🔍 Read the full analysis: The Costs That Can Eat Into AI Coding’s Payoff on ThorstenMeyerAI.com
Get smart everyday buys 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 27 workplaces found AI coding budgets ranging from about $150 to $3,000 per developer per month, while companies report widely different changes in output. Reported gains are strongest for tasks such as triage and backlog cleanup, but growing usage bills, rework and security risks can reduce the return. Many employers still lack a reliable way to connect spending with delivered value.
Engineering teams are spending about $150 to $3,000 per developer each month on AI coding tools, but a new survey of 27 workplaces finds that many companies cannot yet say what that spending has delivered. The findings arrive as organizations set 2027 budgets and face changing subscription terms, uneven productivity results and costs tied to reviewing and securing AI-generated code.
The survey was reported by Gergely Orosz in The Pragmatic Engineer on October 20. It draws on interviews with engineering leaders and senior engineers at 27 workplaces, alongside aggregate figures from AI code-review vendors CodeRabbit and Gitar. The publicly available material describes spending caps ranging from $150 per developer monthly at one company to roughly $3,000 at a small European software agency. Some companies set caps between $1,000 and $4,000, while typical spending at those firms was reported at around $1,000 per developer.
Output measures do not point to a single result. CodeRabbit says developers on its platform created 113% more pull requests per week than a year earlier, a vendor-reported figure from teams using its service. A Microsoft study of tens of thousands of engineers during an early-2026 rollout found adopters merged about 24% more pull requests over four months than they otherwise would have, according to the study authors. They cautioned that a merged pull request is not the same as value delivered. Companies in Orosz’s survey reported results ranging from flat pull-request frequency to increases several times over.
Costs can rise with heavier use and may not be limited to subscriptions. The source says API-plan spending peaked in June and July and had roughly halved by the time of the report for some companies, a change Orosz attributes to a model release and a shift to open models. It also reports that some firms’ discounted subscriptions are due to expire, potentially increasing bills if usage and plans do not change. In a separate security finding cited by the source, Veracode reported that 44% of AI coding tasks in its latest test produced a known vulnerability. That result describes the test, not the rate for every company or codebase.
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
The business question is no longer only whether an AI tool can produce code faster. Companies need to know whether faster drafting leads to accepted changes, less rework and completed customer work after human review, testing and security checks. If pull requests grow in size without improving outcomes, a rise in activity may overstate productivity.
That distinction matters as firms move from discounted or fixed-price access toward charges more closely tied to model usage. GitHub said it changed Copilot plans to usage-based billing on June 1, charging tokens at each model’s API rate and removing the cheaper-model fallback. The company described the shift as a response to higher compute needs from agentic use. For employers, the combined bill may include tool charges and the engineering time spent checking output, correcting errors and managing risk.
The source points to more task-specific gains than a universal productivity boost. Teams described faster customer-issue triage, code archaeology and prototyping, as well as backlog cleanup in uniform codebases. Those cases can free time or move work that had been deferred. But results on complex legacy systems may be weaker, and the report does not establish that any one company’s gains will carry over to another.
As an affiliate, we earn on qualifying purchases.
What the Workplace Data Shows
The available evidence mixes different kinds of measurement. Vendor platform activity, company studies, interviews and controlled experiments do not measure the same thing, so their figures should not be treated as directly comparable. CodeRabbit’s pull-request count reflects its own platform; Microsoft’s figure comes from a study of its engineers and a particular rollout. Neither number on its own establishes a lasting change in business value.
Controlled results have also been uncertain. METR reported in early 2025 that experienced open-source developers took 19% longer with AI on tasks in its study. A February follow-up estimated an 18% speedup for returning developers and 4% 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. Those findings concern the study participants and tasks, not all software teams.
Examples in the survey show why task and setting matter. A staff engineer at a SaaS company with about 600 engineers said customer-issue triage that could take a day sometimes became minutes of agent work plus up to an hour of checking. At Ix Technologies, which uses a consistent C# and Blazor stack, the cofounder said AI helped a two-person team clear a long-standing backlog. These are reported company experiences, not controlled comparisons. The article also cites a multinational team that closed three times as many user stories monthly, but the source excerpt does not provide enough detail to establish the measurement method or period.
“A question that few companies seem able to answer – at least for now.”
— Gergely Orosz, summarizing the survey
As an affiliate, we earn on qualifying purchases.
Where the Return Remains Unproven
The survey does not establish a common, independently verified return on investment across the 27 workplaces. Spending figures and output measures vary, and the reported examples do not show how much additional review, rework or maintenance was required. It is also unclear how representative the interviewed companies are of engineering organizations overall.
The source’s claim that individual output is probably up by double digits is the author’s interpretation, not a single result shared by every study. METR’s findings carry stated limitations, while vendor figures come from users of those vendors’ products. The Veracode vulnerability rate applies to its test tasks; the source does not give enough detail here to translate it into a typical company’s production-code risk.
Future costs are also uncertain. The source reports that some discounts are due to expire and that usage-based billing can expose teams to higher charges, but it does not specify which companies will see increases or how large their eventual bills will be. There is no agreed benchmark in the material for comparing the cost of AI-assisted work with work done without these tools.
As an affiliate, we earn on qualifying purchases.
What Budget Owners Will Track
As companies set 2027 budgets, the report argues they should track outcomes rather than tool consumption alone. Measures it identifies include cost per accepted change, rework, change failure rate and the number of user stories completed compared with those arriving. These indicators can help distinguish more code activity from work that is accepted, reliable and useful.
The next test for individual teams will be whether gains persist when subscriptions, model choices and review processes change. Companies facing the end of discounts will need to compare the full cost of tools and oversight with the work delivered, while checking that security and quality practices keep pace. The source does not report a common timetable or standard for those evaluations, so how quickly employers can produce comparable results remains unclear.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The source reports budgets ranging from about $150 to $3,000 per developer per month among examples from a survey of 27 workplaces. Some companies set higher caps, but actual spending varies by employer and usage.
Does the research show that AI coding makes developers more productive?
It shows mixed results rather than a universal effect. Microsoft reported about 24% more merged pull requests among adopters in its study, while some companies in the survey reported flat activity. Pull-request counts do not by themselves establish improved product or business value.
What kinds of work appear to benefit in the reported examples?
Teams described gains in customer-issue triage, code archaeology, prototyping and backlog cleanup, particularly on simple, consistent technology stacks. These are reported experiences, not proof that the same results will occur at other companies.
What costs can reduce the payoff?
Usage-based charges, expiring discounts, human review, rework and security testing can all affect the total cost. The source does not quantify those costs consistently across the companies surveyed.
How should companies measure AI coding results?
The report suggests tracking cost per accepted change, rework, change failure rates and stories completed compared with stories coming in. Those measures focus on delivered work rather than tool usage or raw code activity alone.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
