🔍 Read the full analysis: The Money Behind AI Coding: Costs, Returns, And Metrics on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A survey of 27 workplaces found AI coding budgets ranging from $150 to about $3,000 per developer per month, while reported productivity results vary widely. Companies report gains in tasks such as triage and backlog cleanup, but many lack measures linking tool use to business value.
Engineering teams spend from $150 to about $3,000 per developer per month on AI coding tools, but many companies cannot say what that spending has delivered, according to a review published by Thorsten Meyer on Oct. 23. The findings draw on Gergely Orosz’s survey of 27 workplaces and vendor and company data, as firms set budgets for 2027 and weigh higher usage-based bills against uneven productivity evidence.
Orosz’s Oct. 20 report for The Pragmatic Engineer gathered accounts from engineering leaders and senior engineers at 27 workplaces, alongside aggregate information from AI code-review vendors CodeRabbit and Gitar. The source article says its figures come from the freely readable portion of Orosz’s report. Spending caps and actual bills differ sharply: a publicly traded company with 1,500 engineers caps monthly spend at $150 per developer, while European software agency Ix Technologies, a two-person .NET firm, spends about $3,000 per developer. Other companies reported caps from $1,000 to $4,000, with typical monthly spending around $1,000.
The output measures do not tell one consistent story. CodeRabbit reported that developers on its platform created 113% more pull requests per week than a year earlier. In a study of tens of thousands of engineers during an early-2026 rollout, Microsoft found adopters merged about 24% more pull requests over four months than they otherwise would have, according to the source article. Those figures measure activity, not necessarily business value: Microsoft’s researchers cautioned that a merged pull request is not the same as value delivered. At several surveyed companies, pull-request counts remained flat; one reported that PRs became three times larger.
The evidence from controlled tests is less settled. The source article reports that 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 also described the follow-up data as unreliable, in part because 30% to 50% of developers held back tasks they did not want to do without AI. These results do not establish a single productivity effect across workplaces.
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
Where AI Spending Can Pay Off
The reported gains appear most concrete in bounded tasks such as customer-issue triage, code archaeology, prototyping and backlog cleanup, particularly on simple, consistent software stacks. A staff engineer at a SaaS company with about 600 engineers told Orosz that triage which could take a day sometimes now takes minutes of agent work followed by up to an hour of human checking. The engineer also said tracing why code was written a certain way can take minutes rather than a day or two, and that prototypes can be built in hours instead of one to two weeks.
Ix Technologies’ co-founder Rogier van der Hee said the agency’s uniform C# and Blazor stack helps it clear old backlog work, and described the pair as feeling able to do the work of five to seven people before AI. These are company accounts, not controlled comparisons. They suggest where teams may find useful applications, but do not prove the same returns will apply elsewhere.
Costs can also rise quickly when usage grows or billing changes. One quant trading firm’s reported AI spending climbed from $500,000 a month in January to $8 million in September; the article says its biggest spenders were traders rather than engineers. That firm evaluates AI as part of a trading desk’s profit-and-loss results, a clearer measure than most engineering departments have. For other organizations, counting tokens or pull requests alone cannot show whether the extra work is accepted, reliable or valuable.
The source article also cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability. That result adds a quality and security dimension to the cost question: more generated code may require more review and remediation. The source does not provide further details of the test setup in the supplied material, so the percentage should not be treated as a universal rate for AI-generated software.
As an affiliate, we earn on qualifying purchases.
Budgets Meet New Billing Models
The question has become more pressing as companies prepare 2027 budgets and some discounts approach their end. According to the source article, several companies said discounted AI subscriptions would soon expire, forcing them either to control use or face bills they said could rise five- to tenfold. Orosz attributed a roughly halving of API-plan spending since its June and July peak to a move to open models and the availability of Opus 5.5. The article does not give a common spending baseline across the companies, so this is an account of reported changes, not a market-wide estimate.
Billing has also shifted toward usage. GitHub changed all Copilot plans to usage-based billing on June 1, charging tokens at each model’s API rate and removing the fallback to a cheaper model, according to the source. GitHub said agent use brings higher computing and inference demands. That change makes the amount and type of model use more directly relevant to a team’s bill, while making comparisons with earlier subscription pricing harder.
Experiments and self-reports offer different signals. In a METR survey of 349 technical workers, the median respondent reported a 1.4-to-2-times change in the value of their work. METR cautioned that participants in its 2025 work overestimated AI’s effect by 40 percentage points. The source article’s author interprets the evidence as suggesting individual output is probably up by double digits, while stressing that delivered value depends on the team. That is an interpretation, not a settled finding across the studies.
“Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.”
— GitHub
developer productivity tracking software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Returns Beyond Activity Counts
The available evidence does not establish a typical return on investment for AI coding tools. The 27 workplaces reported different spending and output patterns, vendor figures describe users of a particular platform, and the supplied material does not provide a shared method for measuring quality-adjusted output across them. It is also unclear how much reported time saved turns into faster releases, lower costs or better customer outcomes.
Security and rework costs remain difficult to compare from the information provided. The Veracode result is a warning about tested tasks, but the source material gives no sample size or breakdown by model, language or task. Nor does it quantify how often AI-generated code leads to defects, review delays or production failures across the surveyed companies. Those unknowns limit conclusions about net returns.
As an affiliate, we earn on qualifying purchases.
What Companies Will Measure
As 2027 budgets are finalized and discounted subscriptions expire, companies will need to compare spend with outcomes rather than usage alone. The source article recommends tracking cost per accepted change, rework, change failure rate, and stories closed versus stories coming in. Those measures can help distinguish faster code production from improvements that reach users and hold up in production.
The next useful evidence would include longer-running comparisons across teams, with consistent measures of code quality, security, review effort and delivery outcomes. Until that information is available, companies’ reported gains and costs remain useful indicators of where tools may fit, not a reliable forecast of what every team will save or produce.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The source review reports budgets ranging from $150 to about $3,000 per developer per month among examples from 27 workplaces. It also describes caps of $1,000 to $4,000 at some companies, with typical spend around $1,000 in those accounts; these are reported figures, not a market-wide average.
Do the figures prove AI coding makes developers more productive?
No. Some company and vendor measures show more pull requests, while several workplaces reported flat counts. Controlled METR estimates were uncertain, and pull-request volume does not by itself show quality or business value.
Which coding tasks were reported as useful?
Accounts in the source describe gains in triage, code archaeology, prototyping and backlog cleanup, especially on simple, uniform software stacks. These are reported experiences and may not apply to complex legacy systems or every team.
What should engineering teams track besides AI usage?
The source article points to cost per accepted change, rework, change failure rate and whether teams close more user stories than arrive. These measures connect spending to completed work and quality more directly than token use or raw PR counts.
What remains unknown about AI coding returns?
There is no common, verified estimate of net returns across companies. The available material does not establish how gains compare with subscription costs, review and security work, rework, or the value of software delivered.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
