TL;DR
A report by The Pragmatic Engineer examined AI-tool spending at 27 workplaces and reviewed aggregate pull-request data from CodeRabbit and Gitar. CodeRabbit data showed weekly PRs up 113% year over year, but the report found that companies generally cannot yet say whether the added output justifies their AI costs.
The Pragmatic Engineer has published findings on software companies’ spending on AI tools, drawing on interviews with engineering leaders and senior engineers at 27 workplaces and aggregate data from CodeRabbit and Gitar. The report found that pull-request counts rose in CodeRabbit’s platform data, but that few companies could provide a clear return-on-investment figure linking AI expenses to business gains.
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
CodeRabbit’s anonymized data showed an average 113% rise in weekly pull requests 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 platform measurements of PR activity, not direct evidence that AI tools caused the increase or that the additional work improved product quality or business results.
Gitar, an AI code-verification agent, shared data from hundreds of customers showing wide differences in PR output per developer. The report said some teams categorized as “AI-native” shipped more than 60 PRs per developer per week. The Pragmatic Engineer’s author suggested workplace norms may influence how much developers ship; that is an interpretation, not a measured cause in the supplied data.
Spending also varied sharply among the 27 workplaces. The report described a two-person .NET company spending $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped monthly spend at $150 per developer. It also reported a quant firm whose total AI spending rose from $500,000 a month in January to $8 million in September, while average spend reached $1,600 per developer monthly. That firm’s biggest spenders were traders, not engineers.
More PRs, Unsettled Payback
The findings show why technology leaders are trying to put budgets around AI tools as usage grows: per-person costs can differ dramatically, and subscription pricing may obscure the true expense. Some companies in the report use spending limits, while others expect costs to rise as subsidized subscriptions end. The author said several organizations may face bills five to 10 times higher when subsidies stop, though the report does not establish how widespread that outcome will be.
PR counts offer a readily available measure of engineering activity, but they do not reveal whether code is useful, reliable, or valuable to customers. A higher count could reflect changes in workflow, team expectations, or the amount of work divided into separate requests. Without comparable measures of quality, delivery time, maintenance costs, and business outcomes, more PRs cannot by themselves establish positive ROI.
For readers evaluating company claims about AI productivity, the distinction matters: the report offers evidence of increased activity in particular datasets and examples of high spending, not a controlled comparison proving that the tools produced the additional output. It leaves the central financial question open: whether measurable benefits outweigh tool costs and any added review or maintenance work.
Products Worth Considering
Jackery Explorer 2000 v2 Portable Power Station with Two 200W Solar Panels
EcoFlow DELTA 3 Classic Solar Generator with 220W Solar Panel
Bose QuietComfort Wireless Headphones – Twilight Blue (Limited Edition)
As an affiliate, we earn on qualifying purchases.
How Spending and Output Were Measured
The Pragmatic Engineer said its author gathered spending and PR details from engineering leaders and senior engineers at 27 workplaces, anonymizing most companies at their request. The report also included aggregate data supplied by CodeRabbit and Gitar. These sources provide a cross-section of reported practices, not a representative survey of all software companies.
The article framed PR throughput as a simple before-and-after measure available at many workplaces. CodeRabbit compared platform activity with the previous year and separately tracked 20,000 developers. Gitar supplied customer-level categories of PRs per developer per week. The figures come from different datasets and should not be treated as a single industry-wide estimate.
The report said spending peaked around June and July at many of the companies it contacted and had roughly halved for some API-plan users. It attributed the decline, based on company accounts, to use of Opus 5.5 and a move toward open models. The supplied material does not specify the full measurement period or detailed methodology behind every spending estimate.
“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 quant firm, quoted by The Pragmatic Engineer
The Missing Business-Outcome Data
The report does not establish that AI tools caused the rise in PR counts. It also does not show whether those PRs were merged, improved software quality, shortened delivery times, or generated revenue. ROI remains unquantified for most of the workplaces discussed, and the source material does not give a common formula used to calculate returns.
The interviews and platform datasets are limited in scope, and most company names were withheld. The supplied material also does not provide enough methodological detail to determine how representative the 27 workplaces or platform customers are. Spending figures may not be directly comparable where companies use subsidized subscriptions, API billing, internal infrastructure, or different spending caps.
Companies Face Cost Reviews
The report points to companies continuing to review AI budgets as usage expands and subsidized subscriptions expire. Some organizations may need to decide whether to raise spending limits, restrict usage, or move workloads to other models; the article reports these as current cost pressures, not as a single industry-wide policy shift.
For a clearer ROI assessment, companies would need to compare spending with outcomes beyond PR volume, such as review and rework effort, delivery speed, reliability, and the value of shipped work. The report does not announce a follow-up study or provide a date for broader results. Whether firms will adopt shared methods for measuring those outcomes remains unclear.
Key Questions
What did the report find about AI coding ROI?
The Pragmatic Engineer found that few companies could clearly quantify ROI. It reported spending patterns and higher PR counts in platform data, but did not establish that the tools caused the increases or that they produced business gains exceeding their cost.
How much did PR activity rise in CodeRabbit’s data?
CodeRabbit reported a 113% year-over-year increase in weekly PRs, a 127% rise in monthly PRs, and a 128% increase for a tracked group of 20,000 developers. These are platform observations, not proof of a causal effect from AI tools.
How much were companies spending on AI tools?
Spending varied widely in the report’s 27-workplace sample. Examples ranged from a $150 monthly cap per developer at a large public company to $3,000 per developer per month at a two-person startup. The figures are not a representative industry average.
Does a higher number of pull requests prove AI made engineers more productive?
No. PR counts measure activity, not necessarily code quality, customer value, or time saved. The report does not prove that AI caused the increases in the platform datasets.
What remains unknown about AI spending?
It remains unclear how much of the reported output is attributable to AI, whether it improves business outcomes, and how costs compare across different billing models. The report says most companies it contacted could not yet give a reliable ROI answer.
Source: rss
