🔍 Read the full analysis: How To Know Whether AI Coding Is Paying Off 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 report based on interviews at 27 workplaces finds AI coding spending ranges from about $150 to $3,000 per developer per month, with no clear link between company size and budget. Output measures vary widely, so teams trying to set 2027 budgets need to compare delivery outcomes, rework and failures with costs rather than treating tool activity or pull-request volume as proof of return.
Engineering teams are spending from about $150 to $3,000 per developer per month on AI coding tools, but a new survey of 27 workplaces found no consistent way to show what that spending delivers. The findings, reported by The Pragmatic Engineer on October 20, come as companies set 2027 budgets and face changes in AI subscription pricing.
The Pragmatic Engineer’s Gergely Orosz gathered responses from engineering leaders and senior engineers at 27 workplaces, alongside aggregate figures from AI code-review vendors CodeRabbit and Gitar. The source article says spending caps range from $150 to $4,000, while typical spending at some companies is around $1,000 per developer monthly. Company size did not predict where a workplace fell on that range.
Reported output measures conflict. CodeRabbit said pull requests created on its platform rose 113% year over year; Microsoft research found adopters merged about 24% more pull requests over four months during an early-2026 rollout. These measures do not establish equivalent gains in delivered value. Microsoft’s authors cautioned that a merged pull request is not the same as the value it delivers. Some surveyed companies reported flat pull-request frequency, including one where the requests became larger.
Evidence from controlled tests also differs. 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 the confidence intervals included zero; METR called the data unreliable, citing task selection issues. The source advises teams to track cost per accepted change, rework, change failure rate and whether completed stories outpace incoming work.
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
Budgeting Beyond Pull-Request Counts
For engineering leaders, the central issue is not whether a tool can generate code, but whether its use improves the work the organization needs done at a cost it can justify. A higher pull-request count can reflect more activity without showing that changes are useful, safe, or reaching customers. Measuring accepted changes alongside rework and production failures gives teams a more grounded basis for deciding whether to expand, limit, or change their tool use.
The timing matters because some companies say discounted subscriptions are nearing expiration, while API-based plans can make costs vary with usage. The source reports one quant trading firm’s monthly AI spending increased from $500,000 in January to $8 million in September; the largest spenders there were traders, and the firm used desk-level profit and loss as its measure. Most engineering organizations do not have an equally direct financial scoreboard. Without one, rising usage can be mistaken for a return on investment.
The reported gains also appear task- and codebase-dependent. The article identifies triage, code archaeology, prototyping and backlog cleanup on simple, uniform systems as promising uses. It describes a SaaS company where overall merged-pull-request velocity barely changed even as staff reported faster customer-issue triage and research into why code had been written a certain way. That distinction matters: local time savings may be real without appearing in a team-wide throughput measure.
As an affiliate, we earn on qualifying purchases.
Why AI Coding Bills Are Changing
The October 20 report arrives amid shifting pricing and model choices. According to the source, spending on API plans peaked in June and July and had roughly halved by the time of publication for some companies. Orosz attributed the decline to the arrival of Opus 5.5 and a move to open models. Those are his explanations for the reported pattern, not a controlled comparison establishing why every company’s bill changed.
The source also says GitHub moved every Copilot plan to usage-based billing on June 1, 2026, charging tokens at each model’s API rate and ending a fallback to a cheaper model. GitHub explained the change by saying, “Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.” This pricing shift gives companies another reason to compare spending with work delivered, rather than relying on a fixed subscription price.
Results vary sharply by workplace. A two-person European .NET agency, Ix Technologies, reported spending around $3,000 per developer monthly and using AI on a uniform C# and Blazor stack to clear older backlog work. The source contrasts this with a public company of 1,500 engineers that capped spending at $150 per developer. Neither case establishes a universal budget or a result that will transfer to different teams.
“Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.”
— GitHub, explaining its Copilot billing change
As an affiliate, we earn on qualifying purchases.
What the Productivity Numbers Miss
The available figures do not provide a single, comparable estimate of AI coding’s effect across engineering teams. Vendor platform data, workplace reports and controlled studies measure different populations and outcomes. CodeRabbit’s increase describes activity on its platform; Microsoft’s figure concerns merged pull requests among adopters; METR’s estimates carry uncertainty and limits acknowledged by the researchers. The source article’s interpretation that individual output is probably up by double digits is an assessment, not a consensus finding established by these figures.
It is also unclear how often reported time savings translate into faster releases, better software, lower staffing costs or improved customer outcomes. The source notes that one company’s pull-request volume remained flat while requests grew larger, and reports security concerns: a cited Veracode test found 44% of AI coding tasks produced a known vulnerability. The supplied material does not give the test’s methodology or a comparison baseline, so that number should not be treated as a general rate for all AI-generated code.
Several companies reportedly face subscriptions losing discounts, but the source does not identify all affected firms or specify when each subsidy ends. Nor does it establish which model, workflow or spending level works best across different codebases. Claims of efficiency remain dependent on local conditions, review practices and how teams define successful delivery.
As an affiliate, we earn on qualifying purchases.
How Teams Can Set 2027 Budgets
Engineering leaders setting 2027 budgets can begin by recording tool costs and a baseline for the work their teams already deliver. The source recommends tracking cost per accepted change, rework, change-failure rates, and stories closed compared with stories arriving. Teams can also separate work types—such as triage, prototyping and maintenance—so results from one task are not assumed to apply to another.
Those measures should be read alongside quality and review capacity. A rise in generated code or pull requests is not a useful result if it brings extra revisions, unresolved security issues, or more work for reviewers. The source’s survey and study summaries do not prescribe a universal measurement period or threshold; companies will need to state their own baseline and compare like with like before changing budgets.
The next decision point is whether teams can show that measured gains persist as pricing changes and tools move from discounted subscriptions to usage-based billing. The October report offers examples and warnings, not a settled verdict. Its clearest takeaway for budget owners is to require evidence tied to accepted, reliable work before treating usage growth as proof that AI coding is paying off.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The October 20 report describes budgets ranging from about $150 to $3,000 per developer per month, with some reported caps reaching $4,000. These are examples from 27 workplaces, not a representative industry-wide average.
Do more pull requests prove 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 changes, rework and failures.
Where does AI coding appear most useful in the report?
The source identifies triage, code archaeology, prototypes and backlog cleanup on simple, uniform codebases as promising areas. It also describes mixed results across companies, so these examples do not guarantee similar gains elsewhere.
What should an engineering team measure?
The source recommends cost per accepted change, rework, change failure rate, and whether the team closes more stories than arrive. Teams should compare those measures with a baseline and account for the cost of review and quality checks.
Are the reported productivity gains settled?
No. Vendor data, company reports and controlled studies use different measures and produce mixed results. METR said its February estimates had confidence intervals that included zero and described the data as unreliable.
Source: ThorstenMeyerAI.com
Evergreen bestsellers Picks
bestsellers
As an affiliate, we earn on qualifying purchases.
