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 productivity information from engineering leaders and senior engineers at 27 workplaces, alongside aggregate data from CodeRabbit and Gitar. CodeRabbit reported roughly doubled pull-request output among users over a year, but the report says few companies can yet determine whether the additional output justifies their AI costs.
The Pragmatic Engineer has published a survey of AI spending and software pull-request output across 27 workplaces, finding wide variation in monthly costs and evidence of higher PR volumes in platform data. But the report says few companies can answer whether the added output produces a measurable return on what they spend.
The report draws on conversations with engineering leaders and senior engineers at 27 workplaces, most of whose names were withheld at their request. It also includes aggregate usage information shared by CodeRabbit and Gitar, providers of AI-assisted code review and verification tools. The findings are an informal, limited sample rather than a representative survey of the software industry.
CodeRabbit’s anonymized data showed a 113% increase in weekly pull requests and a 127% increase in monthly PRs. For a tracked group of 20,000 developers, the platform reported a 128% rise over a year. Separately, Gitar’s data from hundreds of customers showed that some teams categorized as “AI-native” were shipping more than 60 PRs per developer per week. These are output measures, not proof that AI tools caused the changes or that the PRs improved software quality.
Spending varied sharply: a two-person .NET business reportedly spent $3,000 per developer each month, while a publicly traded company with 1,500 engineers capped spending at $150 per developer monthly. One trading firm reported spending $8 million in September, up from $500,000 in January, and average outlay of $1,600 per developer monthly. The report also describes common spending caps among medium and high spenders, and says some organizations’ subsidized subscriptions may end, potentially raising their bills.
Why Higher PR Counts Are Not ROI
The figures show why companies are trying to put a number on AI coding costs: tool use is expanding, while bills can differ substantially between organizations and business units. A higher PR count may suggest more work is being submitted, but by itself it does not show whether engineers finish valuable work faster, whether the code is reliable, or whether review and maintenance costs rise.
For managers deciding whether to expand access, the distinction matters. Spending and throughput are measurable; the business value of the work is harder to establish. Companies would need to connect tool costs and output to measures such as delivery time, defects, rework, and the importance of shipped changes before concluding that an increase in PRs represents a financial return. The report does not provide that kind of comparable, controlled analysis.
As an affiliate, we earn on qualifying purchases.
How the Report Measured Output
The report uses pull requests, or PRs, as a readily available measure that can be compared before and after AI adoption. Its author describes PR throughput as a metric that is accessible at many companies and may be less vulnerable to gaming than some other productivity measures. Still, PR counts measure activity: teams differ in how they split work into changes, and a larger count does not necessarily mean more completed or better work.
The report also found no clear relationship between company size and AI spending in its sample. It says spending peaked around June and July at many companies, then fell by roughly half for some organizations on API plans. The author attributes much of that decline to a mix of model changes and a move to open models. Those patterns come from the workplaces contacted, not a market-wide accounting of AI costs.
“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, as quoted in The Pragmatic Engineer
software development 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 report does not establish that AI tools caused the reported rise in PRs. The platform figures compare usage over time, but the material provided does not describe a control group or isolate AI use from other changes in staffing, workflow, company policy, or the way teams define a PR. Causation remains unproven, and the data do not establish whether code quality or delivery outcomes improved.
It is also unclear how closely the 27 workplaces represent the broader industry, since the sample is limited and many companies are anonymous. The report does not provide a common accounting method for subscriptions, API usage, internal infrastructure, or employee time. Consequently, cost comparisons and ROI estimates may not be directly comparable. Some spending trends are described as changing, but the report does not establish how they will develop across 2027.
As an affiliate, we earn on qualifying purchases.
Companies Face Cost Reviews
The report describes companies reviewing AI bills as they prepare for higher usage and, in some cases, the end of subsidized subscriptions. Organizations with caps may need to decide whether to raise limits, justify exceptions, or shift workloads to less expensive models. One high-spending trading firm told the report it was building internal GPU and CPU capacity, but the supplied material does not give a timeline or projected savings.
The next meaningful measure of ROI will depend on whether companies track more than tool adoption and PR totals. Readers should watch for data that links AI costs to delivery speed, quality, and business outcomes, with a clear comparison period and baseline. The report does not announce a shared industry standard for that measurement, so the return on engineers’ AI spending remains an open question.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI tools per engineer?
The report found a wide range in its 27-workplace sample, from a $150 monthly cap per engineer at one company to reported average spending of $1,600 per developer monthly at a trading firm. A two-person .NET business reportedly spent $3,000 per developer monthly. These examples are not industry averages.
Did AI tools cause pull-request output to rise?
The report presents higher PR counts in CodeRabbit platform data, including a 113% rise in weekly PRs. The figures do not establish that AI caused the increase; the supplied report does not isolate AI use from other possible changes.
Do more pull requests mean better productivity?
Not on their own. PR counts measure submitted changes, not their quality, business value, or the time needed for review and maintenance. The report does not provide comparable quality or outcome data.
Can companies currently calculate AI’s return on investment?
The report says few companies can answer that question so far. It describes available spending and throughput measures but does not establish a common method for connecting them to financial or product outcomes.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
