🔍 Read the full analysis: Why AI Coding Costs Vary—and How To Measure The Benefits 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
AI coding budgets in a survey of 27 workplaces ranged from about $150 to $3,000 per developer each month, with costs shaped by usage and tools rather than company size alone. Output measures conflict, and higher pull-request counts do not by themselves show delivered value; useful measures include accepted changes, rework and change failures.
Engineering teams are spending about $150 to $3,000 per developer each month on AI coding tools, but a review of 27 workplaces and available productivity studies found no consistent way to tell how much of that spending becomes useful software. The findings matter as companies set 2027 budgets and some discounted subscriptions near expiration, potentially raising costs for teams that do not change plans or usage.
The range comes from engineering leaders and senior engineers interviewed by Gergely Orosz of The Pragmatic Engineer, alongside aggregate figures from code-review vendors CodeRabbit and Gitar. The examples do not establish a typical industry-wide bill: one publicly traded company with 1,500 engineers caps spending at $150 per developer monthly, while a two-person European .NET agency reports spending about $3,000 per developer. Other companies set caps between $1,000 and $4,000, with typical individual spending around $1,000 in some cases.
Usage and output measures tell different stories. CodeRabbit says developers on its platform create 113% more pull requests per week than a year earlier, while a Microsoft study of tens of thousands of engineers found adopters merged roughly 24% more pull requests over four months. These figures come from different settings and are not directly comparable. Microsoft’s researchers cautioned that a merged pull request is not the same as the value it delivers; some surveyed companies reported flat PR counts, and one saw PRs grow in size without more frequent submissions.
The source article points to practical gains in issue triage, code archaeology, prototyping and backlog cleanup, particularly where codebases use simple, consistent stacks. It also warns that higher token use or more generated code can add cost without improving results. Veracode’s latest test, as reported in the source, found that 44% of AI coding tasks produced a known vulnerability. That result is a warning about the tested tasks, not a measure of the security of all AI-written code.
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 Outcome Measures
The budget question is becoming more urgent because usage-based pricing can make spending rise with agent activity, while discounts and subsidies may expire. The source reports that some companies expect their bills to increase five- to tenfold when discounted subscriptions end; that is a projection from those companies, not a universal price forecast. GitHub also changed Copilot billing on June 1 to charge tokens at each model’s API rate, citing higher compute and inference demands from agentic use.
For engineering leaders, the key distinction is between activity and delivered value. More prompts, tokens or pull requests are consumption and throughput measures; they do not show whether changes were accepted, reduced work elsewhere, or created extra review and repair. The source recommends tracking cost per accepted change, rework, change-failure rate, and stories completed compared with stories arriving. Used together, those measures can show whether faster code production is helping the whole delivery process or shifting work to reviewers and maintainers.
Results are likely to vary by task and codebase. A team working through a uniform backlog may see a different return from one changing complex legacy software. That means company-wide averages can conceal where a tool is useful and where it adds expense or risk. Teams need to assess outcomes against their own baseline rather than treat vendor productivity figures as a guaranteed return.
As an affiliate, we earn on qualifying purchases.
Why AI Coding Bills Differ
The October 23 article draws on a report Orosz published on October 20, based on interviews across 27 workplaces and vendor data. The findings are a snapshot of reported company practices, not a representative census of all engineering organizations. The source says spending peaked in June and July and had roughly halved since for some companies using API plans; Orosz attributed that shift to Opus 5.5 and a move to open models.
Reported output gains also have important qualifications. METR’s early-2025 experiment 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 the confidence intervals included zero, and METR described the data as unreliable. The organization said task selection may have affected the results: 30% to 50% of participants held back tasks they preferred to do without AI. Survey responses can diverge from measured performance too; METR said participants in a 2025 survey overestimated AI’s effect by 40 percentage points.
Examples in the workplace interviews illustrate why a single productivity figure is difficult. One SaaS company reported little change in merged-PR velocity but said AI helped cut customer-issue triage from as long as a day to minutes of agent work plus human checking. Ix Technologies, which works in C# and Blazor, reported clearing a long-running backlog. These are company accounts, not controlled comparisons, and may not transfer to teams with different systems or review practices.
“a question that few companies seem able to answer – at least for now.”
— Gergely Orosz, summarizing the workplace reporting in The Pragmatic Engineer
As an affiliate, we earn on qualifying purchases.
Returns Still Vary by Team
There is no agreed, comparable measure showing the financial return from AI coding tools across companies. The workplace examples are self-reported, vendor statistics describe users of particular platforms, and controlled studies have produced estimates with substantial uncertainty. Pull-request volume, stories closed and self-reported productivity each capture only part of the work.
The source material does not provide full details for every company’s costs, tool mix, task selection or baseline performance. It also does not establish whether the reported security-test result applies to other models, workflows or software tasks. The size of any bill increase after discounts end will depend on each company’s subscription terms, model choices and usage.
As an affiliate, we earn on qualifying purchases.
Track Costs Against Delivery
Companies setting 2027 budgets can begin by establishing a baseline for spend and delivery, then compare teams and task types over time. Useful measures include cost per accepted change, rework, change-failure rates, review effort and the balance between incoming and completed work. Teams should also monitor security findings rather than count generated code as a benefit on its own.
The next useful evidence will come from company-level tracking that links tool use to accepted, maintained software and its downstream costs. The source article does not identify a single forthcoming study or deadline that will settle the return question. For now, organizations must judge results within their own workflows and revisit budgets as discounts, billing arrangements and model choices change.
AI coding security vulnerabilities
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 coding tools?
The 27-workplace reporting cited in the source describes budgets ranging from about $150 to $3,000 per developer per month. Those examples are not an industry-wide average, and some companies set higher caps.
Do more pull requests prove that AI coding tools are paying off?
No. Pull-request counts measure activity, but a higher count does not show whether changes are valuable, need extensive rework or create additional review and maintenance costs. Microsoft’s researchers made that distinction in their study.
Which work appears to benefit from AI coding tools?
The source reports practical gains in triage, code archaeology, prototypes and backlog work, especially for simple, consistent codebases. These examples are company reports and should not be assumed to apply to every team.
What should engineering teams measure?
Track cost per accepted change, rework, change-failure rates and completed stories compared with incoming stories. Review effort and security findings can help reveal costs that raw output counts miss.
Is there a proven productivity gain across all developers?
No. Available results vary by study and setting. METR reported uncertain estimates in its follow-up experiment, while company and vendor reports describe gains in specific environments; none establishes a universal effect.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
