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 examined AI spending and pull-request (PR) output at 27 workplaces, alongside aggregate data from CodeRabbit and Gitar. CodeRabbit reported PR counts more than doubling among surveyed users, but the report found that few companies could say whether higher AI spending is producing a measurable return.
Companies are spending widely different amounts on AI tools for software engineers, but many still cannot say whether the outlay is paying off, according to a report by The Pragmatic Engineer based on interviews and information from 27 workplaces. Separate vendor data in the report shows higher pull-request counts among developers using AI tools, but those counts alone do not establish that AI caused the increase or improved business results.
The report combines responses from engineering leaders and senior engineers with aggregate figures supplied by CodeRabbit and Gitar, providers of AI code-review and verification tools. It describes spending that ranges from companies setting low monthly allowances to a two-person .NET business reportedly spending $3,000 per developer per month. The sample is not presented as representative of all employers, and most participating companies were anonymized.
CodeRabbit said developers using its platform submitted 113% more weekly PRs on average than a year earlier. Its figures also show a 127% increase in monthly PRs; for a tracked group of 20,000 developers, the reported increase was 128% over a year. Gitar, drawing on hundreds of customers, reported that some teams categorized as AI-native submit more than 60 PRs per developer per week. These are vendor-provided measurements, not an independent controlled study.
Among companies in the report, spending patterns varied sharply and did not track company size in any simple way. The author said spending caps were common at medium- and higher-spending workplaces, although respondents described the limits as easy to bypass. Several companies also said subsidized subscriptions were ending, which could force them to cut costs or absorb substantially higher bills.
Higher PR Counts, Unsettled Returns
The figures matter because engineering leaders are trying to decide whether to expand AI access, restrict costly usage or shift workloads to cheaper models. PR volume is easy to count and can offer a before-and-after signal, but it does not show whether the code is useful, safe, maintainable or reaching customers faster. More submissions can also mean smaller changes rather than more completed work.
The report’s central finding is a gap between visible usage and measurable business value. Without comparable evidence on quality, review burden, delivery time and operating costs, companies risk treating activity as productivity. The reported variation in spending also suggests a single budget benchmark per engineer may not fit different roles, tools or workloads.
AI code review tools for developers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Report Measured Output
The author said an engineer had asked what companies spend per developer and whether PR throughput changed after adopting AI. The report uses PR counts because most companies can retrieve them and compare periods. It also notes that output norms differ among teams: the author suggested that workplace culture can shape how frequently engineers submit changes.
The spending snapshot includes companies on both subscriptions and usage-based API plans. According to the report, monthly API costs peaked around June and July for many of the companies interviewed and later fell by roughly half in some cases. Respondents attributed lower bills to model changes and increased use of open models. The report also found Claude and Opus 5.5 frequently named among major spending destinations, alongside Cursor, Codex and GitHub Copilot.
One high-spending trading firm cited in the report said its total AI bill reached $8 million in September, up from $500,000 in January, and averaged $1,600 per developer per month. Those figures are the company’s account as relayed by an engineer; the report says traders, rather than developers, were its biggest spenders. The firm reportedly planned internal computing capacity, underscoring that AI costs can extend beyond individual coding subscriptions.
“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 by The Pragmatic Engineer
What the PR Data Cannot Show
The reported PR increases do not establish that AI tools caused the change. The source material does not provide a controlled comparison that accounts for team size, codebase, project mix, hiring, release practices or other changes over the year. Nor does it say whether higher submission counts corresponded to better software, fewer defects, faster delivery or lower total costs.
The 27-workplace sample is not described as statistically representative, and company identities are mostly withheld. Vendor data reflects users of particular services and may not represent engineers generally. The report also does not provide a consistent accounting method for comparing subscription costs, API usage, internal infrastructure and employer subsidies. A company-wide ROI figure remains unavailable in the material provided.
Budgets Face a Measurement Test
The report points to a near-term budget question: companies whose subsidized subscriptions expire may need to decide whether to pay more, impose tighter limits or move work to lower-cost models. The author also describes spending caps and rate limits as common responses to rising usage. No universal spending threshold or deadline for those decisions is identified.
For a clearer assessment, employers would need to compare spending with outcomes beyond PR volume, such as review time, defects, delivery speed and the cost of maintaining code. The source does not report a shared measurement framework or a scheduled follow-up study. For now, the evidence supports a narrower conclusion: AI use and PR counts have risen in the reported data, while the financial return remains unclear.
Key Questions
How much are companies spending on AI tools per engineer?
The report found a wide range among 27 workplaces, from low monthly allowances to a reported $3,000 per developer per month at a two-person .NET company. It does not establish an industry-wide average.
Did AI increase pull-request output?
CodeRabbit reported that weekly PRs among developers using its platform were up 113% year over year, with similar increases in monthly counts and a tracked group of 20,000 developers. The figures show an association, not proof that AI caused the change.
Does a higher PR count prove AI is worth the cost?
No. PR volume does not by itself measure code quality, review effort, delivery speed or business value. The report says few companies could answer the ROI question.
Why do AI bills differ so much?
Companies use different tools, models, subscription arrangements and API workloads. The report also describes spending by non-engineers and says company size did not reliably predict per-engineer spending in its sample.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
