🔍 Read the full analysis: AI Coding ROI: How To Separate Useful Gains From Rising Costs on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A survey of engineering leaders and senior engineers at 27 workplaces found AI coding budgets ranging from $150 to about $3,000 per developer per month, with no shared way to show what the spending delivers. Reported productivity varies sharply: some teams cite faster triage and backlog work, while other measures are flat, and security and review costs can offset gains.
Engineering teams are spending $150 to about $3,000 per developer per month on AI coding tools, yet many companies cannot say what the outlay has delivered, according to reporting by The Pragmatic Engineer based on interviews at 27 workplaces. The findings arrive as firms set 2027 budgets and some discounted subscriptions approach their end, while data on output remains mixed and the costs of review and security are not captured by code-generation figures alone.
The Pragmatic Engineer’s Gergely Orosz published the survey findings on October 20, combining interviews with engineering leaders and senior engineers and aggregate figures from AI code-review vendors CodeRabbit and Gitar. The article’s freely available material describes monthly budgets ranging from a $150 cap at a publicly traded company with 1,500 engineers to about $3,000 per developer at a two-person European .NET agency. Some companies with higher caps report typical use near $1,000 a month.
Measures of output do not agree. CodeRabbit says weekly pull requests on its platform increased 113% year over year; that is vendor data from organizations using its review tool. 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’s authors. They cautioned that a merged pull request does not, by itself, measure the value delivered. Other companies in the survey reported flat PR counts, while one saw frequency stay level as PRs grew to roughly three times their former size.
Costs are also changing. Orosz reports that API-plan spending at some companies peaked in June and July, then roughly halved as teams moved to open models and newer tools. Several interviewees said discounted subscriptions were due to expire, potentially raising bills five- to tenfold if usage and plans stay unchanged. Separately, GitHub moved Copilot plans to usage-based billing on June 1, charging for tokens at each model’s API rate and removing the cheaper-model fallback. GitHub said agentic use brings higher computing and inference demands.
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 gap between spending and measurable results matters because 2027 budgets are being set now, and usage totals can be mistaken for business value. More generated code or more pull requests may increase work for reviewers without improving delivery. Teams need to account for the full process: accepted changes, rework, failures after release, review time and security remediation.
The source report points to practical areas where teams describe direct gains: customer-issue triage, understanding unfamiliar code, prototypes and backlog cleanup, especially on simple, consistent software stacks. One SaaS company’s staff engineer said triage that could take a day now takes minutes of agent work followed by up to an hour of human checking. That example is an interviewee’s account, not a controlled measurement, but it illustrates why team-level outcomes may matter more than organization-wide PR totals.
There are also potential costs beyond subscriptions. Veracode’s latest test, as cited in the source, found 44% of AI coding tasks produced a known vulnerability. That result does not establish the rate of vulnerabilities in all production code, but it signals why security checks and human review belong in any assessment of net savings.
As an affiliate, we earn on qualifying purchases.
Why AI Coding Results Diverge
Published evidence gives different answers partly because the measures and participants differ. The CodeRabbit figure reflects its platform’s users; Microsoft’s analysis compares adoption outcomes among its own engineers; workplace interviews capture local experience. None alone establishes that AI tools produce the same gains across teams or types of work.
Controlled research has also been uncertain. METR’s early-2025 study found experienced open-source developers took 19% longer with AI on the tasks studied. A 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 withheld tasks they did not want to attempt without AI. In a separate survey of 349 technical workers, the median self-reported change in work value was 1.4 to two times; METR noted that participants in its 2025 research had overestimated AI’s effect by 40 percentage points.
Results also depend on the codebase and the work being done. The source describes a firm using a uniform C# and Blazor stack that reports clearing a decade-old backlog, while teams dealing with complex legacy systems may find changes slower to verify. A quant trading firm’s AI spending reportedly rose from $500,000 a month in January to $8 million in September; the article says traders were its largest spenders and the company evaluates costs against each desk’s profit and loss. That unusual business measure is not available to most engineering 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.
The Net Return Remains Unsettled
The available figures do not establish a single productivity gain that applies across companies. Reported changes range from flat PR counts to large increases, and the metrics vary in what they count. Vendor platform data, company studies, interviews and controlled experiments have different limitations; the findings should not be treated as directly comparable.
It is also unclear how much of the reported time saved becomes shipped value after review, rework, security checks and production failures. The source does not provide a consistent, cross-company accounting of those expenses or of subscription costs per accepted change. The reported 44% vulnerability result comes from a specific Veracode test, not a measure of all AI-written production software. Likewise, the potential five- to tenfold bill increase applies to companies losing discounts if their usage and plans do not change; it is not a forecast for every customer.
As an affiliate, we earn on qualifying purchases.
Track Results Through 2027
As companies finalize 2027 spending plans, the next useful evidence will be team-level comparisons that include both tool costs and downstream engineering work. The source report recommends tracking cost per accepted change, rework, change failure rate, and stories closed versus stories arriving, rather than treating token use or generated code as a return measure.
Companies facing expiring discounts or usage-based billing will also need to decide whether to cap access, change models, or keep current plans. Those choices can be assessed against the same delivery and quality measures. The source material does not identify a common deadline for these budget decisions beyond the current 2027 planning cycle, and it remains to be seen whether companies will publish comparable results.
AI developer coding tools subscription
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 survey covered monthly budgets from $150 to about $3,000 per developer. Some companies set higher caps, but reported typical use at those firms was around $1,000 per developer per month.
Does the data show that AI coding increases productivity?
It shows mixed results, not a universal gain. CodeRabbit reported a 113% year-over-year increase in weekly PRs on its platform, and a Microsoft study found adopters merged about 24% more PRs over four months. Other companies reported flat counts, and these measures do not directly establish the value or quality of work delivered.
Where do teams report the clearest benefits?
The source describes reported gains in issue triage, code archaeology, prototyping and backlog cleanup, particularly on uniform technology stacks. These are examples from company interviews, not proof that every team will see similar results.
What should engineering leaders measure?
The source recommends measuring cost per accepted change, rework, change failure rate and stories closed compared with incoming stories. Teams should include review and security work rather than counting generated code or tool usage alone.
What remains unknown about AI coding ROI?
There is no consistent cross-company measure showing how much reported time savings becomes delivered value after review, rework, security checks and failures. The available studies also use different populations and metrics, limiting direct comparison.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
