TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
The Pragmatic Engineer gathered AI spending and pull-request data from 27 workplaces, alongside aggregate figures from CodeRabbit and Gitar. CodeRabbit reported PR counts roughly doubling among developers in its data, but the report found that most companies cannot yet say whether higher AI spending produces a measurable business return.
The Pragmatic Engineer’s survey of 27 workplaces found that companies’ spending on AI tools varies widely, while many still cannot quantify the return on that spending. Separate data shared by CodeRabbit showed weekly pull requests rising 113% among developers in its sample, but PR volume alone does not establish that AI caused the increase or improved business outcomes.
The report combined interviews with engineering leaders and senior engineers at 27 companies with aggregate data from CodeRabbit and Gitar, providers of AI-assisted coding and code-review tools. The companies’ reported spending ranged from less than $300 per engineer per month to more than $1,000, with one two-person .NET business reporting $3,000 per developer per month. The report said company size did not appear to predict spending in this sample.
CodeRabbit shared anonymized data from engineers using its platform. It reported a 113% rise in weekly PRs and a 127% increase in monthly PRs. For a tracked group of 20,000 developers, the company reported a 128% increase over a year. These are vendor-supplied figures; the report does not establish that AI use alone drove the changes or that the additional PRs translated into more valuable, reliable software.
Gitar’s data, drawn from hundreds of customers, showed substantial variation in PR output between teams. The report also described companies using spending limits, with additional usage sometimes requiring justification. Several interviewees said subsidized subscriptions were ending, potentially exposing costs that had previously been less visible. The author said API-based costs had roughly halved from summer peaks at some companies, citing moves to open models and changes in model use.
Higher Output, Unproven Returns
The findings show why engineering leaders are looking beyond tool adoption to cost per engineer and changes in output. AI bills can vary by thousands of dollars per developer each month, while subscription arrangements may hide the full cost until subsidies end. Companies setting budgets may need to track usage and spending more closely as these costs become direct expenses.
But a higher PR count is an incomplete measure of productivity. Pull requests can differ in size, complexity, review burden and quality; more of them do not automatically mean faster delivery or greater customer value. The report’s central point is that measuring ROI remains difficult: the available numbers can describe spending and activity, but do not by themselves show whether the investment pays off.
As an affiliate, we earn on qualifying purchases.
How the Report Measures Output
The report focuses on pull requests, or PRs, as a readily available measure that companies can compare before and after adopting AI tools. Its author described PR throughput as a simple metric that may be less susceptible to being gamed than some alternatives, while noting that it is still only a proxy for engineering work.
The source material brings together different kinds of evidence: interviews with people at 27 workplaces, CodeRabbit’s platform data and Gitar’s customer data. These sources cover different populations and do not amount to a controlled study. The report’s interviewees also vary in company size, industry and AI spending arrangements, so their figures should not be treated as a representative estimate for all software companies.
“What’s the ROI of AI? A question that few companies seem able to answer – at least for now.”
— The Pragmatic Engineer report
software development PR review tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Figures Cannot Show
The report does not establish a causal link between AI tools and rising PR counts. It also does not provide a common measure of software quality, review time, delivery speed or customer impact across the 27 companies. As a result, the figures cannot show whether added PRs represent more useful work or whether any productivity gains exceed tool costs.
The sample is limited and combines interviews with vendor data, whose methods and customer populations differ. The report does not give a representative industry-wide spending average. It also says some subscriptions obscure direct costs, and that subsidies are ending at certain companies; the eventual bills and how organizations respond remain unclear.
As an affiliate, we earn on qualifying purchases.
Companies Face Cost Reviews
Companies described in the report are reviewing spending caps, usage and model choices as AI bills change. Some may need to budget for higher direct costs when subsidized plans end, while others are moving workloads to open models or building internal computing capacity. The report says spending had fallen from summer peaks at many API-plan companies, but does not establish whether that pattern will continue.
The next useful evidence would compare costs with outcomes over time, including code quality, delivery timelines and maintenance demands—not just PR counts. The report does not identify a shared measurement standard or provide a timetable for when companies will be able to assess returns more reliably.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI tools per engineer?
The 27-company report found a wide range. Some companies spent below $300 per engineer per month, while others reported more than $1,000. One two-person .NET business said it spent $3,000 per developer per month. The sample is not an industry-wide benchmark.
Did AI tools cause the increase in pull requests?
The report does not establish that. CodeRabbit reported rising PR counts among users of its platform, but the data do not isolate AI’s effect from other factors such as team practices, staffing or changes in the work being done.
Does a higher number of PRs prove AI is paying off?
No. PR volume measures activity, not necessarily software quality, customer value or total productivity. A fuller ROI assessment would compare tool costs with outcomes such as delivery time, defect rates and maintenance work.
Why are some companies’ AI costs changing?
The report cites spending caps, model changes, moves to open models and the end of subsidized subscriptions. It says API costs had roughly halved from summer peaks at some companies, but the figures come from the report’s limited set of interviews and are not a forecast.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
