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 gathered AI-spending and pull-request data from 27 workplaces, alongside aggregate data from CodeRabbit and Gitar. The figures show higher PR activity in some datasets and wide variation in AI costs, but the report says few companies can yet establish a clear return on investment.
The Pragmatic Engineer has reported wide differences in what companies spend on AI tools per software engineer, while finding that few organizations can clearly quantify the return. Its survey of 27 workplaces and separate platform datasets show higher pull-request (PR) counts in some groups, but do not establish that AI spending caused the increases or improved business outcomes.
For its report, The Pragmatic Engineer asked engineering leaders and senior engineers at 27 workplaces about AI costs and changes in engineering output. It also gathered aggregate data from CodeRabbit and Gitar, providers of AI code-review and code-verification tools. The author said spending per engineer varied widely and did not track company size in the sample: a two-person .NET business reported spending $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped spending at $150 per person.
CodeRabbit shared anonymized data showing a 113% increase in weekly PRs among data engineers using its platform, and a 127% rise in monthly PRs. In a tracked group of 20,000 developers, the report said PR counts rose 128% over a year. Gitar, meanwhile, reported that some teams categorized as “AI-native” shipped more than 60 PRs per developer per week. These are platform and customer observations; the report does not show that AI alone produced the changes.
Spending also varied over time and by billing arrangement. The report says API costs peaked around June and July and had roughly halved for some companies by the time of reporting, with respondents citing a switch to open models and use of different models. Spending caps were common among medium- and high-spending companies, while subsidized subscriptions sometimes obscured actual costs. Some companies said subsidies were ending, potentially raising bills if usage and plans stayed the same.
Higher PR Counts, Unsettled Returns
The findings describe a measurement challenge for engineering organizations: AI bills are increasingly visible, but a higher number of PRs does not by itself establish greater productivity or financial value. PR throughput can be counted and compared over time, making it one measure companies may use when reviewing tool budgets. The report does not provide a common measure of PR quality, maintenance burden, review time, or revenue impact.
The report does not determine whether additional code changes outweigh model and subscription costs. Companies using subsidized subscriptions may also face higher costs when discounts expire. The reported spending caps and moves to alternative models indicate efforts to manage bills, but the report does not establish whether those measures preserve output or quality.
As an affiliate, we earn on qualifying purchases.
How the Report Measured AI Use
The report was prompted by a reader asking how much other companies spend on AI per engineer and whether PR throughput changed after adoption. Its author gathered responses from engineering leaders and senior engineers, anonymizing most companies at their request, and paired those accounts with aggregate vendor data. The evidence is therefore a mix of company-reported spending and platform-specific activity, rather than a controlled comparison across the industry.
The report describes Claude and Opus 5.5 as frequent spending destinations among the companies interviewed, with Cursor, Codex, and GitHub Copilot also mentioned. It says spending caps were common at medium-and-higher spenders and that some caps could be bypassed. One company described in the report said its monthly AI spend rose from $500,000 in January to $8 million in September, with average spend increasing from about $80 to $1,600 per developer per month. The report said traders, not engineers, were the largest spenders there, underscoring that company-wide AI costs do not necessarily map neatly to software engineering output.
““The question of ROI of AI is one that few companies seem able to answer – at least for now.””
— The Pragmatic Engineer report
software engineering productivity tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Output Figures Cannot Show
The figures do not establish a causal link between AI use and higher PR counts. The report does not provide a consistent comparison group of similar teams that did not adopt AI, nor does it define one shared starting point for all company-level changes. Vendor datasets cover users or customers of particular platforms, so they may not represent software engineers generally.
It is also unclear whether more PRs correspond to better software, faster delivery, fewer defects, or lower total costs. The report does not provide a single standardized ROI calculation across the 27 workplaces. Spending figures can be hard to compare when companies combine per-seat subscriptions, subsidized plans, and API usage. The report’s spending examples and model references are snapshots from its reporting period, not a current price comparison.
AI development platform subscriptions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Companies Still Need Better Measures
The report does not announce a follow-up study or a shared industry standard for measuring returns. Companies evaluating AI budgets will need to compare costs with output and quality measures while accounting for changes in billing plans and usage. The report’s evidence indicates that PR volume is one operational measure, but does not answer the ROI question on its own.
Further assessment would require organizations to track comparable periods and teams, include model and subscription costs, and examine whether changes in code review, defects, delivery time, or business results accompany higher PR counts. The report leaves open how many companies will adopt such measures and how much of the reported activity increase is attributable to AI.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI per engineer?
The 27-workplace sample showed a wide range. Reported examples included $3,000 per developer per month at a two-person startup and a $150 monthly cap per engineer at a larger public company. These examples are not an industry-wide average.
Did AI increase the number of pull requests?
CodeRabbit’s platform data showed increases of 113% in weekly PRs and 127% in monthly PRs; a tracked group of 20,000 developers showed a 128% rise over a year. The data show an association over time, not proof that AI caused the increase.
Does the report show that AI spending pays off?
No. The report says few companies can currently quantify AI’s return. It provides spending reports and output measures, but no consistent calculation connecting costs to quality, revenue, or total productivity gains.
Why are some companies seeing AI costs fall?
The report says some companies’ API costs had roughly halved after peaking around June and July. Respondents attributed the decline to moving to open models and changing model use; the report does not establish that the pattern applies across companies.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
