Can AI Coding Pay Off? Start By Tracking These Costs

🔍 Read the full analysis: Can AI Coding Pay Off? Start By Tracking These Costs on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Engineering teams in a survey of 27 workplaces reported AI coding budgets ranging from $150 to about $3,000 per developer per month, but the findings do not establish a common return on that spending. The report, published Oct. 20 by The Pragmatic Engineer, describes mixed productivity results and argues companies should measure accepted work and rework as they set 2027 budgets.

The spending range came from interviews with engineering leaders and senior engineers at 27 workplaces, supplemented by aggregate data from code-review vendors CodeRabbit and Gitar. The published account says company size did not reliably predict how much firms spent. Some companies set monthly caps, while others reported typical spending around $1,000 per developer. A two-person European .NET agency, Ix Technologies, was cited as spending about $3,000 per developer monthly.

Measures of output differed. CodeRabbit reported that developers on its platform created 113% more pull requests per week than a year earlier. That figure describes activity among teams using the vendor’s AI review product; it is not a controlled industry-wide measure. 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 report. Its authors cautioned that a merged pull request does not, by itself, establish the value delivered.

Company accounts ranged from flat pull-request frequency to increases of several times, with some teams producing larger changes rather than more frequent ones. The report cites work by METR that also gives a mixed picture: an early-2025 study found experienced open-source developers took longer with AI, while a February follow-up estimated possible speed gains but said the confidence intervals included zero and the data was unreliable. The source also points to a Veracode test in which 44% of AI coding tasks produced a known vulnerability, underscoring the need to track security and review alongside speed.

Budgets Need Delivery Measures

The findings matter because AI coding expenses can vary sharply and some discounted subscriptions are due to expire, according to companies interviewed for the report. One firm said its subsidy ending could expose it to a bill five to ten times higher unless it changes its usage or plan. Separately, the report says API-based spending at some companies peaked in June and July and had roughly halved since, which Gergely Orosz attributed to a move to open models and the release of Opus 5.5. These are reported company experiences, not a forecast for all engineering teams.

Raw tool usage and pull-request counts do not show whether a team ships useful work faster or creates extra review burden. The report recommends measuring cost per accepted change, rework, change failure rate and whether teams close more user stories than arrive. Such measures can help budget owners compare spending with delivered results while accounting for quality and maintenance costs.

The source identifies triage, code archaeology, prototyping and backlog cleanup on simple, uniform technology stacks as areas where interviewees described concrete time savings. That is not proof that all organizations will see the same returns. It does suggest companies may get a clearer picture by looking at specific workflows instead of treating one productivity number as a verdict on AI coding.

Why Coding Bills Are Shifting

The report lands as companies prepare 2027 engineering budgets and vendors change how AI tools are billed. According to the source, GitHub moved all Copilot plans to usage-based billing on June 1, charging for tokens at each model’s API rate and removing the fallback to a cheaper model. GitHub explained the change by saying, “Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.”

The report also describes wide gaps in how organizations judge value. A quant trading firm with roughly 2,000 developers reportedly saw AI spending rise from $500,000 a month in January to $8 million in September. The account says the firm assesses AI as part of each trading desk’s cost base against its profit and loss. Most companies cited did not have a comparable scoreboard, and Orosz summarized the return-on-investment question as one few companies could answer at the time.

Reported gains are not interchangeable. Vendor platform data, company observations and controlled studies have different methods and limits. The report notes that METR’s survey of technical workers found high self-reported gains, but METR said participants in its 2025 work had overestimated AI’s effect by 40 percentage points. The evidence therefore supports measurement at the team and task level, not a single universal productivity estimate.

““A merged PR is not the same as the value it delivers.””

— Microsoft study authors, as quoted in The Pragmatic Engineer report

Returns Vary by Team and Task

The report does not establish a single average return on AI coding investment. Its workplace accounts are not presented as a representative sample of all engineering organizations, and vendor statistics reflect users of those vendors’ products. The source says the full Pragmatic Engineer piece is partly paywalled; the account here draws on its freely readable material.

It remains unclear how much reported output growth becomes higher-quality delivered software, and how much is offset by review time, defects, security risks or maintenance. The report cites different results across organizations and warns that pull-request volume alone cannot settle the question. The cited Veracode result also does not tell readers how often vulnerabilities reach production or how the result compares across every type of task.

Some figures are based on company reports or individual assessments, while research findings have their own methodological limitations. METR said its follow-up data was unreliable, partly because 30% to 50% of developers held back tasks they did not want to do without AI. The available evidence therefore does not support a universal promise of faster delivery or a precise return for every company.

Track Results Before Renewals

Companies setting 2027 budgets can compare spending with accepted work before renewals or subsidies change their bills. The report’s proposed measures include cost per accepted change, rework, change failure rate and stories closed relative to stories entering the backlog. Teams can also examine specific workflows—such as customer-issue triage or code archaeology—where the report says interviewees described time savings.

The next useful evidence will come from organizations recording those measures over time and separating model or subscription costs from the engineering work needed to review and maintain outputs. The source does not identify a forthcoming industry-wide study or a date when a settled ROI benchmark will be available. Until comparable outcome data emerges, companies will need to judge the payoff within their own teams and workflows.

Key Questions

How much are companies spending on AI coding tools?

The report says budgets among 27 workplaces ranged from $150 to about $3,000 per developer per month. The figures describe those workplaces, not a universal industry average.

Does the report show that AI coding increases productivity?

It reports productivity gains in some settings, including a Microsoft study that found adopters merged about 24% more pull requests over four months. Other organizations reported flat pull-request counts, and METR research cited in the report has substantial limitations. More pull requests do not automatically mean more delivered value.

What should engineering teams measure?

The report recommends tracking cost per accepted change, rework, change failure rate and stories closed compared with stories coming in. Those measures can help distinguish useful output from higher tool usage or larger pull requests.

Where did interviewees report clearer benefits?

The source describes gains in customer-issue triage, understanding why existing code was written a certain way, prototyping and backlog work on simple, uniform stacks. These are reported examples, not a guarantee that every team will see similar results.

What remains unknown about AI coding ROI?

The available evidence does not establish a single return that applies across companies. It remains unclear how reported speed or output changes translate into software quality and business value after review, rework, security and maintenance costs are counted.

Source: ThorstenMeyerAI.com