product at 3.6%, or 6.5% if faster engineering also speeds research and development. By mid-2026, amid rapid progress in coding agents, the effect had more than doubled from the end of 2025, the authors wrote.
The number measures expectations, not observed output. Lian acknowledged to The Register that markets can be too optimistic or too pessimistic. The Register also noted that other researchers have reported task-level gains of 21% to 56%, and that code review can become a bottleneck that limits gains for a whole team.
Inside companies, the picture is harder. GFT, an IT services firm that sells modernization work, commissioned Wakefield Research to survey 945 CIOs and CTOs at companies with at least $500 million in annual revenue across 19 countries from August 11 to 31. Besides the 84% who reported a canceled AI effort, 93% said running AI on legacy infrastructure without modernizing will eventually trigger an enterprise-wide security crisis. Some 89% worried that global AI investment may be growing faster than the value it can deliver, and only 20% said their fellow executives and board members fully understand the security risks of running AI on old systems.
Progress on the fix is uneven. Only 15% of respondents said modernization was nearly or fully complete, while about a quarter said they had started but felt behind, CIO Dive reported. Rishi Chohan, GFT's U.S. chief executive, told CIO Dive that companies need not modernize everything at once and should first identify the systems blocking specific AI goals.
Talent adds cost. Compensation software company Payscale found that more than 4 in 10 companies cannot find the AI skills they need, and warned that new hires with those skills can command premiums of 20% to 40%, CIO Dive reported.
The numbers
- Priced-in software engineering productivity gain, Nov. 2022 to Dec. 2025 (NBER paper)
- 32.6% (permanent-equivalent)
- Implied effect on GDP level
- 3.6% baseline; 6.5% with R&D effect
- CIOs and CTOs who canceled an AI pilot or project over legacy limits (GFT)
- 84%
- Expect AI on unmodernized systems to trigger a security crisis (GFT)
- 93%
- Say their C-suite and board fully understand the risk (GFT)
- 20%
- Modernization nearly or fully complete (GFT, via CIO Dive)
- 15%
- GFT survey sample
- 945 CIOs and CTOs, 19 countries
Why CEOs should care
For CFOs, the two findings describe the same problem from opposite ends. If investors are already valuing software-heavy companies on engineering gains, boards will expect to see them. Yet the survey suggests many AI efforts stall on systems that were never built to share data with AI tools. Tie each tranche of AI spending to a modernization milestone you can verify, such as a legacy interface retired, a data source exposed through a secure application programming interface (API) or an approval step automated, and report delivery speed and defect rates alongside license counts.
CIOs and CTOs should follow Chohan's sequencing advice: pick the AI outcomes that matter, map the specific systems blocking them and put that modernization inside the AI business case instead of a separate budget line. Do not assume task-level coding gains will reach the whole team if review, testing and release processes stay the same.
Boards should ask for a briefing on the security risk of connecting AI to old systems, given that only 20% of surveyed technology chiefs think their board fully understands it. HR leaders should budget for retention: if new hires with AI skills get 20% to 40% more, the people doing the modernization may leave.
The bigger picture
The distance between what markets expect and what old infrastructure can deliver is where AI budgets are most likely to be cut. The NBER authors say the priced-in effect more than doubled by mid-2026, so expectations are rising faster than most modernization programs, by the survey's own account. Readers should weigh the survey's source: GFT sells the modernization work its findings recommend.
What’s next
Watch for companies that start reporting engineering output metrics, rather than AI adoption figures, when they discuss AI spending with investors. The Register reported that the NBER authors plan follow-up work on AI's economic effects through other industries, which would test whether the software engineering signal holds up.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
