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 engineering leaders and senior engineers at 27 workplaces, alongside aggregate data from CodeRabbit and Gitar. Some measures show substantially more pull requests, but the report says few companies can connect that output to AI spending or establish a clear return on investment.
A report from The Pragmatic Engineer finds that software companies are spending widely varying amounts on AI tools, while few can yet say what return that spending produces. Its comparison of information from 27 workplaces with aggregate platform data shows higher pull-request counts in some datasets, but does not establish that AI tools caused the increases or that they improved business results.
The report’s author spoke with engineering leaders and senior engineers at 27 workplaces about AI spending and changes in pull-request (PR) activity, and collected aggregate data from CodeRabbit and Gitar. PRs are a readily available measure of software changes submitted for review, but they do not by themselves show whether code is better, reaches users sooner, or reduces costs.
CodeRabbit’s anonymized data showed a 113% increase in weekly PRs among data engineers using its platform and a 127% increase in monthly PRs. A separate comparison tracking 20,000 developers over a year found a 128% increase. Those figures describe activity in the reported groups; the material does not show that AI alone produced the increases.
Gitar’s data from hundreds of customers grouped teams by PR output, with its report saying AI-native teams ship more than 60 PRs per developer per week. The article also describes sharply different spending practices: one two-person .NET company spends $3,000 per developer monthly, while a publicly traded company with 1,500 engineers limits monthly AI spending to $150 per developer. The author found no relationship between company size and spend in this sample.
Higher PR Counts Do Not Prove ROI
The figures offer companies a way to track whether software-development activity changes as AI tools spread, but more PRs are not the same as more value. PR counts do not account for complexity, defects, review time, maintenance work, or whether a change helps customers. They can also differ because teams have different working practices and expectations.
The report’s central finding is a measurement gap: organizations are tracking costs and output, but many cannot connect the two to a reliable financial or productivity outcome. That matters to engineers and employers facing decisions about subscriptions, usage limits, and expanding access. Without a clearer comparison, a higher bill or a higher PR count alone cannot show whether a tool is worth its cost.
AI coding tools for software engineers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Wide Spending Differences Across Teams
The interviews describe spending caps as common among companies with medium or higher AI budgets, often with additional spending requiring approval. The report says these limits can be easy to work around. It also says spending at companies using API plans peaked in June and July and had roughly halved by the time of reporting, with respondents citing a move to open models and changes in model use.
Subscription plans can obscure per-engineer costs, according to the report, and some employer subsidies are ending. Companies facing the end of those subsidies may need to reduce usage or absorb higher bills; the report says some expect costs could rise five- to tenfold. Claude and Opus 5.5 were frequently named as major spending destinations in this limited set of responses, alongside Cursor, Codex, and GitHub Copilot.
One anonymized trading firm reported spending $8 million a month in September, up from $500,000 in January, and average spending of $1,600 per developer monthly, compared with about $80 in January. A source at the firm said traders were among the heaviest users and described rate limiting as a concern. These are details from one company, not an industry-wide benchmark.
“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 anonymized trading firm, quoted by The Pragmatic Engineer
As an affiliate, we earn on qualifying purchases.
Causation and Business Returns Remain Unclear
The reported PR increases do not isolate AI’s effect from other changes, such as team growth, new processes, or changing expectations. The sources also do not provide a consistent before-and-after measure of code quality, delivery time, maintenance burden, or revenue across the 27 workplaces. Whether higher PR volume represents a net productivity gain remains unknown.
The data is drawn from a limited set of companies and platform users, and the spending examples are not a representative industry survey. The report does not establish a standard ROI calculation or provide comparable financial outcomes for each company. It also does not clarify how much of the observed change reflects AI tools rather than differences in team culture and work type.
As an affiliate, we earn on qualifying purchases.
Companies Face Cost and Measurement Decisions
Companies will need to decide whether to keep expanding AI access, introduce tighter usage controls, or shift workloads to less expensive models as subscriptions and subsidies change. The report describes some firms preparing for higher bills as subsidized plans end, while others are using spending caps or moving toward open models.
The next useful evidence would compare tool costs with outcomes beyond PR counts, including review and rework time, defects, delivery speed, and the business value of shipped work. Until those measures are tracked consistently, the report’s figures provide a snapshot of spending and activity—not a definitive answer on whether engineers’ AI spending pays off.
software development analytics platform
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What does the report say about AI spending per engineer?
It found wide differences among the 27 workplaces it contacted. Examples ranged from $150 per developer monthly at a large public company to $3,000 per developer monthly at a two-person .NET company; these examples are not industry averages.
Did AI tools cause the increase in pull requests?
The report does not establish that. CodeRabbit’s datasets recorded higher PR counts, but the material does not isolate AI from other possible causes, including changes in team practices or workload.
Do more pull requests prove that AI has a positive ROI?
No. PR volume measures activity, not the value or quality of the work. The report says few companies can yet connect AI spending to a clear return.
Which AI tools or models did companies report using?
Respondents frequently mentioned Claude and Opus 5.5, as well as Cursor, Codex, and GitHub Copilot. The findings reflect the report’s limited set of company responses, not a market-wide ranking.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
