🔍 Read the full analysis: AI Coding Spend: What Pays Back And What Doesn’t 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 based on interviews at 27 workplaces finds AI coding budgets range from $150 to about $3,000 per developer each month, while companies often cannot tie that spend to delivered value. Reported gains are strongest in triage, code research, prototyping and backlog work on uniform systems; evidence on broad productivity remains mixed.
Engineering teams are spending between $150 and about $3,000 per developer per month on AI coding tools, but many companies still cannot show what that spending delivered, according to an October 20 report by The Pragmatic Engineer based on discussions with people at 27 workplaces. The findings arrive as businesses set 2027 budgets and some discounted subscriptions approach renewal, raising pressure to measure results rather than tool usage.
The survey found no clear link between company size and spending. One publicly traded company with 1,500 engineers limits monthly spending to $150 per developer; a two-person European .NET agency spends about $3,000 per developer. Other companies set caps between $1,000 and $4,000, with typical usage around $1,000, the report said. Caps are common, but interviewees said they can be easy to bypass.
Reported output varies as much as budgets. CodeRabbit, an AI code-review vendor, says developers on its platform create 113% more pull requests per week than a year earlier. Microsoft research on its 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. Microsoft’s authors cautioned that a merged pull request does not by itself measure the value delivered.
Other workplaces reported flat pull-request counts, larger changes, or much higher output. The source report says these accounts are not directly comparable: vendor platform data, company observations and controlled studies measure different groups and outcomes. A pull-request increase may reflect more work shipped, but it does not establish that the work was useful, safe or less costly overall.
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
Budgeting Beyond Token Use
The findings matter because AI coding costs can rise quickly while the link between spending and business results remains hard to establish. The source report describes a quantitative trading firm whose monthly AI bill rose from $500,000 in January to $8 million in September, with traders accounting for the largest use. That firm uses desk-level profit and loss as a measure; many engineering organizations lack an equivalent scoreboard.
As companies move from discounted subscriptions or fixed allowances to usage-based charges, budgets may become more exposed to heavy use. GitHub said its June 1 billing change reflected the greater computing demands of agentic use. For engineering leaders, the practical question is whether tools reduce time to resolve issues or deliver accepted changes without adding rework, failures or review burden—not simply whether developers generate more code.
The report recommends tracking cost per accepted change, rework, change-failure rates and whether teams close more work than enters their backlog. Those measures do not settle every question about productivity, but they connect tool spending more directly to engineering outcomes than token counts or raw pull-request totals.
As an affiliate, we earn on qualifying purchases.
Evidence Varies by Task
Across the workplaces in The Pragmatic Engineer’s report, some teams described little change in pull-request frequency while others reported increases ranging from roughly double to five- or tenfold. One company saw pull requests stay flat as their average size tripled. The report draws on interviews and aggregate data from CodeRabbit and Gitar; its author said the cited material came from the freely readable portion of the publication.
Research outside those interviews also points to uncertainty. METR’s early-2025 study found experienced open-source developers took 19% longer when using AI on selected tasks. A February follow-up estimated an 18% speedup for returning developers and 4% for new ones, but both estimates had confidence intervals that included no effect. METR described the data as unreliable in part because 30% to 50% of participants held back tasks they did not want to do without AI.
The report’s clearest examples of value involve specific work: customer-issue triage, tracing why code was written a certain way, building prototypes and clearing backlogs on consistent technology stacks. A SaaS firm with about 600 engineers said triage that could take a day sometimes became minutes of agent work plus up to an hour of human checking. Those examples are company accounts, not controlled comparisons, and should not be treated as a guaranteed result for other teams.
“A question that few companies seem able to answer – at least for now.”
— Gergely Orosz, The Pragmatic Engineer, as summarized by Thorsten Meyer
As an affiliate, we earn on qualifying purchases.
Returns Remain Hard to Compare
The available figures do not establish a single productivity gain that applies across companies. The sources include vendor data, interviews, company-specific observations and research with different methods and task types. Their results cannot be combined into one reliable estimate of AI’s effect on engineering output.
It also remains unclear how much reported output translates into shipped customer value after accounting for human review, security checks, rework and failures. The source material cites a Veracode test in which 44% of AI coding tasks produced a known vulnerability, but it does not provide enough detail here about the test design, task mix or comparison group to generalize that figure to all AI-generated code.
Spending patterns may also change as discounted subscriptions end, billing rules shift and teams move between commercial and open models. The report says some companies expect costs to rise five- to tenfold if they do not optimize, but that is a forecast reported by those companies, not a confirmed increase across the industry.
As an affiliate, we earn on qualifying purchases.
Companies Face 2027 Budget Tests
Companies setting 2027 budgets will need to decide whether to renew discounted plans, adopt spending limits or pay usage-based rates. The source report says API-plan spending at some companies peaked in June and July and had roughly halved by the time of its reporting, which Orosz attributed to Opus 5.5 and a shift to open models. That account describes the companies discussed, not a market-wide trend.
The next useful evidence will come from teams that compare costs with accepted changes, rework and delivery outcomes over time, while accounting for task complexity and review effort. Until those measures are reported consistently, leaders and developers should treat large productivity claims as specific to the tools, teams and work studied.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much are companies spending on AI coding tools?
The workplaces in The Pragmatic Engineer’s report ranged from a $150 monthly cap per developer to about $3,000 in monthly spending per developer. The report does not establish an industry-wide average.
Does the report show that AI coding makes developers more productive?
It shows mixed evidence. Microsoft reported about 24% more merged pull requests among adopters over four months, while some companies saw flat counts and METR’s studies produced uncertain estimates. Pull-request totals alone do not establish delivered value.
Which coding tasks appear to benefit most?
Interviewees described gains in issue triage, code archaeology, prototypes and backlog cleanup, particularly on simple, uniform technology stacks. These are reported company experiences, not guaranteed results for every team.
What should engineering teams measure?
The report suggests measuring cost per accepted change, rework, change-failure rates and whether completed work exceeds incoming work. It also stresses that teams need to account for review and security effort.
Is the reported 44% vulnerability figure representative of all AI-written code?
No. The source cites a Veracode test in which 44% of AI coding tasks produced a known vulnerability, but the supplied material does not describe enough of the test methodology to generalize the result to all code or teams.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
