Where AI Coding Creates Value—and Where It Adds Expense

🔍 Read the full analysis: Where AI Coding Creates Value—and Where It Adds Expense on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Engineering teams are spending between $150 and about $3,000 per developer each month on AI coding tools, but many cannot yet tie those bills to delivered value, according to an October 23 report by Thorsten Meyer that draws on interviews across 27 workplaces. The findings arrive as companies set 2027 budgets and face changing subscription terms, with reported productivity results ranging from flat pull-request counts to substantial gains in specific tasks.

The budget figures come from The Pragmatic Engineer’s October 20 reporting by Gergely Orosz, based on conversations with engineering leaders and senior engineers at 27 workplaces, plus aggregate data from code-review companies CodeRabbit and Gitar. The report describes a publicly traded company with 1,500 engineers that caps spending at $150 per developer monthly, while a two-person European .NET agency, Ix Technologies, spends about $3,000 per developer. Company size alone did not predict spending.

Reported output measures also differ. CodeRabbit said developers on its platform produced 113% more pull requests per week than a year earlier; that figure concerns the vendor’s users, not all developers. Microsoft’s 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 necessarily represent valuable work.

Interviews in the report point to task-specific gains. At one software company with about 600 engineers, overall merged-pull-request velocity barely changed, but a staff engineer said AI reduced customer-issue triage from as long as a day to minutes of agent work followed by up to an hour of checking. Ix Technologies said its uniform C# and Blazor stack helped it clear a long-standing backlog. These are company accounts, not controlled comparisons. The source also cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability; the supplied material does not provide further testing details.

Budgets Need More Than Usage Counts

The figures matter because spending may change as discounts end and billing shifts toward actual model usage. The report says several companies expect discounted subscriptions to expire, with some warning their bills could rise five- to tenfold if they do not optimize. It also says spending on API plans peaked in June and July and had roughly halved by the time of reporting, which Orosz attributed to the release of Opus 5.5 and a move toward open models.

Neither higher tool use nor more pull requests establishes that a company is delivering more useful software. Larger batches can increase review demands, while security checks and rework can offset time saved during initial coding. Meyer’s report recommends tracking cost per accepted change, rework, change-failure rates and whether teams close more work than arrives. Those measures connect tool spending more directly to outcomes, though the source does not report a common measurement standard across the 27 workplaces.

From Flat PRs to Faster Backlogs

The evidence cited in the report is mixed by design and source. Company interviews describe local experiences; vendor figures reflect users of particular products; and Microsoft’s study compares adopters with an estimated counterfactual. These measures are not directly interchangeable. In the companies Orosz interviewed, pull-request frequency ranged from flat to five or ten times higher, and one team reported unchanged frequency alongside PRs that were three times larger.

Controlled research cited by Meyer also gives a qualified picture. 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 4% for new ones, but both confidence intervals included zero. METR said the data were unreliable, in part because 30% to 50% of developers withheld tasks they did not want to do without AI. Its survey of 349 technical workers recorded a median self-reported change of 1.4 to two times the value of their work, while METR said participants in its 2025 study had overestimated AI’s effect by 40 percentage points.

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

— Gergely Orosz, summarizing company reporting in The Pragmatic Engineer

How Much Output Becomes Value?

The available evidence does not establish a single productivity effect that applies across engineering teams. Results differ by task, technology stack, experience, measurement method and source. Vendor platform data and company testimonials may not represent the wider workforce, while the cited METR estimates have uncertainty that the research group itself highlighted.

It is also unclear how often higher PR counts lead to faster delivery without added defects, security risk or review work. The source’s Veracode figure signals a potential security concern, but does not include enough information here to assess the tasks, sample or comparison baseline. The report does not provide a consistent accounting of total costs—including review, rework and model use—for all 27 workplaces.

What Companies Will Measure Next

Companies setting 2027 budgets will have to decide whether to renew discounted subscriptions, set usage caps or shift tools and models as pricing changes. GitHub’s move to usage-based Copilot billing, reported as taking effect June 1, is one example of how costs can change with heavier agent use. The source does not identify a shared timetable for the subscription discounts mentioned by interviewees.

The next useful evidence will come from organizations comparing spending with accepted changes, time saved on specific tasks, rework and software failures over a defined period. Until such data are available across more teams, the reported gains support targeted uses—such as triage, code archaeology, prototyping and backlog cleanup on simpler stacks—rather than a reliable general promise of savings.

Key Questions

How much are companies spending on AI coding tools?

The workplaces in the report described budgets from $150 to about $3,000 per developer per month. Those figures are examples from a 27-workplace survey, not a market-wide average.

Does AI coding increase developer productivity?

The evidence cited is mixed. Some sources report more pull requests or faster work on particular tasks, while other teams reported flat PR counts. METR’s cited estimates were uncertain, so the material does not establish one productivity gain for all developers.

Where does the report find the clearest gains?

Interviewees described gains in customer-issue triage, understanding unfamiliar code, prototyping and backlog cleanup, particularly where codebases use a simple, consistent stack. These are reported experiences, not controlled results across the industry.

Why can AI coding add expense?

Costs can rise with heavier usage or usage-based billing, while larger code submissions may require more review. Security issues and rework can also reduce the value of time saved during code generation.

What should companies measure?

The report recommends tracking cost per accepted change, rework, change-failure rates and whether teams close more work than enters their backlog. These measures focus on delivered outcomes rather than tokens or generated code alone.

Source: ThorstenMeyerAI.com