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 pull request (PR) output. Data shared by CodeRabbit and Gitar show higher PR volumes among users or teams, but the report says few companies can yet answer whether that extra activity produces a measurable financial return.
The Pragmatic Engineer has published a survey of AI-tool spending at 27 workplaces, alongside platform data showing higher pull request (PR) volumes among developers using AI tools. The report finds that reported spending varies widely and that few companies can yet quantify AI’s return on investment, leaving higher output counts short of proof that the tools deliver equivalent business gains.
Aggregate data shared by CodeRabbit showed a 113% increase in weekly PRs and a 127% increase in monthly PRs among developers using its AI code-review platform, compared with a year earlier. For a tracked group of 20,000 developers, the report cited a 128% rise over a year. These are platform observations, not a controlled test establishing that AI caused the increases.
Gitar, an AI code-verification agent, shared data across hundreds of customers. The report says some teams categorized as “AI-native” ship more than 60 PRs per developer per week. The figures describe different teams and workplace practices; PR counts alone do not show whether code is more reliable, whether work is more valuable, or how much time it saves.
The 27-workplace spending sample included companies with sharply different budgets. A two-person .NET company reported spending $3,000 per developer per month, while a publicly traded company with 1,500 engineers capped monthly spending at $150 per developer. One trading firm reported total AI spending of $8 million per month in September, up from $500,000 in January, and average spend of $1,600 per developer per month. The report says traders, rather than engineers, accounted for the largest costs there.
Higher PR Counts, Unproven Returns
The report highlights a measurement problem facing companies expanding access to coding assistants: spend is easier to count than value. A rising bill can be tracked directly, while connecting that bill to faster delivery, lower costs, better software or increased revenue requires additional evidence.
PR volume offers a readily available activity measure, but it is not a complete productivity measure. A greater number of smaller changes may raise the count without increasing the amount of useful work shipped. Quality, rework, review burden and the business impact of completed changes also matter. The report does not establish that the platform-reported increases translate into those outcomes.
The spending range also suggests that companies are making decisions under different circumstances. Some impose per-person caps and require additional spending to be justified; others subsidize subscriptions or allow much larger budgets. The report says some subsidies are ending, potentially exposing costs that were previously less visible. For engineering leaders, the practical issue is how to set budgets and evaluate results without treating raw usage or PR counts as proof of ROI.
As an affiliate, we earn on qualifying purchases.
How the Report Measured AI Use
The Pragmatic Engineer says the investigation began with a reader asking what companies spend on AI per engineer and how PR throughput changes before and after adopting AI. The author spoke with engineering leaders and senior engineers at 27 workplaces and gathered aggregate statistics from CodeRabbit and Gitar. Most company names were withheld at their request.
The comparison uses PR throughput because it is commonly available in engineering systems and can be compared across time. However, the sources described different datasets: CodeRabbit supplied anonymized user data, while Gitar shared figures from customer teams. The report’s year-over-year comparisons do not amount to a uniform, independently controlled study of AI adoption.
Spending patterns in the interviews included heavy use of Claude and Opus 5.5, as well as Cursor, Codex and GitHub Copilot. The report also says some companies’ costs on API plans fell after summer peaks, which respondents attributed to a mix of model changes and moves to open models. These are reported company experiences, not a single market-wide spending index.
““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 trading firm, quoted in The Pragmatic Engineer
software development AI assistants
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Output Figures Cannot Show
The report does not establish that AI tools caused the higher PR counts, or that the additional PRs represent more valuable work. It does not provide a common measure of code quality, defects, review time, delivery speed or financial benefit across the 27 workplaces. The platform figures also cover their own users or customer teams, rather than a representative sample of all software engineers.
Spending comparisons may be incomplete where companies rely on subsidized subscriptions, and the report says some subsidies are approaching their end. It is not clear how much reported spending includes non-engineering use, as the trading-firm example indicates that other employee groups can be major users. The source material also does not provide a consistent time period or accounting method for every company’s figures.
AI-powered code verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Companies Face Budget Reviews
The report points to upcoming budget decisions as subsidies expire and companies decide whether to absorb higher bills, limit usage or shift workloads to other models. Some respondents described spending caps, with additional expenditure requiring justification; the trading firm said it was building internal GPU and CPU capacity and developing models. The report does not give a timeline or cost comparison for that work.
For a clearer ROI assessment, companies will need to compare spending with outcomes beyond PR counts, using consistent periods and accounting for changes in team size, project mix and code quality. Whether the reported productivity gains persist—and whether they offset tool, infrastructure and review costs—remains unanswered by the available data.
developer productivity tools with AI
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 tools per software engineer?
The report found a wide range among 27 workplaces, from a company cap of $150 per engineer per month to a two-person firm reporting $3,000 per developer per month. These examples are not a market average.
Did AI tools cause PR volumes to rise?
The report cites higher PR counts in data shared by CodeRabbit and Gitar, but those observations do not prove that AI caused the increase. They also do not show whether the extra PRs improved software quality or business results.
What did CodeRabbit’s data show?
CodeRabbit shared anonymized platform data showing weekly PRs up 113% year over year and monthly PRs up 127%. A tracked group of 20,000 developers recorded a 128% increase over a year.
Do companies know the ROI of AI coding tools?
The report says few companies it contacted could answer that question. Many can track spending and activity, but the evidence presented does not establish a consistent financial return.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
