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 surveyed engineering leaders and senior engineers at 27 workplaces about AI-tool spending and changes in pull-request output. Spending varied widely, and data from CodeRabbit and Gitar showed higher PR volumes among some users and teams, but the report found that few companies could clearly quantify AI’s return on investment.
The Pragmatic Engineer has published a report on AI-tool spending at 27 workplaces, finding that costs per engineer vary widely while many companies still cannot say whether the spending produces a measurable return. The report also cites platform data showing more pull requests (PRs) among developers using AI tools, but that output measure alone does not establish that AI caused the increase or improved business results.
Among the companies surveyed, monthly spend ranged from below $300 per engineer at lower-spending organizations to more than $1,000 per engineer among some high spenders. The report found no clear relationship between company size and spending: it describes a two-person .NET shop spending $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped spending at $150 per developer. Most company names were withheld at their request.
The report also describes changing costs. Several companies said spending peaked in June and July and later fell; for those on API billing, costs were roughly halved, according to the report. Respondents attributed reductions to a combination of using open models and changes in model use. Subsidized subscriptions can make per-tool costs less visible, while some companies said subsidies were ending and could leave them facing bills five to ten times higher if they did not cut costs or change plans.
CodeRabbit shared anonymized data from engineers using its platform. It reported a 113% rise in average weekly PRs and a 127% increase in monthly PRs; in a tracked group of 20,000 developers, monthly PRs rose 128% over a year. Separately, Gitar, an AI code verification agent, reported that some “AI-native” teams ship more than 60 PRs per developer per week. These are platform observations, not controlled evidence that AI tools alone produced the increases.
Costs Rise Faster Than ROI Evidence
The findings matter to companies deciding how much to spend on AI coding tools and whether to make them standard for engineering staff. A large range in reported costs suggests that a single per-engineer budget may not reflect different workloads, tools, billing models or company policies. Spending limits are common among medium and higher spenders, but respondents told the report that caps can often be avoided.
Higher PR volume may indicate faster or more frequent code changes, yet it does not by itself show that software is more reliable, that teams complete valuable work faster, or that revenue or productivity has improved. PR counts can also reflect differences in team practices and what each organization considers a pull request. The report’s central finding is that few companies can yet give a firm ROI figure, leaving decision-makers to weigh visible invoices against less certain productivity effects.
As an affiliate, we earn on qualifying purchases.
How the Report Measured Spending
The Pragmatic Engineer says it gathered spending and PR information through conversations with engineering leaders and senior engineers at 27 workplaces, anonymizing most companies. It also collected aggregate figures from CodeRabbit and Gitar. The evidence combines company-reported costs with platform-level output data; it is not described as a representative survey of the software industry or a controlled comparison of AI users with non-users.
The report uses PR throughput because pull requests are readily available in many engineering organizations and can be compared across time. But its author also notes that team culture affects how many PRs developers submit: norms differ between organizations, and employees may adapt to local expectations. The report says Claude and Opus 5.5 featured prominently in respondents’ spending, alongside Cursor, Codex and GitHub Copilot. Those observations reflect this limited sample, not a ranking of tools across the industry.
““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 a trading firm, quoted by The Pragmatic Engineer
software engineering productivity tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What PR Counts Cannot Establish
The reported PR increases do not show how much of the change was caused by AI. The source material does not provide a matched comparison of AI users and non-users, nor does it establish whether higher PR volume improved code quality, reduced delivery time or raised profits. Platform datasets may also reflect the behavior of their own users rather than developers generally.
Company spending figures are self-reported and the sample covers 27 workplaces, with many identities anonymized. The report does not provide a common method for calculating ROI across those organizations. It also remains unclear how costs will change as subsidies end, tool use shifts, and companies move between commercial and open models.
As an affiliate, we earn on qualifying purchases.
Budgets and ROI Tracking Ahead
The report describes companies continuing to review AI budgets, set or revise spending caps, and prepare for the end of some subsidized subscriptions. Organizations may face a choice between limiting usage, optimizing model selection, or accepting higher bills when discounted access ends. The source does not identify a single upcoming date when these changes will take effect across the surveyed companies.
For readers comparing tools or budgets, the next useful evidence would pair spending with outcomes such as delivery time, defects, maintenance work and business value—not PR volume alone. The report provides a snapshot of current practices, but whether AI spending pays off will depend on company-level tracking that the surveyed organizations have not yet consistently reported.
developer pull request management 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 engineer?
The report describes a wide range: some companies spend under $300 per engineer per month, while examples among high spenders exceed $1,000. One two-person shop reported $3,000 per developer monthly. These are examples from 27 workplaces, not an industry-wide average.
Did AI tools cause the reported increase in pull requests?
The report does not establish causation. CodeRabbit reported higher PR counts among developers using its platform, but the data does not isolate AI’s effect from changes in team practices, workloads or other factors.
Does more PR activity prove AI is improving productivity?
No. PR volume measures how frequently code changes are submitted, but does not show whether those changes are useful, reliable or less costly to maintain. The report says few companies can clearly quantify overall ROI.
Why could AI bills increase for some companies?
Some organizations rely on subsidized subscriptions that are ending, according to the report. It says companies facing that change may need to optimize spending or could see bills rise five to ten times, depending on their arrangements.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
