TL;DR
Get privacy and security gear 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 figures from CodeRabbit and Gitar. The data points to higher PR volumes among some AI-tool users and widely varying costs, but the report says few companies can yet show whether that activity produces a financial return.
The Pragmatic Engineer’s report on AI spending found that 27 workplaces vary widely in what they pay for AI coding tools, while companies and vendors point to rising pull-request (PR) volumes as one possible measure of increased output. The report does not establish that the higher PR counts were caused by AI or that they translate into financial returns: few companies surveyed could answer the ROI question.
Among the aggregate figures supplied by CodeRabbit, developers using its AI code-review platform had a 113% increase in weekly PRs, according to anonymized platform data. The company also reported a 127% increase in monthly PRs and a 128% increase when tracking the same group of 20,000 developers over a year. These are changes in PR counts, not direct measures of code quality, engineering time saved or business value.
Gitar, an AI code-verification agent, shared data from hundreds of customers showing that some teams categorized as “AI-native” shipped more than 60 PRs per developer per week. The report’s author cautioned that company norms and culture may also shape how frequently engineers submit PRs. The figures are vendor-supplied observations; the source material does not describe a controlled comparison that isolates AI’s effect.
Spending among the 27 workplaces ranged from below $300 per engineer per month to more than $1,000. The report says company size did not predict spend: a two-person .NET business reportedly paid $3,000 per developer monthly, while a publicly traded company with 1,500 engineers capped monthly AI spending at $150 per developer. Spending caps were common among medium and high spenders, though the report says employees could often work around them.
Higher PR Counts, Unsettled Returns
For engineering leaders, the report highlights a gap between measurable tool usage and a demonstrated return. PR volume is relatively easy to count, but more submissions do not by themselves show that software ships faster, contains fewer defects, costs less to maintain or generates more revenue. Those outcomes matter when companies decide whether to expand AI budgets or renew subsidized subscriptions.
The range of reported costs also makes a single “spend per engineer” benchmark difficult to apply. A firm paying thousands monthly may have different workloads, internal rules and financial capacity from one with a tight cap. The report describes companies preparing for subsidized subscriptions to end, with some expecting bills could rise five- to tenfold if they do not cut costs or change plans. That makes tracking both spending and outcomes more pressing, but the report does not quantify a company-wide ROI.
As an affiliate, we earn on qualifying purchases.
How the Report Gathered Its Figures
The report was prompted by a reader asking how much companies spend on AI per engineer and whether PR throughput changed after adoption. Its author said they spoke with engineering leaders and senior engineers at 27 workplaces, anonymizing most company names at their request, and collected aggregate information from CodeRabbit and Gitar.
The report describes spending as having peaked around June and July at many companies. Among firms on API plans, costs had roughly halved by the time of reporting, which the author attributed to a mix of model changes and use of open models. Claude and Opus 5.5 were frequently cited in the limited sample, alongside Cursor, Codex and GitHub Copilot. The source does not provide a representative industry-wide spending survey, so its findings should be read as a set of reported examples rather than a market average.
““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 an anonymized trading firm, quoted in The Pragmatic Engineer
AI-powered code verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the PR Data Cannot Prove
The reported PR increases do not establish that AI tools caused the change. The source material does not give a control group, a consistent definition of a PR across participating companies, or evidence that the added PRs were merged, improved product quality or reduced delivery costs. Vendor platform data may also reflect the users and customers represented in those services, rather than software engineers as a whole.
The report does not provide a common ROI calculation across the 27 workplaces, including comparable data on tool costs, time saved, defect rates and business outcomes. It also describes some spending and model details as company-reported or anonymized. The precise scale and persistence of the reported patterns across the wider industry remain unknown.
software engineering PR management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Tracking Costs Against Outcomes
The report points to a practical next step for companies: compare AI bills with outcomes beyond raw PR counts, while documenting what is included in each cost figure. That could include engineering time, review effort, software quality and the cost of maintaining AI-generated code. The source does not announce a shared measurement standard or a follow-up study with a set release date.
Companies facing the end of subsidized subscriptions will also need to decide whether to absorb higher bills, impose tighter limits or shift workloads to other models. Whether any of those choices produces a positive return remains a question the report leaves open.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did the report find about AI spending per engineer?
Spending varied widely across the 27 workplaces surveyed, from below $300 to more than $1,000 per engineer monthly in the examples described. The report says company size did not correspond to spending levels.
Did AI tools cause PR volumes to rise?
The report gives vendor data showing higher PR counts among some platform users, including CodeRabbit’s reported 113% rise in weekly PRs. It does not establish that AI caused the increase; the figures are observational and may reflect other factors.
Does more PR activity mean a positive ROI?
Not on its own. PR counts measure submission activity, not whether code is better, takes less time to review, reduces costs or improves business results. The report says few companies could quantify AI’s ROI.
Why might AI bills change?
The report says API costs had roughly halved from summer peaks at some companies, which its author linked to model changes and open-model use. It also says some subsidized subscriptions are ending, potentially increasing bills for companies that do not adjust their use or plans.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
