🔍 Read the full analysis: How Much Does AI Coding Pay Off? A Practical Measurement Guide 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 report based on interviews at 27 workplaces finds AI coding budgets range from about $150 to $3,000 per developer monthly, while output measures vary sharply. The evidence points to useful gains in tasks such as triage and backlog work, but does not establish a universal return on investment; teams need to track accepted changes, rework, failures and completed work.
Engineering teams are spending about $150 to $3,000 per developer per month on AI coding tools, but a review published October 23 says companies still lack a consistent way to show what that spending delivers. Drawing on interviews at 27 workplaces and data from vendors and other studies, the report finds uneven productivity results and recommends measuring accepted work, rework and delivery outcomes as companies set budgets for 2027.
The budget range comes from reporting by The Pragmatic Engineer’s Gergely Orosz, who spoke with engineering leaders and senior engineers at 27 workplaces and included aggregate data from AI code-review companies CodeRabbit and Gitar. Spending does not appear to track company size neatly: examples range from a publicly traded company capping its 1,500 engineers at $150 per person monthly to a two-person European agency spending about $3,000 per developer.
Measures of output tell different stories. CodeRabbit says pull requests created by developers on its platform rose 113% year over year; Microsoft’s study of tens of thousands of engineers found adopters merged about 24% more pull requests over four months than they otherwise would have. Those figures come from different populations and methods. At several companies interviewed, pull-request frequency stayed flat, while one reported that pull requests became three times larger.
Controlled-study evidence is less settled. 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 estimates had confidence intervals that included zero; METR said the data was unreliable, in part because some participants avoided tasks they preferred to do without AI. The report’s central measurement advice is to track cost per accepted change, rework, change failure rate and whether completed stories exceed incoming work, rather than treating usage or raw PR volume as proof of value.
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
Why Budget Owners Need Outcome Data
The question is becoming more immediate as companies prepare 2027 budgets and some discounted subscriptions approach expiration. The source report says several companies expect costs to rise five- to tenfold if they do not optimize their plans. GitHub also 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, according to the source.
More code or more pull requests do not automatically mean more useful software. Microsoft’s study authors cautioned that “a merged PR is not the same as the value it delivers.” If review and security work cannot keep pace, faster generation may shift effort downstream or introduce more rework. Veracode’s latest test, cited by the source, found 44% of AI coding tasks produced a known vulnerability; that result is a test finding, not a measure of the vulnerability rate in all deployed AI-assisted code.
Outcome measures can help leaders compare the cost of a tool with changes that users or the business actually receive. They also make it easier to spot where AI assists existing work and where it merely increases consumption. No single productivity figure in the report settles that question across teams.
As an affiliate, we earn on qualifying purchases.
What the Productivity Studies Measure
The source report combines evidence that should not be read as one comparable experiment. CodeRabbit’s figure reflects activity on its own platform, while Microsoft studied adoption during an early-2026 rollout of Claude Code and GitHub Copilot CLI. Company interviews add operational examples, but they are not controlled trials. The reported numbers therefore describe different groups, tools and definitions of output.
The examples suggest task and codebase matter. A staff engineer at a software company with about 600 engineers said customer-issue triage that could take a day became minutes of agent work plus up to an hour of human review. The same engineer described code archaeology taking minutes rather than a day or two, and a prototype taking hours rather than one to two weeks. Ix Technologies, a small firm using a uniform C# and Blazor stack, said it was clearing a long-standing backlog. These are reported company experiences, not general productivity guarantees.
The source also describes a quant trading firm whose AI spending grew from $500,000 monthly in January to $8 million in September, with traders accounting for the largest use. That company evaluates spending against desk-level profit and loss, a direct business measure that many engineering organizations do not have. The contrast helps explain why a single industry-wide ROI figure is difficult to establish.
““A question that few companies seem able to answer – at least for now.””
— Gergely Orosz, summarizing the companies he interviewed
best AI-assisted code review software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Where the Return Remains Unproven
The evidence does not establish a universal productivity gain or a reliable average return on investment. Vendor activity data, company accounts and controlled studies measure different things, and the report does not provide a common baseline across the 27 workplaces. It also does not establish whether higher pull-request counts lead to better software, lower total costs or improved customer outcomes.
Several details remain open: how much time saved on drafting is offset by review and maintenance; whether benefits persist as teams change tools or workflows; and how often AI-generated vulnerabilities reach production. The source reports that spending on API plans roughly halved after peaking in June and July, attributing the shift to Opus 5.5 and movement to open models, but it does not provide enough detail to compare costs or performance across those choices. The 44% Veracode finding should not be generalized beyond the test described.
As an affiliate, we earn on qualifying purchases.
What Teams Should Track Next
For budget decisions, the report’s practical next step is to establish a team-level baseline before comparing AI-assisted work with earlier performance. Track spend per accepted change, rework, change failure rates and completed stories against incoming work, while noting review time and the type of task. These measures can show whether faster coding translates into more delivered work or simply larger changes and added checking.
Companies will also need to reassess costs as subscription discounts end and usage-based billing changes the relationship between activity and price. The source report offers no single deadline or promised industry benchmark; it points instead to budget reviews underway for 2027. How much AI coding pays off will remain a team-specific question until organizations connect tool costs to delivery quality and business results over time.
software for tracking developer output
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 workplaces in the source report range from a $150 monthly cap per developer to about $3,000 per developer per month. These are examples from 27 workplaces, not a representative industry average.
Does the evidence show that AI coding increases productivity?
Some studies and companies report higher output, but results differ. CodeRabbit reported 113% more weekly pull requests on its platform year over year, while Microsoft found about 24% more merged pull requests among adopters over four months. Those figures use different methods, and neither alone proves an equivalent increase in business value.
Which coding tasks appear to benefit most?
Company accounts in the report identify triage, code archaeology, prototyping and backlog cleanup on simple, uniform codebases as useful applications. These examples are reported experiences, not guarantees for other teams.
What should engineering teams measure?
Track cost per accepted change, rework, change failure rate and whether teams close more stories than arrive. Include review and maintenance effort so increased code volume is not mistaken for improved delivery.
Is there a confirmed industry-wide ROI figure?
No. The report says few companies can answer the ROI question, and the available studies vary in population, task and measurement. A reliable, comparable industry-wide return has not been established by the evidence presented.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
