🔍 Read the full analysis: AI Coding: A Cost-and-Value Check For Your Team on ThorstenMeyerAI.com
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
TL;DR
A survey of engineering leaders and senior developers at 27 workplaces found monthly AI coding budgets ranging from $150 to about $3,000 per developer, with spending and productivity results varying sharply. The reporting points to clear gains in some tasks but says many companies cannot yet connect AI tool costs to delivered value.
Engineering teams surveyed across 27 workplaces reported monthly AI coding budgets ranging from $150 to about $3,000 per developer, while companies reported mixed changes in output and often lacked a clear return-on-investment measure. The findings, published by Thorsten Meyer on October 23, come as firms prepare 2027 budgets and some discounted subscriptions are due to expire.
The survey and reported examples were drawn from an October 20 article by Gergely Orosz of The Pragmatic Engineer, which included interviews with engineering leaders and senior engineers at 27 workplaces, plus aggregate figures from AI code-review companies CodeRabbit and Gitar. Meyer said the figures cited in his account came from the freely readable portion of Orosz’s article. The available material does not establish that the workplaces form a representative sample of all software companies.
Spending differed sharply even among organizations with large engineering teams. One publicly traded company with about 1,500 engineers capped spending at $150 per developer monthly. A two-person European .NET agency, Ix Technologies, spent about $3,000 per developer monthly. Another example was a quantitative trading firm whose monthly AI spend reportedly rose from $500,000 in January to $8 million in September; the article said traders, rather than engineers, were its largest spenders.
Reported productivity measures did not line up. CodeRabbit said developers on its platform created 113% more pull requests per week than a year earlier, while a Microsoft study of tens of thousands of engineers during an early-2026 rollout found adopters merged roughly 24% more pull requests over four months than they otherwise would have. Microsoft’s authors cautioned that a merged pull request is not the same as delivered value. Among the surveyed workplaces, results ranged from flat pull-request frequency to increases of several times.
Infographic 01 · Where the money goes
From a $150 cap to $3,000 a month per developer.
Anecdotal data from 27 workplaces, mostly anonymized. It shows how wide the range is, not an industry average.
Spend per developer per month
Examples from the free part of the article · scale to $4,000
actual or typical spendbudget cap
What changed in 2026
Why the 2027 budget needs a range
- Peak in June and July: spend has roughly halved since for companies on API plans (Opus 5.5 and open models).
- Caps are common at medium and high spenders, and most people say they are easy to get around.
- Subsidies are ending: some companies must optimize or see the bill rise 5–10×.
- GitHub Copilot bills by token since 1 June 2026, at each model’s API rate, with no fallback to a cheaper model.
- Company size does not predict spend.
ROI is “a question that few companies seem able to answer – at least for now.”Gergely Orosz, The Pragmatic Engineer, 20 Oct 2026
Sources: The Pragmatic Engineer, “What’s the ROI of software engineers’ spending on AI?” (20 Oct 2026, free part); GitHub (27 Apr 2026).
Powered by Thorsten Meyer AI
Why AI Coding Bills Need Outcome Measures
The spending range matters because subscription prices alone do not describe total AI costs. Usage-based billing, model choice and agent activity can change bills, while a cap may limit spending without showing whether work improved. Meyer’s source says some companies expect discounted subscriptions to end, potentially raising their costs if usage and plans remain unchanged; the cited account does not provide a common, independently verified estimate of how many firms face that change.
The evidence also highlights a gap between activity and value. More pull requests can reflect faster delivery, but can also mean larger review queues or more rework. The source recommends that teams compare spending with measures such as cost per accepted change, rework, change-failure rates and whether completed work keeps pace with incoming stories. Those measures do not, by themselves, establish that AI caused an improvement, but they offer a more direct basis for budget decisions than token use or raw code volume.
According to the reported examples, teams described gains in customer-issue triage, understanding unfamiliar code, prototyping and backlog cleanup, especially on simple, consistent technology stacks. The reported benefits were less clear for complex legacy systems. That distinction can help engineering leaders decide where to test or expand tools, while avoiding a blanket conclusion that the same budget or productivity result applies across teams.
