The Business Case For AI Coding: Costs, Returns, And Measurement

🔍 Read the full analysis: The Business Case For AI Coding: Costs, Returns, And Measurement on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

Engineering teams surveyed across 27 workplaces report AI coding budgets ranging from about $150 to $3,000 per developer per month, while many companies cannot yet say what those costs have delivered, according to an analysis published Oct. 23 by ThorstenMeyerAI.com. The review brings together company accounts, vendor data and research findings as organizations set 2027 budgets and face changes to subscription pricing.

The analysis draws on an Oct. 20 report by Gergely Orosz of The Pragmatic Engineer, who spoke with engineering leaders and senior engineers at 27 workplaces and included aggregate figures from code-review vendors CodeRabbit and Gitar. Its spending examples vary sharply: one publicly traded company with 1,500 engineers caps usage at $150 per developer monthly, while a two-person European agency, Ix Technologies, spends about $3,000 per developer. Several companies described typical monthly usage around $1,000, despite higher caps.

Output measures also differ. CodeRabbit reports that developers on its platform create 113% more pull requests per week than a year earlier; that is vendor data from organizations using its AI review service. 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. The study’s authors cautioned that a merged pull request does not itself establish the value delivered.

Other company accounts range from flat pull-request frequency to reports of several-fold increases, sometimes alongside larger changes rather than more changes. The analysis also cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability. It recommends tracking outcomes such as cost per accepted change, rework, change failure rates and whether teams close more work than enters their queues, rather than relying on token use or pull-request counts alone.

Budgets Need Outcome Measures

The spending gap matters because companies may soon pay more for tools whose returns are not captured by a single productivity metric. The analysis says some organizations have been using discounted subscriptions that are due to expire; those companies could face bills rising five- to tenfold if they do not change plans or usage. That projection comes from accounts reported in the source, not a universal estimate for all customers.

Pull-request volume can obscure whether software is better, safer or reaching users faster. A team may produce more code while also creating more review work or defects. The cited vulnerability result makes the quality question material: teams need to account for security checks and rework alongside speed. The analysis points to stronger use cases in issue triage, code archaeology, prototyping and backlog cleanup, especially where systems are simple and consistent.

For engineering leaders, the practical question is whether AI changes the cost and pace of useful work, not whether employees use more tokens or generate more code. Comparing accepted changes, work completed, failures and review burden can help distinguish productivity improvements from added activity. The source does not establish one measurement method that works for every company.

Pricing and Productivity Evidence

The timing reflects both budget planning and changes in how AI tools are billed. According to the source, GitHub moved every Copilot plan 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 said the change reflected increased compute and inference demands as agentic use became more common. The supplied material does not specify the year for that June 1 date.

Reported productivity findings are not directly interchangeable. METR’s early-2025 study found experienced open-source developers took 19% longer with AI on the tasks studied. A February follow-up estimated an 18% speedup for returning developers and a 4% speedup for new developers, but both confidence intervals included zero. METR called the data unreliable, in part because 30% to 50% of participants withheld tasks they did not want to attempt without AI.

In a separate METR survey of 349 technical workers, the median respondent reported a 1.4-to-2-times change in the value of their work. METR noted that participants in its 2025 research had overestimated AI’s effect by 40 percentage points. Those caveats underline why survey impressions, controlled task studies and vendor platform metrics should not be treated as equivalent evidence.

“A question that few companies seem able to answer – at least for now.”

— Gergely Orosz, summarizing the companies he surveyed

Returns Remain Hard to Verify

The evidence does not establish a single typical productivity gain across engineering teams. Company reports vary widely, vendor metrics cover their own platforms, and research findings carry limits involving task selection, sample design and uncertain estimates. The analysis’s view that individual output is probably up by double digits is its interpretation, not a settled finding shared by all cited studies.

It is also unclear how frequently higher output translates into faster releases, lower costs or better software. The source does not provide comparable, audited financial returns across the 27 workplaces, nor does it establish how much the reported vulnerability rate changes in production. Company-specific pricing, usage, review capacity and codebase complexity all affect the result.

Companies Set 2027 Budgets

Companies are setting 2027 engineering budgets while some discounted AI subscriptions approach renewal and usage-based billing changes affect costs. The next useful evidence will come from teams tracking spending against accepted work, rework, failures and delivery outcomes over time, rather than comparing raw usage or isolated productivity claims.

The source does not identify a common industry reporting standard or a scheduled follow-up study. For now, budget decisions will depend on each company’s own results, including whether its tools fit the codebase and whether reviewers and security processes can handle the resulting work.

Key Questions

How much are companies spending on AI coding tools?

The workplaces covered in the analysis reported budgets from about $150 to $3,000 per developer per month. These are examples from a 27-workplace survey, not a representative industry-wide average.

Does the evidence show that AI coding increases productivity?

Some company and vendor measures show more pull requests or stories completed, while other teams report flat output. Research results are mixed and include uncertainty, so the evidence does not establish one productivity gain that applies to all developers or tasks.

Where does the analysis say AI coding is most useful?

Reported gains are clearest in triage, code archaeology, prototyping and backlog cleanup, particularly on simple, uniform technology stacks. These examples are company accounts, not proof that every team will see the same results.

What should companies track to judge returns?

The analysis recommends measuring cost per accepted change, rework, change failure rates and work completed compared with work coming in. Pull-request volume and token consumption alone do not show whether software delivered value.

What remains uncertain about AI coding’s business case?

It remains unclear how often reported activity gains translate into lower costs, faster delivery or safer software. The cited studies and company reports use different methods, and the analysis provides no audited, comparable return figures across the surveyed workplaces.

Source: ThorstenMeyerAI.com