🔍 Read the full analysis: Is AI Coding Paying Off For Your Business? Here’s How To Measure on ThorstenMeyerAI.com
Get privacy and security gear 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 report published by The Pragmatic Engineer on October 20 surveyed engineering leaders and senior engineers at 27 workplaces, finding AI coding budgets range from about $150 to $3,000 per developer per month. Output measures vary widely, so teams should judge returns by accepted work, rework and delivery outcomes—not tool usage or pull request counts alone.
A survey of 27 workplaces found that engineering teams spend from about $150 to $3,000 per developer per month on AI coding tools, while reported changes in output range from flat to several times higher. The findings, published October 20 by The Pragmatic Engineer, underscore a budgeting problem for companies planning 2027 spending: many track tool costs and activity but cannot show whether AI-assisted work produces more useful, reliable software.
The survey drew on interviews with engineering leaders and senior engineers, plus aggregated figures from AI code-review vendors CodeRabbit and Gitar. The report says company size did not predict spending. A publicly traded company with 1,500 engineers capped monthly use at $150 per developer, while a two-person European .NET agency, Ix Technologies, spent about $3,000 per developer. Other companies set caps between $1,000 and $4,000, with typical reported use around $1,000.
Output figures differ by source and workplace. CodeRabbit reported that developers on its platform created 113% more pull requests per week than a year earlier, but that is vendor data from teams using its review product. A Microsoft study of tens of thousands of engineers during an early-2026 rollout found adopters merged about 24% more pull requests than they otherwise would have over four months. The study’s authors cautioned that a merged pull request does not necessarily represent delivered value.
Among the 27 workplaces, some reported flat pull-request frequency, while others described increases ranging from twofold to five- or tenfold. In one company, pull-request frequency stayed flat as average pull-request size tripled. The source article also cites a Veracode test in which 44% of AI coding tasks produced a known vulnerability, a finding that makes security and review capacity part of any productivity calculation.
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
Budgets Need Delivery Measures
The figures matter because AI coding costs are changing as usage shifts toward agents and more compute-intensive tasks. The source article says spending on API plans peaked in June and July and had roughly halved by the report’s publication, which author Gergely Orosz attributed to Opus 5.5 and a move to open models. It also says some discounted subscriptions are due to expire, potentially raising costs for companies that do not change plans or usage.
For budget owners, more code or more pull requests are not enough to establish a return. Teams need to connect expenditure to accepted changes, rework, change-failure rates and stories completed versus new stories arriving. Those measures can reveal whether AI is reducing cycle time and clearing useful work—or creating larger review queues and additional fixes. Security testing and human review must be counted alongside speed.
The source describes stronger results for bounded tasks such as issue triage, code archaeology, prototypes and backlog work on simple, consistent technology stacks. A SaaS company with about 600 engineers saw little change in merged-pull-request velocity, but a staff engineer said customer issue triage fell from sometimes a full day to minutes of agent work plus up to an hour of checking. That distinction shows why company-wide output measures can miss task-level gains, and why local improvements should not automatically be treated as overall financial returns.
As an affiliate, we earn on qualifying purchases.
Why AI Coding Costs Are Shifting
The Pragmatic Engineer’s findings arrive as companies set 2027 engineering budgets and reconsider subscriptions, usage limits and model choices. The report says some companies have caps that developers can get around, while API-plan spending has changed alongside model choices. These details are based on the article’s interviews and reporting; they do not establish a single spending pattern across the industry.
Billing changes add pressure to understand consumption. The source says GitHub moved every Copilot plan to usage-based billing on June 1, charging tokens at each model’s API rate and removing the fallback to a cheaper model. GitHub explained: “Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.” That makes usage tracking financially relevant, but it does not by itself show whether the resulting work benefits a business.
Research cited in the source is mixed. METR’s early-2025 study found experienced open-source developers took 19% longer with AI. Its February follow-up estimated an 18% speedup for returning developers and a 4% speedup for new developers, but both confidence intervals included zero. METR said the data was unreliable in part because 30% to 50% of developers held back tasks they did not want to do without AI. The results caution against treating any single productivity estimate as universal.
““a question that few companies seem able to answer – at least for now.””
— Gergely Orosz, The Pragmatic Engineer
As an affiliate, we earn on qualifying purchases.
Productivity Gains Remain Uneven
The available evidence does not establish a dependable return on investment across companies. The survey combines reported experiences from different workplaces, while vendor figures describe users of specific products and may not represent engineering teams generally. Microsoft’s result measures merged pull requests, not the financial or customer value of each change.
It also remains unclear how much reported output gains persist after accounting for review time, defects, security work and maintenance. The source article cites Veracode’s vulnerability finding but does not provide the test’s full methods or enough detail here to compare results across models and task types. METR’s estimates are uncertain, and its researchers flagged limitations in the data. A company’s own baseline and measurement method are needed before drawing conclusions about its results.
Some reported spending and output figures are individual company accounts rather than audited, comparable measures. The source material does not specify a common measurement window or baseline for every workplace figure. In particular, claims of much higher output should be treated as reported observations, not a general prediction for other teams.
AI-powered software development platforms
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Track Results Through 2027
Companies setting 2027 budgets can establish a baseline before expanding or cutting AI tool access: record spending, time to accepted changes, rework, change failures, security findings and the number of completed stories relative to incoming work. Comparing similar tasks and teams over a defined period can help separate a genuine improvement from a change in workload or reporting habits.
The next useful evidence will be whether those measures hold as subscriptions change, discounts expire and teams adopt more agent-based workflows. The source article does not identify a single follow-up study or reporting date. For now, businesses will need to evaluate their own results and account for both the time saved on routine work and the cost of review, correction and risk.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The survey cited by The Pragmatic Engineer found reported budgets ranging from about $150 to $3,000 per developer per month across 27 workplaces. It also described caps and spending patterns that varied by company; the figures are not a universal market average.
Do more pull requests prove that AI coding is paying off?
No. Pull-request counts measure activity, not necessarily useful or reliable delivery. Microsoft’s study authors said a merged pull request is not the same as the value it delivers. Teams should also track accepted work, rework, failures and whether completed work keeps pace with incoming requests.
Where did the surveyed companies report the clearest gains?
The source article describes reported benefits in customer issue triage, understanding existing code, prototyping and clearing backlogs on simple, uniform technology stacks. These are examples from particular companies, not proof that every team will see the same results.
What should a business measure when evaluating AI coding?
Track tool costs alongside cost per accepted change, rework, change-failure rates, security findings and stories completed versus stories arriving. A consistent baseline and measurement period help teams judge whether changes reflect AI use or other factors.
Are the research findings conclusive?
No. The reported results differ across workplaces and research methods. Some are vendor or company reports, and METR described limitations in its estimates. The evidence does not yet establish a single productivity gain or return that applies across engineering teams.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