Products Worth Considering
DJI Mini 3 Drone with DJI RC and 3‑Axis Mechanical Gimbal
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
Samsung 32-inch Odyssey G5 G51F QHD 180Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How Spending and Productivity Evidence Diverges
The underlying discussion comes as companies revisit both AI tool pricing and their internal budgets. The source says spending at companies on API plans peaked in June and July and had roughly halved since, attributing the decline to a switch to a newer model and a move toward open models. That explanation is an account of Orosz’s reporting, not a controlled comparison of costs across companies.
Pricing arrangements are also changing. The source reports that on June 1, GitHub moved Copilot plans to usage-based billing, charging tokens at each model’s API rate and removing a fallback to a cheaper model. GitHub’s stated rationale was: “Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.” The change makes usage and model selection relevant to forecasting costs, although the source does not detail how the change affected each customer’s bill.
Other productivity research cited in Meyer’s account offers a mixed picture. METR’s early-2025 study found experienced open-source developers took 19% longer with AI. A February follow-up estimated an 18% speedup for returning developers and a 4% speedup for new developers, but both confidence intervals included zero. METR described the data as unreliable, in part because developers held back some tasks they preferred to do without AI. These results make broad claims about productivity uncertain.
“A question that few companies seem able to answer – at least for now.”
— Gergely Orosz, summarizing company return-on-investment evidence in The Pragmatic Engineer, as reported by Meyer
Where the Return Evidence Falls Short
The reported survey does not show that its 27 workplaces represent the wider software industry, and the available source material does not provide a single standardized measure of AI’s financial return. Spending figures and productivity reports are not directly comparable: companies use different tools, billing plans, tasks and definitions of output.
Vendor figures from CodeRabbit describe activity on its own platform, while Microsoft’s study measures merged pull requests rather than downstream customer or business results. METR’s estimates have substantial uncertainty and disclosed limits. The cited Veracode test found that 44% of AI coding tasks produced a known vulnerability, according to the source material, but the supplied account does not include enough information about the test’s methods or task set to apply that rate to all AI-written code.
It is also unclear how much reported time saved becomes additional delivered work after review, testing, security checks and maintenance. The source includes examples of faster triage and backlog work, but does not provide comparable cost-and-quality data for those cases. The timing and scale of subscription subsidy expirations, and their impact on company budgets, remain unspecified.
What Teams Can Track in 2027
Companies setting 2027 budgets can compare spending with outcomes at the level of specific teams and tasks. Measures suggested in the source include cost per accepted change, rework, change-failure rates and the balance between stories completed and stories arriving. Tracking quality and review effort alongside volume can show whether faster code production is creating useful, maintainable work.
The next step is to assess those measures over time and across comparable work, while recording which tools, models and billing plans were in use. That would help separate tool-related changes from differences in task difficulty or team practices. The available reporting does not identify an agreed industry-wide measurement standard or a scheduled follow-up study, so how companies will report returns remains open.
Key Questions
How much are teams spending on AI coding tools?
The 27-workplace survey cited by Meyer reported monthly budgets ranging from $150 to about $3,000 per developer. Those are examples from the surveyed workplaces, not an industry-wide average.
Does the evidence show that AI coding makes developers more productive?
It shows mixed results. CodeRabbit and a Microsoft study reported increases in pull-request activity, while results across surveyed companies ranged from flat output to much higher counts. Pull-request volume does not by itself establish improved quality or business value.
Where did teams report the clearest benefits?
The source describes reported gains in issue triage, code archaeology, prototyping and backlog cleanup, particularly with simple, uniform technology stacks. These are examples reported by organizations, not controlled findings that apply to every team.
What should companies measure before expanding AI tool budgets?
The source recommends tracking cost per accepted change, rework, change-failure rates and completed stories compared with incoming work. Teams can also track review and security effort so higher code volume is not mistaken for more delivered value.
Source: ThorstenMeyerAI.com
