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 Pragmatic Engineer report drawing on interviews at 27 workplaces and aggregate data from CodeRabbit and Gitar finds that AI-tool spending varies widely, while pull-request output has risen in some datasets. The report says few companies can yet establish whether those changes amount to a financial return on investment.
A report by The Pragmatic Engineer, based on conversations with engineering leaders and senior engineers at 27 workplaces and aggregate data from two software companies, finds that spending on AI coding tools varies sharply while some measures of pull-request output have more than doubled. But the report says few companies can answer whether the higher spending produces a measurable return.
CodeRabbit shared anonymized data from engineers using its AI code-review platform. It reported a 113% increase in weekly pull requests and a 127% rise in monthly PRs. In a separate tracking of the same 20,000 developers over a year, the increase was 128%. Those figures describe activity among platform users; they do not, by themselves, show that AI tools caused the rise or that the extra work improved software quality or business results.
Gitar, an AI code-verification agent from Sonar, shared data from hundreds of customers and said teams it classifies as “AI-native” ship more than 60 PRs per developer per week. The report also presents a wide range of company budgets: a two-person .NET startup said it spends $3,000 per developer per month, while a publicly traded company with 1,500 engineers caps monthly spend at $150 per developer. The examples are reported figures from individual organizations, not a representative industry average.
Costs also shifted over time for some companies. One trading firm reported total AI spending of $8 million per month in September, up from $500,000 in January, and average spending of $1,600 per developer per month, compared with about $80 in January. The source says traders, rather than engineers, were the largest spenders at that firm. These company-specific numbers should not be read as typical software-industry spending.
Higher PR Counts, Unsettled Returns
The figures show why companies are trying to track AI use as budgets grow: pull requests are relatively easy to count, and the report’s author describes them as a readily available measure that may be useful for comparing activity over time. But PR volume is not a direct measure of productivity or value. A higher count does not establish that code is better, that work takes less time, or that a company earns more than it spends on tools.
For engineering teams, the distinction affects decisions about subscriptions, usage caps and infrastructure. The report says spending caps are common among medium- and high-spending companies, and that some subsidized subscriptions are ending. Companies may then face a choice between paying substantially more, limiting use or shifting to less costly models. The reported possibility of bills rising five to 10 times concerns companies losing subsidies, not a universal forecast.
As an affiliate, we earn on qualifying purchases.
How Companies Track AI Use
The report was prompted by a reader asking how much companies spend per engineer and whether PR throughput changes after AI tools are introduced. The author interviewed leaders and senior engineers at 27 workplaces, anonymizing most company names at their request, and added aggregate figures supplied by CodeRabbit and Gitar. The source does not describe this group as a statistically representative survey.
Across those accounts, the report says company size did not track with AI spending: a small startup reported very high per-person costs, while a much larger organization set a relatively low cap. It also reports that spending on API plans peaked for many companies in June and July and later roughly halved, with respondents citing a move to open models and a change in model use. Claude and Opus 5.5 were frequently named in the interviews, alongside Cursor, Codex and GitHub Copilot. These are observations from the report’s limited sample, not a ranking of tools across the industry.
““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 PR Counts Cannot Show
The report does not establish that AI tools caused the increases in PR activity reported by CodeRabbit, or that those increases translate into faster delivery, fewer defects or lower overall costs. The vendor datasets cover platform users or customers, and the report does not provide a controlled comparison with teams not using AI. Quality, rework, security outcomes and business impact are not resolved by the PR counts presented.
Spending figures also come from interviews and selected company examples, with many organizations anonymized. The source does not provide a complete, comparable account of each company’s tool costs, engineer headcount, usage, productivity or savings. It remains unclear how much of the reported spending is attributable to coding, how sustainable the observed output changes are, and whether results differ by team or type of work.
Budgets Face Their Next Test
The report describes companies continuing to adjust AI budgets as costs, subscriptions and model choices change. Some respondents said subsidies are due to end, while the trading firm cited plans to build internal GPU and CPU capacity and develop models. Those plans are reported intentions; the source does not give a timeline or show what savings or performance they may deliver.
The next useful evidence will depend on companies comparing costs with outcomes beyond PR counts, including code quality, delivery time, maintenance and business results. The report does not announce a shared industry standard or a future measurement date. For now, its findings offer a snapshot of reported spending and activity, rather than a settled calculation of AI’s return.
Key Questions
What did the report find about AI spending?
Spending differed widely among the 27 workplaces discussed. Examples ranged from a $150 monthly per-engineer cap at one large company to reported spending of $3,000 per developer per month at a two-person startup; these are individual cases, not an industry average.
Did pull-request output rise?
CodeRabbit reported year-over-year increases of 113% in weekly PRs and 127% in monthly PRs among its platform data. Tracking 20,000 developers for a year showed a 128% increase. Those figures do not establish that AI caused the changes.
Does the report prove that AI tools pay for themselves?
No. The report says few companies can answer the ROI question. PR counts measure activity, not financial returns, code quality or the total cost of development.
Why are some companies reviewing their AI budgets?
The report says some subsidies are ending and companies are using spending caps or changing models. Respondents at some companies reported lower costs after moving to open models, while others anticipate higher bills when subsidies expire.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
