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
A report by The Pragmatic Engineer draws on interviews with engineering leaders and senior engineers at 27 workplaces, plus aggregate data from CodeRabbit and Gitar, to examine AI coding costs and output. CodeRabbit data showed more than twice as many weekly pull requests per developer as a year earlier, but the report found that few companies could say whether the spending produced a measurable return on investment.
Companies are spending widely varying amounts on AI tools for software engineers, while available platform data shows substantial growth in pull-request output; however, few organizations can yet quantify the return on investment, according to a report from The Pragmatic Engineer. The report draws on interviews at 27 workplaces and aggregate figures from CodeRabbit and Gitar, offering a snapshot of costs and output rather than proof that AI tools caused the increase.
CodeRabbit, an AI code-review platform, shared anonymized data from engineers using its service. It reported a 113% rise in average weekly pull requests and a 127% increase in monthly pull requests. For a tracked group of 20,000 developers, monthly pull requests rose 128% over a year. Those figures compare platform activity across time; they do not establish that AI tools alone drove the change or that the additional pull requests led to better software or lower costs.
Gitar, an AI code-verification agent, supplied data from hundreds of customers showing large differences in weekly pull requests per developer among teams. The report says some AI-native teams ship more than 60 pull requests per developer per week. Its author cautioned that team norms and company culture may shape output rates, making comparisons between organizations difficult.
Spending also varied sharply in the 27-company sample. One two-person .NET business reportedly spent $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped spending at $150 per developer monthly. A quantitative trading firm reported spending $8 million per month in September, up from $500,000 in January, and average spending of $1,600 per developer monthly. These are examples reported by the article, not a representative industry-wide cost survey.
More Code, Unproven Returns
The figures highlight a central challenge for engineering leaders: increased activity is not the same as demonstrated productivity or financial return. Pull-request counts are relatively easy to track, and can help organizations compare activity before and after introducing tools. But they do not show whether code is more reliable, whether features reach customers sooner, or whether review and maintenance work has increased.
The spending range also suggests that companies are making very different choices about access, budgets and oversight. Some impose per-person caps; others subsidize subscriptions or permit extensive usage. The report says several companies expect subsidized plans to end, which could raise bills five- to tenfold unless users or employers change tools or reduce consumption. For readers, the practical issue is not only whether AI is being used, but whether its costs and effects can be measured against meaningful engineering outcomes.
As an affiliate, we earn on qualifying purchases.
How the Spending Snapshot Was Built
The Pragmatic Engineer said it collected information from engineering leaders and senior engineers at 27 workplaces, anonymizing most company names at their request. It also gathered aggregate platform statistics from CodeRabbit and Gitar. The article describes the results as a snapshot of companies contacted, not a controlled study or a representative survey of all software firms.
According to the report, AI spending rose at many of the companies it contacted, with monthly costs peaking around June and July before falling roughly by half among some API-plan users. The author attributed the reductions in part to a move toward open models and changes in model use. Claude and Opus 5.5 were commonly named among the tools receiving spending, alongside Cursor, Codex and GitHub Copilot. These observations reflect the report’s sample and the period it covered; they should not be read as a current price comparison across all products.
““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 quantitative trading firm, as quoted by The Pragmatic Engineer
Output Does Not Prove Productivity
The available figures do not establish a causal link between AI tools and higher pull-request counts. The report does not provide a controlled comparison that accounts for staffing changes, project mix, company practices or other factors that could affect output. Nor do PR counts measure code quality, defects, review burden, delivery time or customer value.
The spending information is also self-reported and covers a limited, anonymized set of workplaces. The report does not give a standardized company-by-company accounting of subscriptions, API usage, internal infrastructure or employee time. A comparable, independently verified ROI measure remains absent, and the data cannot show whether increased spending is worthwhile for a typical employer.
Companies Face Cost and Measurement Choices
As some subsidized subscriptions expire and companies reconsider API bills, employers are likely to face decisions about spending caps, model choices and access to tools. The report describes caps as common among medium- and high-spending companies, though interviewees said they could often be bypassed. The trading firm cited in the article was also building internal computing capacity and models, an approach that may change costs but is not shown to deliver savings.
The next useful evidence would compare costs with outcomes beyond PR volume, such as time to deliver work, defect rates, maintenance demands and the value of completed projects. The report does not set out when such evidence will become available. For now, its findings document rising use and varied spending, while leaving the core question—whether those costs produce a measurable net benefit—open.
Key Questions
How much are companies spending on AI coding tools?
The report found wide variation among 27 workplaces, from $150 per engineer per month at one large company to much higher spending elsewhere. One two-person business reported $3,000 per developer monthly. These examples are not an industry average.
Did AI increase the number of pull requests?
CodeRabbit reported 113% growth in average weekly pull requests and a 128% increase in monthly pull requests for a tracked group of 20,000 developers over a year. The figures show an association in platform data, not proof that AI caused the increase.
Does more pull-request activity prove AI is worth the cost?
No. Pull-request counts measure activity, not code quality, delivery speed, maintenance costs or business value. The report says few companies could quantify AI’s overall return on investment.
Are the report’s spending figures representative?
No such claim is made. The spending details came from interviews at 27 workplaces, with most company names withheld, and the report presents them as a snapshot rather than a representative survey.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
