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 figures from CodeRabbit and Gitar. CodeRabbit reported weekly PRs up 113% among surveyed users, but the report says companies generally cannot yet establish whether higher output outweighs AI costs.
The Pragmatic Engineer’s survey of 27 workplaces found sharply different levels of spending on AI coding tools, while companies rarely had a clear measure of financial return. Separately, platform data shared with the publication showed more pull requests among developers using AI tools, but that output measure alone does not establish that the spending paid off.
The report combines interviews with engineering leaders and senior engineers at 27 companies with aggregate data supplied by CodeRabbit and Gitar. The organizations were mostly anonymized at their request. The figures show no apparent link between company size and per-engineer spending: one two-person .NET business reportedly spent $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped spending at $150 per engineer per month.
CodeRabbit shared anonymized data from engineers using its platform. It reported an average 113% rise in weekly pull requests and a 127% increase in monthly PRs. For a tracked group of 20,000 developers, weekly PR output rose 128% over a year. These are platform observations, not a controlled finding that AI caused the increase. Gitar, an AI code verification agent, reported that some teams classified as “AI-native” shipped more than 60 PRs per developer per week; the report does not establish that this rate applies broadly.
Spending patterns also shifted during the period covered. Interviewees said costs peaked around June and July and later fell by roughly half for some companies on API plans, which they attributed to model and infrastructure changes, including use of open models. The report says spending caps were common among medium and higher spenders, while some subsidized subscriptions were nearing their end, potentially exposing companies to substantially higher bills.
Why PR Counts Do Not Settle ROI
The findings point to a measurement problem for engineering teams: tool bills are visible, but productivity gains and their financial value are harder to establish. A rise in PRs may indicate that developers are shipping more changes, yet it does not by itself show that those changes are useful, maintainable, or delivered faster overall.
For companies, that distinction matters as AI subscriptions and API usage scale. The report’s examples range from modest monthly limits to thousands of dollars per engineer, and it says some subsidized plans may end. Leaders deciding whether to expand access need to weigh spending against outcomes such as delivery time, review burden, defects, and business value—not rely on PR volume alone. The report does not supply a common accounting method or a verified return figure across the 27 organizations.
As an affiliate, we earn on qualifying purchases.
How the Report Compared Spending
The Pragmatic Engineer said an engineer asked how much other companies spend per developer and whether PR throughput changed before and after adopting AI. The publication sought answers from leaders and senior engineers, then added platform-level data from CodeRabbit and Gitar. It described PR counts as a readily available measure that can be compared over time, while acknowledging that team cultures differ in their normal pace of work.
The report also described variation in tools and payment models. Interviewees named Claude and Opus 5.5 among commonly used models, as well as Cursor, Codex, and GitHub Copilot. Some companies pay for API usage, while others use subscriptions or subsidies, making direct comparisons difficult. One high-spending trading firm cited in the report said its overall AI expenditure had reached $8 million per month in September, compared with $500,000 in January; the source attributed much of its spending to traders, not engineers.
““Lots of traders are using Claude. Engineering is learning the importance of rate limiting for traders who accidentally run expensive queries without realizing it.””
— A current engineer at the trading firm, quoted by The Pragmatic Engineer
As an affiliate, we earn on qualifying purchases.
What the Available Metrics Miss
The data does not show whether AI tools caused the reported rise in PRs. CodeRabbit’s figures cover platform users and a tracked cohort, while Gitar’s figures describe customer teams; neither is presented as a randomized comparison with developers who did not use AI. The report also does not give a consistent definition of an AI user or a comparable baseline for every company.
It remains unclear how much of the increased output represents valuable production work, whether review and maintenance costs changed, and how spending translates into revenue or saved engineering time. The reported spending figures are based largely on interviews and selected company examples, not a representative survey of the software industry. The report’s model and pricing observations are tied to the period it examined and may change.
As an affiliate, we earn on qualifying purchases.
Companies Face Cost Reviews
The report says companies are preparing to review costs as subsidized subscriptions expire and usage bills become more visible. Some interviewees described caps or approval requirements for spending above set limits; others were exploring cheaper models or internal computing capacity. The high-spending trading firm said it was building internal GPU and CPU farms, though the report provides no independent assessment of the expected savings or returns.
The next useful evidence would be company-level comparisons that track spending alongside delivery speed, software quality, developer time, and business outcomes over a defined period. Until organizations publish or establish those measures, the survey offers a snapshot of rapidly varying costs and rising PR counts—not a general answer to whether AI coding tools generate a positive return.
developer productivity analytics tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI per software engineer?
The report found wide variation, from a company cap of $150 per engineer per month to a two-person .NET business reportedly spending $3,000 per developer per month. These examples are not industry averages.
Did AI use increase pull request output?
CodeRabbit reported increases in PR counts among developers using its platform, including a 113% rise in weekly PRs. The report does not establish that AI alone caused the increase.
Does higher PR volume prove AI tools have a positive ROI?
No. PR volume measures activity, not whether the work created business value or reduced total costs. The report says few companies could answer the ROI question clearly.
Why could AI bills rise for some companies?
Some companies rely on subsidized subscriptions that may end, while API usage can vary with how often and how intensively employees use models. The report says spending caps were common among medium and higher spenders.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
