What five developers spend on AI coding tools in a month, and how a small team caps it

A year ago an AI coding assistant was a line item nobody looked at: $10 or $20 per developer per month, the same every month. In 2026 the tools do more than suggest the next line. They run for an hour on their own, read half the repository, write the tests and fix what fails, and the vendors have changed how they charge to match. The seat is still there, but more and more of the bill is usage, and usage is not the same for everyone on the team.

For a five-person team this raises two practical questions: what will we actually spend in a month, and how do we keep it from surprising us without making people ration a tool that saves them time? This article goes through current list prices, why one developer usually dominates the bill, the caps each vendor offers, and what to tell the person who signs off the spend.

What the tools charge now

List prices from the vendors' own pages, read on 15 October 2026. They change often; check the page before you budget.

- GitHub Copilot. Business is $19 per user per month and Enterprise $39, each including the same amount in AI credits (1,900 and 3,900 credits, at $0.01 each), pooled across the organization (GitHub Docs, Plans for Copilot). Since 1 June 2026 usage beyond the pool is billed by the tokens each model uses. GitHub's reason, in its announcement: "a quick chat question and a multi-hour autonomous coding session can cost the user the same amount", which agent use made unsustainable (GitHub blog, Copilot is moving to usage-based billing).
- Cursor. Teams starts at $30 per user per month with pooled usage; third-party models are "not included in the monthly prices shown" and are paid for at what you use, with spending caps you set. Granular and group-level spend limits are listed on the Enterprise plan only (Cursor, Pricing).
- Claude (Team plan). A standard seat is $20 a month billed annually ($25 monthly) and a premium seat $100 ($125 monthly); Claude Code is included in both. Each seat has a usage allowance; past it, members can use paid usage credits if the organization turns them on, with spend limits at organization, group or member level (Claude, Pricing).
- Paying per token instead of per seat. Teams that use Claude Code with an API key pay for every token. Anthropic's cost page gives the averages it sees across enterprise deployments: about $13 per developer per active day and $150 to $250 per developer per month, with 90% of users below $30 per active day (Claude Code Docs, Manage costs).

Read together: the seat price is now the floor, not the bill. For a team of five it is $95 to $150 a month before anyone does anything heavy. What happens above the floor depends on how each person works.

Why one developer is most of the bill

Usage-priced tools spread costs very unevenly. Someone who asks a question now and then, or accepts completions while typing, barely touches the allowance. Someone who hands the agent a whole feature, lets it read the codebase, run tests and try again, and runs two or three of those sessions in parallel, can use more in a day than a colleague uses in a month. Anthropic's averages say the same thing in another way: if 90% of developers stay under $30 a day and the average is $13, the rest are well above it.

A month for an illustrative five-person team paying per token, using figures in the range of those averages:

The numbers are invented for the example, but the shape is the one most teams see. Two things follow. First, an average per developer is the wrong budget unit: it is too high for four people and too low for one. Second, the heavy user is not necessarily the problem. Developer A may be shipping half the team's features. The question is what the $1,200 bought, which is a different conversation from "please use it less".

Caps that work without slowing people down

Every vendor above now has some way to cap spend. The trick is to set them so that ordinary days never hit a wall and a runaway session does.

- Know what happens at the cap. On GitHub, a user-level budget "always enforce[s] a hard stop": the user is blocked from AI-credit features until the next cycle or until an admin raises it, and there is "no automatic fallback to lower-cost models"; completions keep working (GitHub Docs, Budgets for usage-based billing). Paid usage past the pool also needs a policy switched on at all. Decide before someone is blocked at 4pm on a release day who can raise a limit and how quickly.
- Set the team budget from the floor plus one heavy user. For the team above, something like $2,000 a month with an alert at 80% catches a doubling without blocking a normal month.
- Give personal limits headroom. A per-person limit of two to three times a typical month stops a session that loops overnight without stopping the developer who had a heavy week.
- Match the seat to the person. On seat plans with tiers, the person who lives in agent sessions is often cheaper on a premium seat than on a standard seat plus usage credits. Look at a month of usage before deciding.
- Default to the cheaper model. Anthropic's own advice is that its mid-tier model "handles most coding tasks well and costs less" than the top one, and that clearing the session between unrelated tasks is one of the biggest savings, because every request re-sends the whole conversation. Set the default; let people switch up when the task needs it.
- Check the dashboard once a month. Each vendor reports usage per person. Ten minutes a month is enough to see who is at the top and ask what they were working on.

What to tell the founder

The person who signs off the bill will ask what it bought. A per-developer number on its own invites the answer "less". Put it next to two other numbers:

- What the developer costs a month. If a developer costs the company, say, $10,000 a month all in, an extra $1,000 of AI spend pays for itself if it saves about two days of their time. That is a claim you can test against what they shipped.
- What shipped and what it cost to check. More generated code means more code to review; if review becomes the bottleneck, the saving moves rather than disappears. We looked at how teams handle that in who reads the AI's code?

For the founder, one line a month is enough: total AI tool spend, spend per active developer, the largest single user and what they worked on, and whether the budget was hit. If a decision is needed between two tools or two models, test them on your own work rather than arguing from list prices; the method in fifty examples and a spreadsheet works for coding tasks too, with the developer who would do the job doing the marking.

A starting setup for five developers

None of this needs a finance team. It needs someone to own the bill, a limit that is set on purpose rather than left at the default, and the habit of asking what the spend bought instead of how to make it smaller.

Sources

GitHub Docs, Plans for GitHub Copilot (Business $19 and Enterprise $39 per seat, 1,900 and 3,900 AI credits); GitHub blog, GitHub Copilot is moving to usage-based billing (27 April 2026; AI credits from 1 June 2026; why); GitHub Docs, Budgets for usage-based billing (user-level budgets as a hard stop, no fallback model, paid usage policy); Cursor, Pricing (Teams from $30 per user, third-party models billed separately, spend controls by plan); Claude, Pricing (Team standard and premium seats, Claude Code included, usage credits and spend limits); Claude Code Docs, Manage costs effectively (about $13 per developer per active day, $150 to $250 a month, 90% under $30 a day; model choice and clearing sessions). Pages read on 15 October 2026. The five-person team, its monthly figures and the developer cost are our own illustration, not measurements.