🔍 Read the full analysis: AI Coding ROI Questions Every Team Should Ask on ThorstenMeyerAI.com
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
TL;DR
A report drawing on interviews with people at 27 workplaces found AI coding budgets range from $150 to about $3,000 per developer per month, while companies disagree on how much output has changed. The evidence points to gains in some tasks, but does not establish a universal return on investment; teams are urged to track accepted work, rework and delivery outcomes rather than tool usage alone.
Engineering teams are spending from $150 to about $3,000 per developer per month on AI coding tools, but many companies cannot yet say what that spending has produced, according to an October 23 report by Thorsten Meyer. The evidence he cites shows varied results across workplaces, making measurement of delivered value a live issue as companies set 2027 budgets.
The report draws on an October 20 analysis by Gergely Orosz of The Pragmatic Engineer, who spoke with engineering leaders and senior engineers at 27 workplaces and included aggregate data from AI code-review vendors CodeRabbit and Gitar. The budget range was broad: one company capped spending at $150 per developer a month, while a two-person European agency reported spending about $3,000 per developer. Meyer says company size did not predict spending levels.
Output measures also diverged. CodeRabbit reported that developers on its platform created 113% more pull requests per week than a year earlier. In a study of tens of thousands of engineers during an early-2026 rollout, Microsoft found adopters merged about 24% more pull requests over four months than they otherwise would have, according to the report. Microsoft’s researchers cautioned that a merged pull request is not the same as delivered value. Several interviewed companies reported flat PR counts, while others reported increases.
Controlled studies are less conclusive. Meyer cites METR research that estimated experienced open-source developers took 19% longer with AI in early 2025; a February follow-up estimated an 18% speedup for returning developers and 4% for new developers, but the confidence intervals included zero. METR also described the data as unreliable, in part because some participants held back tasks they did not want to attempt without AI. Separately, Veracode’s latest test found 44% of AI coding tasks produced a known vulnerability. These findings do not establish that all AI-generated code is insecure, but they underline the need to account for review and security work.
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
What Teams Should Count as Value
The central issue is that tool activity is not the same as business output. More prompts, tokens or pull requests can indicate increased use, but do not by themselves show that a team shipped more useful work, reduced delays or avoided rework. Microsoft’s qualification about merged PRs reflects that distinction.
Meyer’s report points to task-level benefits in customer issue triage, code archaeology, prototypes and backlog cleanup, particularly on simple, uniform technology stacks. One staff engineer at a company with about 600 developers said customer issue triage that could take most of a day was reduced to minutes of agent work and up to an hour of human review. That is an interviewee’s account, not a controlled measurement. At a small agency using C# and Blazor, the cofounder said AI helped the firm work through a long backlog; the company’s reported pace is not directly comparable with larger teams.
Costs can also rise without matching gains if teams reward token consumption, produce larger but not better changes, or leave human reviewers and security checks overloaded. The report recommends tracking cost per accepted change, rework, change failure rate and whether stories are being closed faster than new ones arrive. Those measures connect spending to completed work and quality more directly than usage totals.
As an affiliate, we earn on qualifying purchases.
Why Budget Decisions Are Due Now
The timing reflects changing prices and payment models as well as annual planning. Orosz’s October 20 report said spending on API plans had peaked in June and July and roughly halved afterward among the companies discussed. He attributed the decline to the availability of Opus 5.5 and a move toward open models. That is his explanation for the pattern, not a universal market measurement.
Several companies told Orosz that discounted subscriptions were due to end, potentially raising their bills substantially if usage and plans stayed unchanged. Separately, GitHub changed Copilot billing on June 1 to usage-based charges tied to models’ API rates and removed 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. These shifts make it harder for teams to set budgets using last year’s flat subscription costs.
The report also describes cases where spending is assessed against a direct financial measure. A quant trading firm’s AI bill reportedly rose from $500,000 in January to $8 million in September, with traders accounting for its largest spending. The firm uses each desk’s profit and loss as its measure. Most engineering organizations do not have such a direct scoreboard, according to the report, so they need other ways to connect tool costs with outcomes.
““a question that few companies seem able to answer – at least for now.””
— Gergely Orosz, in The Pragmatic Engineer report as quoted by Thorsten Meyer
How Much Output Becomes Delivered Value
The available figures do not provide a single, comparable estimate of AI coding’s effect across engineering teams. They come from different settings: vendor platform data, a Microsoft study, company interviews and controlled experiments with their own limitations. A rise in PRs may reflect more completed changes, larger changes, or both; it does not alone establish improved quality, customer impact or cost savings.
The source material does not provide enough detail to determine how the reported gains compare across tools, team types or the full cost of review, rework and security. The 44% Veracode result concerns the tasks in its test and should not be treated as a rate for all AI-written code. The exact return on investment for most companies remains unestablished, as do the lasting effects of changing prices, expiring discounts and shifts to open models.
What Budget Owners Can Measure
Companies setting 2027 engineering budgets will need to decide which tools and usage plans to fund as discounts expire and billing models change. The report does not identify a single forthcoming industry benchmark or confirm that companies will adopt a common ROI method.
A practical next step, based on the measures Meyer lists, is for teams to compare AI-assisted work with their own baseline: track spending against accepted changes, monitor rework and change failures, and check whether completed stories outpace incoming work. Those results can show where a tool helps a particular team and where additional review or security effort offsets the gain. Until teams publish comparable outcome data, claims about broad productivity gains should be treated as dependent on the task, workplace and measurement method.
Key Questions
How much are companies spending on AI coding tools?
The workplaces described in the report ranged from a $150 monthly cap per developer to about $3,000 per developer per month. The figures are examples from the surveyed workplaces, not a market-wide average.
Does the evidence show that AI coding increases productivity?
Some sources report more pull requests or faster work on selected tasks, while other companies reported flat PR counts. Controlled METR estimates cited in the report were uncertain, with confidence intervals including zero. The evidence does not establish one productivity effect for every team.
What should engineering teams track to measure ROI?
The report suggests measuring cost per accepted change, rework, change failure rate and whether stories close faster than new ones arrive. Teams should compare these outcomes with their own baseline rather than treating tokens or PR volume as value on their own.
Does the Veracode result mean 44% of all AI-written code is vulnerable?
No. The report says 44% of AI coding tasks in Veracode’s latest test produced a known vulnerability. That test result should not be generalized to all AI-generated code or every engineering team.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
