🔍 Read the full analysis: AI Coding ROI: Where It Pays And Where It Costs on ThorstenMeyerAI.com
Get hardware and tech essentials 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 engineering leaders and staff at 27 workplaces found monthly AI coding budgets ranging from $150 to about $3,000 per developer, with no consistent measure of return. Reported output gains vary, and the clearest benefits described were in triage, code archaeology, prototyping and backlog work on simple, uniform software stacks. Higher PR counts do not by themselves establish business value, while security, review capacity and rising usage costs remain concerns.
Engineering teams at 27 workplaces are spending between $150 and about $3,000 per developer per month on AI coding tools, yet many companies still cannot say what the spending has delivered, according to an October 23 report by Thorsten Meyer. The findings arrive as companies set 2027 budgets and weigh reported productivity gains against usage charges, code review demands and security risks.
The report draws on interviews with engineering leaders and senior engineers at 27 workplaces, alongside aggregate figures from AI code-review vendors CodeRabbit and Gitar. The underlying survey is not a controlled comparison of companies, and the report says spending varies widely without a clear link to company size. One publicly traded company with 1,500 engineers sets a $150 monthly cap per developer; a two-person European .NET agency, Ix Technologies, spends about $3,000 per developer.
Output measures also differ. CodeRabbit says developers on its platform created 113% more pull requests per week than a year earlier. That is vendor data from teams using its tool. A Microsoft study of tens of thousands of engineers during an early-2026 rollout found adopters merged roughly 24% more pull requests over four months than they otherwise would have, according to the report. The Microsoft researchers cautioned that a merged pull request does not necessarily represent delivered value.
Other workplaces reported flat pull-request frequency, or more activity alongside larger changes. Controlled-study results cited in the report are mixed: METR’s February follow-up estimated an 18% speedup for returning developers and 4% for new developers, but both estimates had confidence intervals that included zero. METR said the data was unreliable in part because 30% to 50% of participants held back tasks they did not want to do without AI.
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
Budget Decisions Need Outcome Measures
The findings matter because AI coding costs can scale with usage while activity metrics can give an incomplete picture of returns. The source says API-plan spending at some companies peaked in June and July and had roughly halved by the time of reporting, a decline it attributes to the release of Opus 5.5 and a move to open models. Other firms are facing the end of discounted subscriptions; several told the report their bills could rise five- to tenfold if they do not optimize. These are reported experiences, not a forecast for every buyer.
Some companies have a direct way to judge the expense. A quant trading firm described in the report saw monthly AI spending rise from $500,000 in January to $8 million in September, with traders accounting for the largest bills. It evaluates the spending against each desk’s profit and loss. Most engineering organizations do not have an equivalent scoreboard, the report says.
The report’s practical implication is to track outcomes rather than tokens or raw PR volume. Suggested measures include cost per accepted change, rework, change failure rate and whether teams close more stories than arrive. Those measures can help distinguish faster production from work that creates additional review or maintenance burden. They do not, on their own, establish that AI caused a change; teams also need a meaningful comparison over time.
As an affiliate, we earn on qualifying purchases.
Why AI Coding Bills Are Shifting
The immediate budget question follows changes to both subscription pricing and the tools teams use. On June 1, GitHub moved every Copilot plan to usage-based billing, charging for tokens at each model’s API rate and removing the fallback to a cheaper model, according to the source. GitHub explained the change by saying: “Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.”
That shift makes the gap between a flat allowance and intensive agent use more relevant to budgeting. The report says spending caps are common and, by accounts it collected, easy to get around. Meanwhile, teams may see different results depending on their codebase and task. Reported wins include customer-issue triage, finding the history behind existing code, prototypes and backlog cleanup, particularly on simple, consistent stacks. The source contrasts those tasks with complex legacy code, where gains are less clear.
For example, a SaaS company with about 600 engineers saw little movement in merged-PR velocity, but a staff engineer described customer-issue triage falling from sometimes a full day to minutes of agent work plus up to an hour of human checking. Ix Technologies, which uses C# and Blazor, said it was clearing a backlog dating back roughly a decade. These are company accounts, not independent measurements of productivity across the industry.
As an affiliate, we earn on qualifying purchases.
Productivity Gains Remain Hard to Verify
The available figures do not settle how much AI coding improves overall business performance. Vendor platform data, company reports, self-reported productivity and controlled experiments measure different things and cover different groups. The report notes that METR’s survey of 349 technical workers found a median self-reported change of 1.4 to 2 times the value of their work, but METR also said participants in its 2025 survey overestimated AI’s effect by 40 percentage points.
Security is another unresolved cost. The source cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability. The report does not provide further details here about the test’s sample, task types or comparison baseline, so the figure should not be treated as a rate for all AI-generated code. It does point to a need to account for security checks and human review when evaluating time or cost savings.
It also remains unclear how representative the 27 workplaces are, how companies define an accepted change, and whether reported gains persist after review, rework and maintenance. The source’s conclusion that individual output is probably up by double digits is the author’s interpretation, not a single finding shared by all studies.
As an affiliate, we earn on qualifying purchases.
Companies Face 2027 Budget Choices
Companies setting 2027 budgets will need to decide whether to renew discounted plans, impose usage controls or pay for higher levels of agent use. The report does not identify a common approach or a settled ROI benchmark. Teams can compare spending with accepted changes, rework, failure rates and work completed relative to incoming work, while separately tracking review and security costs.
Further evidence will depend on companies publishing comparable results that connect tool use to delivered software and business outcomes. Until then, the current record supports a narrower conclusion: AI coding appears useful for some well-defined tasks and stacks, but higher output counts alone do not show that a company is receiving a return equal to its bill.
software development productivity tools
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 report describes budgets ranging from $150 to about $3,000 per developer per month across 27 workplaces. Those figures are reported examples, not an industry-wide average.
Do the reported productivity gains prove AI coding has a positive ROI?
No. Higher pull-request counts or self-reported speed do not by themselves show business value. The cited studies use different methods, and some experimental estimates had confidence intervals that included zero.
Which tasks were reported to benefit most?
The report describes gains in customer-issue triage, code archaeology, prototyping and backlog cleanup, particularly on simple, uniform software stacks. These examples are company accounts rather than a guarantee for other teams.
What risks should companies include in their calculations?
Companies need to account for usage-based charges, human review, rework and security. The source cites a Veracode test in which 44% of AI coding tasks produced a known vulnerability, but does not give enough test detail to apply that rate to all AI-generated code.
What should teams measure instead of raw usage?
The report suggests tracking cost per accepted change, rework, change failure rate and whether stories are closed faster than they enter the backlog. These measures provide more operational detail than token consumption alone.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
