🔍 Read the full analysis: A Team’s Guide To Tracking AI Coding Costs And Benefits on ThorstenMeyerAI.com
TL;DR
A report published by The Pragmatic Engineer on Oct. 20 surveyed engineering leaders and senior engineers at 27 workplaces, finding monthly AI coding budgets ranging from $150 to about $3,000 per developer. Reported productivity results vary widely, while researchers and companies caution that more pull requests do not necessarily mean more delivered value. Teams are being urged to compare costs with accepted changes, rework and delivery outcomes.
A survey of 27 workplaces found that engineering teams spend from $150 to about $3,000 per developer per month on AI coding tools, while reported productivity results range from flat pull-request counts to substantial increases. The findings, published Oct. 20 by The Pragmatic Engineer, arrive as companies set 2027 budgets and face the question of whether the tools’ costs are matched by useful work.
The survey, by The Pragmatic Engineer’s Gergely Orosz, draws on interviews with engineering leaders and senior engineers, plus aggregate data from code-review vendors CodeRabbit and Gitar. The source article says the cited findings come from its freely readable portion. Spending does not map neatly to company size: examples range from a publicly traded company with 1,500 engineers capping monthly spend at $150 per developer to a two-person European agency spending about $3,000 per developer.
Reported output measures also differ. CodeRabbit said developers on its platform produced 113% more pull requests per week than a year earlier. Microsoft research tracking 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. Microsoft’s authors cautioned that a merged pull request does not establish the value of the work delivered.
Across the surveyed workplaces, pull-request activity ranged from unchanged to five or ten times higher. In one company, frequency stayed flat while pull requests became three times larger. The report also describes specific uses where teams said they see gains: customer-issue triage, understanding why existing code was written, prototyping and clearing backlogs on simple, uniform technology stacks. Those examples are reported experiences, not a single controlled measure of return.
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 cost range matters because companies are deciding whether to renew subscriptions, expand access or absorb higher usage charges. The source article says several companies expect discounted subscriptions to end; some could see bills rise five- to tenfold if they do not optimize or change their use. It also reports that spending on API plans peaked in June and July and had roughly halved since, which Orosz attributed to Opus 5.5 and a shift to open models.
Counting tokens or pull requests alone can reward activity without showing whether work reaches users or reduces engineering effort. The article recommends tracking cost per accepted change, rework, change-failure rates, and stories closed compared with stories arriving. These measures could help teams distinguish faster production from valuable delivery, while accounting for the time spent reviewing and correcting AI-generated code.
There are also quality concerns. The source cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability. That result does not establish the security rate for every tool or workplace, but it underlines why review capacity and security checks belong in any assessment of cost and benefit.
As an affiliate, we earn on qualifying purchases.
Billing Changes Meet Mixed Evidence
The report lands amid a shift toward usage-based pricing. It says GitHub changed every Copilot plan on June 1 to charge for tokens at each model’s API rate and removed the fallback to a cheaper model. GitHub explained the change by saying agentic use is becoming the default and carries higher compute and inference demands.
Evidence about productivity remains unsettled. 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 4% for new developers, but both estimates had confidence intervals that included no effect. METR also described the data as unreliable, citing task selection that may have favored work developers were willing to do without AI.
Survey responses can tell a different story from controlled tests. In a METR survey of 349 technical workers, the median reported a 1.4-to-2-times change in the value of their work. METR said participants in its 2025 research had overestimated AI’s effect by 40 percentage points. These measures use different methods and populations, so they should not be treated as directly comparable estimates of a single productivity gain.
As an affiliate, we earn on qualifying purchases.
Productivity Gains Still Lack a Common Measure
The survey does not establish a single productivity effect that applies across engineering teams. Its examples differ in company size, software stack, work type and measurement method. Vendor platform data may reflect teams already using AI review tools, while self-reports and controlled experiments have different limitations.
It is also unclear how often faster coding leads to more accepted, secure changes or better outcomes for customers. The source material does not provide a consistent accounting of tool costs alongside review time, rework, defects and delivery value for all 27 workplaces. The cited vulnerability test likewise does not show that 44% of all AI-generated code is vulnerable; it refers to tasks in Veracode’s test.
Some company-specific figures, including a quant firm’s reported rise in monthly AI spending from $500,000 in January to $8 million in September, are examples in the source article rather than independently established industry-wide totals. The basis for comparing spending across firms is not fully specified.
As an affiliate, we earn on qualifying purchases.
Teams Face 2027 Budget Decisions
Engineering leaders are setting budgets for 2027 while some discounted subscriptions approach expiry and vendors charge more directly for model usage. The next practical step for a team is to establish a baseline before changing access or spending: record costs and delivery measures over a defined period, then compare similar work with and without AI support where feasible.
Teams can track accepted changes alongside rework, failure rates, review time and the flow of work into and out of backlogs. They can also separate results by task, since the source examples suggest that triage, code archaeology, prototypes and backlog cleanup may behave differently from work on complex legacy systems.
The available source material does not give a common industry timetable or a settled standard for these measurements. More comparable results will depend on companies reporting costs and outcomes together, rather than relying on usage totals or pull-request volume alone.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The surveyed examples range from a $150 monthly cap per developer to about $3,000 per developer per month. The report says company size alone did not predict spending.
Do the findings show that AI coding increases productivity?
They show mixed results, not one reliable gain for every team. Microsoft and CodeRabbit reported higher pull-request counts in their respective data, while some surveyed companies saw flat counts. A pull request is an activity measure, not proof of delivered value.
What should teams measure besides AI usage?
The source recommends tracking cost per accepted change, rework, change-failure rates and stories closed compared with stories arriving. Review time and security checks can help show whether faster code generation creates additional work elsewhere.
Where does AI coding appear most useful in the report?
Reported examples include triaging customer issues, tracing the history behind existing code, building prototypes and clearing backlogs on simple, uniform stacks. These are workplace accounts, not a guarantee that the same gains will apply to other teams.
What remains unknown about AI coding returns?
The report does not provide a consistent, cross-company calculation linking tool spending to accepted, secure work and business outcomes. The size and durability of productivity gains across different tasks and engineering teams remain unsettled.
Source: ThorstenMeyerAI.com
