TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Pragmatic Engineer’s report, based on conversations with engineering leaders and senior engineers at 27 workplaces, found major differences in AI spending per developer, with no clear relationship to company size. CodeRabbit and Gitar data show higher pull-request output among some AI-tool users and teams, but the report says few companies can yet explain whether that added output justifies the cost.
The Pragmatic Engineer has published a report on AI-tool spending at 27 workplaces, finding that monthly costs vary widely and that few companies can clearly measure return on investment. Data shared by CodeRabbit and Gitar points to higher pull-request output among some developers and teams, but the report does not establish that AI tools caused the increases or that they outweigh the costs.
The report’s author said the inquiry began after an engineer asked how much other companies spend per developer and whether pull-request (PR) throughput changed after adopting AI tools. The author spoke with engineering leaders and senior engineers at 27 workplaces, anonymizing most company names at their request. The report also included aggregate figures from CodeRabbit and Gitar, which provide AI code-review or verification tools.
CodeRabbit’s anonymized platform data showed a 113% increase in weekly PRs and a 127% rise in monthly PRs. In a separate comparison tracking the same 20,000 developers over a year, monthly PRs increased 128%, the report said. These are vendor-supplied usage figures, not a controlled study of productivity or proof that AI tools alone produced the change.
Gitar data, drawn from hundreds of customer organizations, showed that some teams categorized as “AI-native” ship more than 60 PRs per developer per week. The report also described large differences in reported spending. A two-person .NET business spent $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped monthly AI spending at $150 per developer. The report said caps were common among medium- and higher-spending companies, though employees could often work around them.
Higher Output, Unproven Payback
Companies are making decisions about budgets, access limits and software-development practices while the financial payoff remains hard to quantify. PR counts offer a readily available way to track code changes, but they do not show whether the code is more valuable, whether it needs more review, or whether teams spend less time overall.
The contrast in reported spending also shows why a single industry-wide figure may mislead. The report found no clear link between company size and spending per developer; budgets appear to depend on local policies, tools, and willingness to absorb costs. For businesses, the practical question is not only how many PRs teams create, but whether faster output improves delivery without adding expense or maintenance work.
The report’s findings can help leaders benchmark their own costs and ask better questions, but they do not provide a reliable ROI formula. The underlying evidence combines interviews and vendor datasets rather than a standardized comparison of costs and outcomes across all 27 workplaces.
As an affiliate, we earn on qualifying purchases.
How Spending Patterns Are Shifting
The report describes a period of rapid change in company AI budgets. Respondents said spending peaked around June and July and then fell at companies paying for API use. The author attributed much of that decline to a mix of model changes and shifts toward open models; the report did not provide a single verified market-wide spending series.
Subscription plans can make per-engineer costs less visible than API bills, the report said. Some companies also told the author that subsidized subscriptions were ending, prompting them to prepare for higher charges or reduce usage. In the report’s account, Claude and Opus 5.5 were frequently named as major spending destinations, alongside tools such as Cursor, Codex and GitHub Copilot. These observations reflect the limited group interviewed, not a representative survey of all software companies.
One high-spending trading firm reported monthly AI costs rising from $500,000 in January to $8 million in September, which the report described as a 16-fold increase. Its average spend was about $1,600 per developer per month, up from roughly $80 in January, although traders were said to account for the largest costs. A company engineer told the report that traders sometimes ran expensive queries without realizing it, leading engineering staff to focus on rate limits.
“Lots of traders are using Claude. Engineering is learning the importance of rate limiting for traders who accidentally run expensive queries without realizing it.”
— An engineer at the trading firm, quoted in The Pragmatic Engineer
As an affiliate, we earn on qualifying purchases.
What the Output Figures Cannot Show
The report does not establish whether higher PR counts were caused by AI adoption. It also does not show whether those changes translated into faster releases, better software, lower labor costs or increased revenue. PR volume is a measure of activity, and differences in team culture or how work is divided could affect the totals.
The 27-workplace sample is not presented as a representative survey, and most companies are unnamed. CodeRabbit and Gitar supplied data from their own platforms, so the figures cover users or customers of those services rather than the entire software workforce. The report also does not provide enough comparable information to calculate a common cost-per-productivity-gain figure. How spending will change as subscriptions, model choices and internal computing capacity shift remains uncertain.
As an affiliate, we earn on qualifying purchases.
Companies Face Budget and Measurement Choices
The report points to two near-term decisions for employers: whether to maintain or change spending limits, and how to track outcomes beyond the number of PRs. Companies whose subsidized subscriptions are ending may need to choose between paying more, restricting use, or changing tools and models. The author also reported that some organizations are experimenting with open models or building internal computing capacity, but the report does not establish how common or effective those approaches are.
For now, the next test is whether organizations can connect AI costs to measurable engineering or business outcomes over time. That would require comparing spending with factors such as delivery time, review and maintenance effort, and the value of work shipped. The report offers no confirmed timetable for when a broader, standardized ROI assessment will be available, and the companies it describes may reach different conclusions depending on their work and cost structures.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did the report find about AI spending per developer?
Spending varied widely among the 27 workplaces the author spoke with. Examples ranged from $150 per developer per month at a publicly traded company to $3,000 at a two-person .NET business; these examples are not an industry-wide average.
Did AI tools cause the reported rise in pull requests?
The report does not establish cause and effect. CodeRabbit’s data showed increases in PR counts among developers using its platform, but the figures do not isolate AI’s role from other changes in teams or working practices.
How much did CodeRabbit’s weekly PR figure increase?
CodeRabbit data cited by The Pragmatic Engineer showed a 113% increase in weekly PRs. A separate analysis tracking 20,000 developers for a year showed a 128% increase in monthly PRs.
Can companies calculate AI’s ROI from PR counts alone?
No. PR volume measures output activity, but does not show whether software quality, delivery speed, maintenance effort or business results improved enough to cover AI costs.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
