🔍 Read the full analysis: AI Coding’s Payoff Depends On What You Measure on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A report drawing on interviews at 27 workplaces finds AI coding budgets range widely, while reported output gains vary from flat to several times higher. Vendor figures and company examples point to gains in tasks such as triage and backlog work, but higher activity does not establish business value; the report recommends measuring accepted changes, rework and delivery outcomes.
Engineering teams are spending $150 to about $3,000 per developer per month on AI coding tools, but a review of 27 workplaces finds no consistent answer to whether that spending produces business value. The report, published by Thorsten Meyer on October 23 and drawing on The Pragmatic Engineer’s interviews and vendor data, says output measures range from flat pull-request counts to large reported increases, making the choice of metric central as companies set 2027 budgets.
The figures come from The Pragmatic Engineer’s October 20 report by Gergely Orosz, which gathered input from engineering leaders and senior engineers at 27 workplaces and included aggregate data from code-review vendors CodeRabbit and Gitar. Meyer says his account uses material from the report’s freely readable portion. Spending does not appear to track simply with company size: one publicly traded company with 1,500 engineers caps spending at $150 per developer monthly, while two-person European agency Ix Technologies spends about $3,000 per developer.
Reported output varies by source and workplace. CodeRabbit says weekly pull requests on its platform rose 113% year over year; that is vendor data from teams using its AI code-review product. Microsoft researchers, studying tens of thousands of engineers during an early-2026 rollout of Claude Code and GitHub Copilot CLI, found adopters merged about 24% more pull requests over four months than they otherwise would have. The researchers cautioned that a merged pull request does not by itself show the value delivered.
Across the 27 workplaces, the report describes results from unchanged pull-request frequency to increases of five to ten times. In one company, the frequency stayed flat while pull requests grew to three times their previous size. Those differences make raw activity an incomplete measure: a larger or more frequent change can still require substantial review or rework, and the cited figures do not establish a common, controlled productivity effect.
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 Activity Is Not Business Value
The budgeting question matters because AI tool costs can change quickly, while a count of prompts, tokens or pull requests cannot show whether customers receive better software sooner. A company can record more code activity without closing more useful work, and larger changes may place added pressure on review and security teams. The report cites Veracode testing in which 44% of AI coding tasks produced a known vulnerability; that result is a reason to track quality alongside speed, not evidence that every AI-generated change is unsafe.
Several examples in the report suggest the benefits may be clearest in specific workflows. A staff engineer at a SaaS company with about 600 engineers said customer-issue triage that could take a day was reduced to minutes of agent work plus up to an hour of human checking. The engineer also described code-history research taking minutes rather than a day or two, and a complex prototype taking hours rather than one or two weeks. These are reported experiences, not a company-wide controlled estimate of savings.
Meyer’s account recommends tracking cost per accepted change, rework, change-failure rates and whether teams close more stories than arrive. These measures connect AI use to accepted work and delivery consequences, rather than consumption alone. That distinction is relevant as some companies’ discounted subscriptions approach expiry and usage-based pricing puts more of the cost directly on teams.
As an affiliate, we earn on qualifying purchases.
Budgets Meet Mixed Productivity Evidence
Orosz reported that spending on API plans peaked in June and July and had roughly halved afterward for some companies, which he attributed to the release of Opus 5.5 and a move to open models. Several companies told him that discounted subscriptions were due to end, creating a risk of bills rising five- to tenfold if usage and plans were not adjusted. These are reported company experiences, not a forecast for every buyer.
Pricing has also shifted. GitHub moved all Copilot plans to usage-based billing on June 1, according to the source account, charging 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 brings higher compute and inference demands. The change makes usage and cost more visible, but does not itself measure whether the work is useful.
Research cited in Meyer’s report adds uncertainty to headline productivity numbers. 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 called its data unreliable, noting that 30% to 50% of developers held back tasks they preferred to do without AI. In a survey of 349 technical workers, the median reported a 1.4-to-2-times change in work value, while METR said its 2025 participants overestimated AI’s effect by 40 percentage points.
““Agentic usage is becoming the default, and it brings significantly higher compute and inference demands.””
— GitHub, explaining its Copilot billing change
How Much Output Becomes Value?
The available figures do not establish one reliable productivity gain that applies across teams. They combine vendor platform data, company reports, a large Microsoft study, controlled experiments with limitations and workers’ self-assessments. The results concern different populations, tasks and time periods, so they cannot be treated as directly comparable measures.
It is also unclear how much of any increase in activity becomes accepted, maintained software or measurable customer benefit. The source report does not provide a shared baseline for cost per accepted change across the 27 workplaces, nor a consistent accounting of rework, security issues or long-term maintenance. Company accounts may show where teams see practical gains, but do not prove that the tools caused the same outcomes elsewhere.
What Companies Should Track Next
Companies setting 2027 budgets will need to compare tool spending with measures tied to completed work and quality, rather than treating usage as a result. The report points to tracking cost per accepted change, rework and change failures, alongside whether completed stories outpace incoming work. Teams can also compare particular workflows—such as triage or backlog cleanup—against their own earlier performance, while keeping task difficulty and review time visible.
The source material does not name a common industry standard or a scheduled follow-up study. For now, companies will have to decide which outcomes fit their products and teams, and test whether gains persist as discounts end, pricing changes and AI-generated work moves through review and into production.
Key Questions
How much are companies spending on AI coding tools?
The report describes budgets ranging from $150 to about $3,000 per developer per month among the workplaces cited. The figures vary by company and are not a market-wide average.
Do the reported figures prove AI coding makes developers more productive?
No single productivity effect is established. The sources report different results, from flat pull-request counts to increases, and the studies cover different teams and tasks. A higher count of merged changes does not by itself show greater business value.
Where did teams report the clearest practical gains?
Examples in the report include customer-issue triage, researching why legacy code was written a certain way, prototyping and backlog cleanup on simple, uniform technology stacks. These are reported cases, not guarantees for other teams.
What should an engineering team measure?
The report recommends connecting cost to outcomes, including cost per accepted change, rework, change-failure rates and whether teams close more user stories than arrive. Review effort and security should also be considered when interpreting output counts.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
