🔍 Read the full analysis: The Economics Of AI Coding: Spend, Savings, And Measurement 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 per month, while reported productivity results varied sharply. Research and company examples suggest gains depend on the work and team; many organizations still lack measures that show whether extra output becomes delivered value.
A survey of 27 workplaces found that companies spend between about $150 and $3,000 per developer per month on AI coding tools, but many cannot yet show what the spending has delivered. The report, published October 20 by The Pragmatic Engineer, also found sharply mixed productivity results, as companies prepare 2027 budgets and some face the end of discounted subscriptions.
The survey by The Pragmatic Engineer’s Gergely Orosz drew on interviews with engineering leaders and senior engineers, plus aggregate data from code-review vendors CodeRabbit and Gitar. The article’s freely readable portion describes spending caps, usage patterns and output measures; its author says that return on investment remains a question few companies can answer. Spending varies widely and company size did not predict where a workplace fell in the reported range.
Reported output measures do not line up. CodeRabbit said developers on its platform created 113% more pull requests per week than a year earlier, a vendor statistic from teams using its AI review product. 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. Microsoft’s authors cautioned that a merged pull request does not by itself show the value delivered.
Accounts from the surveyed companies ranged from flat pull-request frequency to increases of several times. In one workplace, pull requests became three times larger while their frequency stayed flat. A separate example from a SaaS company with about 600 engineers described faster customer-issue triage and code-history research even though merged-pull-request velocity barely changed. These are reported company experiences, not a common controlled measure of productivity.
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
Higher Bills, Uneven Engineering Gains
The budget question matters because tool costs can change quickly when usage is billed by model consumption rather than a fixed subscription. The source report says some companies’ discounted subscriptions are due to end, leaving them to reduce usage or face bills the author says could rise five- to tenfold. It also reports that GitHub moved Copilot plans to usage-based billing on June 1, charging tokens at each model’s API rate.
Higher output counts do not necessarily mean more useful software reaches customers. Larger pull requests can increase the burden on reviewers, while security and rework costs can offset time saved during drafting. The source cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability. That figure describes the test results and should not be read as a rate for all AI-generated code in production.
For engineering leaders, the practical choice is not simply whether to buy AI tools. It is whether the gains in particular tasks outweigh subscriptions, inference costs, review time and any added defects. Measures such as cost per accepted change, rework, change-failure rate and stories closed compared with stories arriving can connect spending more closely to outcomes than token use or raw pull-request totals.
As an affiliate, we earn on qualifying purchases.
How Spending and Output Diverged
The Pragmatic Engineer report describes a broad spread in budgets: a publicly traded company with 1,500 engineers capped monthly spending at $150 per developer, while a two-person European .NET agency reported about $3,000 per developer. Other companies had caps between $1,000 and $4,000, with typical reported spending around $1,000. One quant trading firm said its monthly AI spending climbed from $500,000 in January to $8 million in September; the source says its largest spenders were traders, and that desks assessed the cost against profit and loss.
Evidence from controlled studies is less settled than some platform statistics. 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 both confidence intervals included zero. METR also described the follow-up data as unreliable because some participants held back tasks they did not want to attempt without AI. In a survey of 349 technical workers, participants reported a median change of 1.4 to 2 times the value of their work; METR cautioned that its 2025 participants had overestimated AI’s effect by 40 percentage points.
The source’s examples suggest gains are easier to identify in triage, code archaeology, prototyping and backlog cleanup, especially on simple, consistent software stacks. A .NET agency said it was clearing a backlog built up over about ten years. Another company reported one team closing three times as many user stories each month. Those cases describe particular organizations and should not be treated as a guaranteed result for other engineering teams.
As an affiliate, we earn on qualifying purchases.
What the Productivity Figures Miss
The available figures do not establish a single, comparable productivity gain across engineering teams. Vendor platform data, company reports, controlled experiments and self-reported estimates use different populations and methods. The source does not provide enough detail to reconcile those measures or determine how much additional coding output became customer value.
It also remains unclear how the end of discounts and changing billing models will affect total spending across companies. Reported caps may not reflect actual consumption, and the source says caps can be easy to get around. The Veracode result concerns its test tasks; it does not establish the vulnerability rate for all AI-assisted development. Long-term effects on maintenance, defects and review workloads are not settled by the figures presented.
developer productivity analytics tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Metrics Behind 2027 Budgets
Companies setting 2027 budgets will need to compare costs with outcomes at the team and task level, rather than treating tool adoption or token consumption as proof of return. The report recommends tracking cost per accepted change, rework, change failures and whether teams close more work than enters their backlog. Those measures could help distinguish useful acceleration from activity that adds review or correction work.
Further evidence will depend on whether companies publish comparable results as billing arrangements change and discounted plans expire. For now, the source report supports a narrower conclusion: AI coding can save time on some tasks, but the size and value of the gains vary, and many organizations have not yet built the measurement systems needed to justify their spending.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The survey described monthly budgets ranging from about $150 to $3,000 per developer. Some companies reported caps between $1,000 and $4,000, with typical spending around $1,000 in those cases.
Does AI coding reliably increase developer productivity?
The figures are mixed. Microsoft reported about 24% more merged pull requests among adopters in its study, while company reports ranged from flat pull-request frequency to much higher counts. METR’s estimates were uncertain, with confidence intervals that included no effect in its follow-up.
What work appears to benefit most in the report?
Reported examples point to triage, code archaeology, prototyping and backlog cleanup, particularly on simple and uniform software stacks. These are examples from specific teams, not a guarantee of similar gains elsewhere.
How should a company measure the return?
The source recommends comparing spending with outcomes such as cost per accepted change, rework, change-failure rates and stories closed relative to stories coming in. Pull-request or token counts alone do not establish delivered value.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
