AI coding tool budgets: seats, usage, and overages

Say your team has 20 seats at $40 a month, with a shared $600 usage allowance. The dashboard reports $900 of usage. If the contract bills only the excess, you owe $800 for the seats and $300 in overages: $1,100 before tax or credits. Adding the full $900 to the seat fee would count the included usage twice. These are fictional prices for the calculation, not a vendor quote.

Reconcile the accounts and billing periods

A developer can use a company Cursor workspace, an individual Claude Code subscription, and a separate API account in the same month. Those charges may appear on different invoices, with different billing dates. The budget needs to include each account the company pays for, including reimbursed subscriptions.

The workspace where engineers work is not always the account that pays. An enterprise account may cover several organizations. Match the invoice to the contract and billing period before assigning its charges to a team.

Keep the invoice separate from usage estimates

The invoice shows what the provider charged. A usage report may value tokens at list prices, before contract discounts or included usage. Both amounts can be correct without being equal.

Subscription allocations answer another question. Dividing a seat fee across users or requests is an accounting rule; it does not mean each request caused that much additional expense.

Keep the source and calculation with each amount. In Token Intelligence, cost-source labels distinguish these views. The integration coverage page lists which sources supply billing data, usage estimates, or subscription allocations.

Example: a pooled $600 allowance

In the fictional contract above, each $40 seat includes $30 of usage. The allowances are pooled, so 20 seats provide $600 for the team. The seat fee stays at $800 whether usage is $450 or $900. Only the overage changes.

The calculation is seat fees + max(usage - allowance, 0). Taxes, credits, discounts, and proration are left out of this example.

This formula only applies to that contract. A plan with a per-user cap behaves differently: one engineer can exceed their allowance while another leaves theirs unused. Request limits, token limits, and model-specific rates need their own calculations. For a completed billing period, use the invoice's actual usage charge once it is available.

Forecast seats and usage separately

Seat commitments give you the fixed part of the budget. Recent usage gives you a starting point for the variable part. Use invoices and activity from the same accounts and billing periods, then account for planned seat changes, model changes, and new agent workflows.

Headcount alone misses how uneven usage can be. If a few engineers run most agent sessions, their usage per active user may be far higher than the team average. Forecast from active users, then model how usage might change as more engineers adopt the tools. Write down what changes between the low and high estimates.

The budget worksheet has columns for planned and billed amounts, sources, and missing data. It is a blank template without vendor prices or formulas. Record shared charges separately unless you have an allocation rule you can explain.

What the public cost average includes

The Weave Index AI cost per employee series divides monthly attributed AI spend by employees with AI spend or code output that month. It blends Weave customer telemetry across organizations; it does not price your team's seats or usage. The public snapshot does not include monthly population counts for this series.

The Index's spend-to-output ratios use a different group: engineers with both attributed spend and merged output in the same month. Dividing the cost-per-employee average by the output-per-employee average will not reproduce those ratios. The benchmark methodology explains the populations and formulas.

Comparing the budget with engineering output

The full company bill can be larger than the spend you can assign to engineers. For example, the Index's matched spend-to-output ratios exclude organization-level overages that cannot be attributed to a person. Keep those charges in the budget even when they cannot enter the matched calculation.

For a renewal decision, compare the same team, repositories, and reporting period, and inspect the merged work, review effort, and rework alongside the cost. Weave measures output in estimated expert hours, which are neither recorded working time nor time saved by AI. The AI ROI methodology explains the calculation and what else you need to evaluate a financial return.