GitHub Copilot Business costs $19 per user per month. Cursor Teams Standard costs $40 month to month. Claude Enterprise starts at $20 per seat, with usage billed separately at API rates. Those numbers give finance a starting point. They don’t show what the AI stack will actually cost per developer once usage starts varying across tools and workflows.
Gartner says the shift from seat-based licensing toward consumption-based pricing is creating highly variable cost structures for AI coding workloads. Anthropic makes the same point more specifically for Claude Code: per-developer costs vary widely based on model choice, codebase size, parallel usage, and automation.
The real question is how spend is distributed across developers using a stack of fixed-price and consumption-based tools.
Multiplying seat price by headcount assumes that AI tooling costs are roughly the same for every developer.
AI coding tools increasingly break that assumption.
GitHub combines paid seats with AI credits. Cursor combines seat pricing with included model usage and on-demand consumption. Claude Enterprise combines a seat charge with usage billed according to the models and tasks developers run.
That means two developers with access to the same tools can create very different bills.
Anthropic reports that Claude Code averages about $150 to $250 per developer per month across enterprise deployments, but it explicitly warns that individual costs vary widely with model selection, codebase size, multiple instances, and automation.
We see the same concentration inside real enterprise environments. In one Larridin customer environment, one engineer generated 65% of the team’s AI spend during one week while several teammates spent between $0 and $300.
The team average would still be mathematically correct. It just wouldn’t show what was actually driving the bill.
These are patterns that can exist at the same time inside one engineering organization. They’re not universal spending tiers.
For some developers, the fixed access cost makes up most of the AI bill.
They may have licenses for approved tools, but only use them occasionally. Their token or credit consumption stays low, so the bigger question is utilization: is the organization paying for access that isn’t becoming part of regular work?
Seat cost shows what the organization pays for access; consumption shows how much developers actually use.
Low usage costs can still hide wasted spend if the organization is paying for several overlapping tools that rarely get used.
Other developers move between AI tools depending on the task.
A developer might use Copilot for inline assistance, Cursor for IDE-based agent work, and Claude Code for more complex sessions. The fixed costs may be predictable, while variable usage changes from week to week.
That creates a measurement problem if every vendor dashboard is treated separately.
Our guide to tracking AI coding costs by team recommends normalizing identities and costs across the stack so one developer using several tools doesn’t become several disconnected usage records.
For this group, the useful number is the combined developer-level cost across the stack rather than each tool’s cost in isolation.
The largest differences are from developers running sustained or parallel agentic workflows.
Longer sessions, multiple instances, automation, larger codebases, and different model choices can all increase consumption. Anthropic specifically lists these factors as reasons Claude Code costs can vary substantially between developers.
Gartner says this shift toward consumption-based AI coding is making costs harder for enterprises to predict and control because usage can vary significantly across development tasks.
A heavy user isn’t automatically an inefficient user. What’s important is whether the additional spending is producing enough delivery, quality, or business value to justify it.
But that evaluation is impossible if the cost is buried inside a company-wide average.
Start with the predictable part: the minimum cost of the stack before variable consumption begins.
Track which licenses and subscriptions are assigned to each developer and what the organization pays for that access. Include overlapping tools instead of treating each procurement decision separately.
Add the consumption generated above the fixed baseline.
That might include AI credits, token charges, API usage, agent activity, or other metered consumption depending on the tool.
Our Token Spend & Insights consolidates AI spend across tools and providers so organizations can see both billed spending and the usage driving it.
Don’t stop at the average. Look at how total spend is distributed across developers. How much comes from the highest-consuming users? How wide is the gap between light and heavy users? Is a small group consistently driving most of the variable bill?
That distribution shows where costs are concentrated and which spending patterns need a closer look.
Our AI token spend attribution guidance shows why this matters: a team total can hide extreme differences between individual developers.
Finally, compare what the organization buys with what developers actually use.
A developer who meaningfully uses all three paid tools looks very different from one who regularly uses only one.
Look for dormant seats, overlapping subscriptions, and tools whose fixed access cost isn’t translating into sustained usage.
Our Token Spend & Insights surfaces unused seats and consolidates spend across multiple sources rather than leaving those costs scattered across vendor reports.
Once organizations discover that sticker price understates AI coding spend, there’s a temptation to replace it with a more realistic industry average. That still misses the point.
Anthropic’s $150 to $250 monthly enterprise average is useful context for Claude Code, but Anthropic itself recommends establishing a baseline from your own deployment before scaling.
The same principle applies across a multi-tool stack.
Use external numbers to understand what’s plausible. Use your own developer-level data to understand what’s happening.
The goal is to find out which developers are using AI lightly, which are combining several tools, which are driving high variable consumption, and whether those different spending patterns are producing value.
It’s different for everyone. The total depends on the tools assigned, pricing structure, included usage, model selection, task type, agent activity, automation, and how heavily the developer uses each tool.
Start with fixed access costs, then measure actual variable consumption across the stack.
Not as a universal AI coding benchmark. Anthropic says $150 to $250 is the average monthly Claude Code cost across its enterprise deployments, but costs vary widely among developers. Use it as a Claude Code reference point, not a benchmark for Copilot, Cursor, or a multi-tool AI coding stack.
Connect activity from each tool to the same developer identity. Then combine fixed access cost and variable consumption across the stack rather than reviewing each vendor independently.
This avoids double counting developers and shows what the organization is actually spending to support one person’s AI-assisted workflow.
Investigate the work before deciding the spending is excessive.
High consumption could reflect waste, but it could also come from valuable agentic workloads or unusually productive use of the tools. Compare the spending with delivery, quality, and business outcomes before setting limits.
Look for both unused access and unexplained consumption.
Dormant seats, overlapping subscriptions with little incremental use, and sustained high consumption without corresponding outcomes deserve different responses. Developer-level attribution helps distinguish among them.
The pricing page tells you what access starts at. It doesn’t tell you how that cost will be distributed once developers start using several tools in very different ways.
Our Token Spend & Insights consolidates AI spending across providers and connects it to the developers, teams, agents, workflows, and use cases driving the bill.
That lets finance and engineering see the lightly used seats, regular multi-tool users, and heavy consumption patterns hidden inside one average.
Book a discovery call to see what your developer AI stack actually costs.