🔍 Read the full analysis: The Real Cost Of AI Coding: How To Measure Its Return on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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 from about $150 to $3,000 per developer each month, while companies often cannot quantify what the spending delivers. Output measures vary, and the article recommends tracking accepted changes, rework, failures and backlog progress rather than tool usage alone.
Engineering teams are spending roughly $150 to $3,000 per developer per month on AI coding tools, but many cannot say what that spend has delivered, according to an October 20 report by Gergely Orosz of The Pragmatic Engineer. Drawing on interviews at 27 workplaces and data from code-review vendors, the report describes conflicting output measures and points companies toward business outcomes such as accepted changes, rework and software failures as they set budgets for 2027.
The budget figures vary sharply even among companies of different sizes. One publicly traded company with about 1,500 engineers caps spending at $150 per developer a month; a two-person European .NET agency reported spending about $3,000 per developer. Other companies set caps between $1,000 and $4,000, with typical monthly usage around $1,000, according to the report. Company size alone did not explain the differences.
Output data does not point to one consistent productivity gain. CodeRabbit, an AI code-review vendor, reported that developers on its platform created 113% more pull requests per week than a year earlier. 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 Microsoft researchers cautioned that a merged pull request does not by itself measure the value delivered.
Across the companies interviewed, pull-request counts ranged from flat to several times higher. Some firms reported larger changes rather than more frequent ones. The article also cites a Veracode test in which 44% of AI coding tasks produced a known vulnerability. That finding concerns the tasks in the test; it does not establish that the same share of AI-assisted code in production contains vulnerabilities.
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
Budgets Need Outcome Measures
The disagreement over output matters because more code activity is not automatically more useful software. A rise in pull requests may reflect faster work, but it can also leave teams with larger changes to review, more rework or added security exposure. If organizations use raw tool consumption or code volume as a proxy for value, they may reward activity without knowing whether customers receive improvements sooner.
The report recommends tracking measures closer to delivery: cost per accepted change, rework, change-failure rate and whether teams close more user stories than enter the backlog. Those indicators can help connect spending to outcomes, although the source does not provide a standardized measurement method or a universal threshold for a worthwhile return. Companies will need to compare results against their own baseline and account for work quality and risk.
That is increasingly relevant as software budgets are set for 2027 and some discounts are due to expire. The article says several companies expect subsidized subscriptions to end, potentially raising bills five- to tenfold if usage and pricing are not managed. Such increases are reported expectations, not a guaranteed price change for every customer.
Products Worth Considering
DJI Osmo 360 Road Cycling First-Person Combo Action Camera
Bose QuietComfort Headphones – Wireless Bluetooth Over-Ear Noise Cancelling Headphones, Petal Pink Limited Edition
LG UltraGear 45GX950A-B 45-Inch 5K2K Curved OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Why AI Coding Bills Are Shifting
Orosz’s report combines interviews with engineering leaders and senior engineers at 27 workplaces with aggregate information from CodeRabbit and Gitar. The article describing it says its citations come from the report’s freely readable portion; some of the original report is paywalled. Vendor-platform figures and company interviews may not represent all engineering teams.
According to the article, spending by companies on API plans peaked in June and July and had roughly halved by the time of the report. Orosz attributed the decline to Opus 5.5 and a move toward open models. The article also says spending caps are common and often easy to bypass, making a stated limit an imperfect measure of actual usage.
Pricing changes add pressure to understand consumption. The source reports that GitHub changed all Copilot plans to usage-based billing on June 1, charging for tokens at each model’s API rate and removing a fallback to a cheaper model. GitHub explained the change by saying agentic use was becoming the default and required more computing and inference capacity. This is the company’s stated rationale, not independent evidence about how much any particular customer will pay.
“a question that few companies seem able to answer – at least for now.”
— Gergely Orosz, The Pragmatic Engineer
Productivity Gains Still Vary
The available evidence does not settle how much AI coding raises productivity across the industry. Company reports range from unchanged pull-request frequency to much higher counts, while vendor data reflects users of those vendors’ products. Neither type of measure alone establishes improvements in software quality, delivery time or customer value.
Controlled studies cited in the article also carry caveats. METR first found 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 both estimates had confidence intervals that included zero. METR called the data unreliable, noting that 30% to 50% of developers held back tasks they did not want to attempt without AI. The article also says METR participants’ self-reported estimates may overstate AI’s effect. These findings do not produce a single dependable return figure.
It remains unclear how many companies have established baselines for the suggested outcome measures, how the reported vulnerability results translate to production systems, and what actual costs will be after discounts end. The source provides examples of gains in particular teams, not proof that the same results will generalize to other codebases or organizations.
Track Delivery as Costs Change
Companies setting 2027 budgets can compare AI-assisted work with existing baselines, separating usage by task and tool from the outcome of the work. The measures proposed in the article—accepted changes, rework, change failures and backlog flow—offer a way to test whether higher spending is associated with better delivery rather than simply more generated code.
Teams will also need to account for review and security capacity. The source identifies triage, code archaeology, prototypes and backlog cleanup on simple, uniform technology stacks as areas where interviewees described clear benefits. It says complex legacy code is a less reliable setting for gains. Those are reported patterns, not a guarantee of savings; companies will need to check whether improvements persist as work moves into review and production.
The next practical test is whether organizations can maintain those outcomes as subscription subsidies expire and usage-based bills change. The report does not provide a common industry return-on-investment figure or a specific forecast for future pricing. That leaves budget owners with a local task: measure the cost and quality of delivered work, then decide whether the results justify the expense.
Key Questions
How much are companies spending on AI coding tools?
The report describes budgets from about $150 to $3,000 per developer per month across the workplaces discussed. These are reported examples, not an industry-wide average.
Does AI coding increase developer productivity?
Some companies and vendors report more pull requests or faster work, but the findings vary. The cited controlled studies have important limitations, so the evidence does not establish one productivity gain that applies across teams.
What should companies measure to judge returns?
The source recommends tracking cost per accepted change, rework, change-failure rate and whether teams close more user stories than enter the backlog. These measures focus on delivery outcomes rather than tool usage alone.
Where did teams report the clearest benefits?
Interviewees pointed to triage, understanding unfamiliar code, building prototypes and clearing backlogs on simple, uniform technology stacks. The article does not establish that these gains will apply to every team.
Is AI-generated code safe to use?
The article cites a Veracode test in which 44% of AI coding tasks produced a known vulnerability. That result does not show that 44% of AI-generated production code is vulnerable, but it supports measuring security outcomes and maintaining review.
Source: ThorstenMeyerAI.com
