Pay for it, sheet 08 of 13
Seats versus metered usage, per-person caps that mostly don't exist, and plan terms that decide what you may expense.
Find the one place your platform enforces a hard per-person cap. Everywhere else, a cap means a gateway you run.
← → move to the next sheet over, Esc zooms out.
There are two ways to pay for AI at work, and most teams use both. A seat is a flat monthly price per person. Metered usage bills by the token or the request, as it happens. Agents live on the meter, and an agent never gets tired.
A seat is a gym membership: one price, use it as much as you like. Metered usage is a taxi meter, and the agent is a passenger who never asks to get out. Most companies are paying for both, on different invoices, owned by different people.
Budgets and alerts for the whole organization are easy to find. A hard cap per person is rarer, and where it exists it usually lives in one product or one tier. Everywhere else, a per-person cap means a gateway you run yourself.
Every link opened on 11 Nov 2026. The numbers match the squares on the map.
Reference, Anthropic
Every Claude plan in one table: individual tiers, Team for 2 to 150 seats in Standard and Premium, and Enterprise, with API pricing alongside.
The catch. Prices and limits change. Re-read the page before every renewal, not just the first one.
Built in
How Claude Code bills, how to track usage, and how to set team spend limits, with what each plan shows people about their own spend.
The catch. Limits differ by plan. Check which ones apply to yours before you promise a cap.
Built in
Budgets across GitHub products. For metered products such as Copilot AI credits, a budget can stop usage once the threshold is reached.
The catch. A budget alone only alerts. And GitHub notes you can overshoot in the first cycle after you create one.
Open source
An open-source gateway in front of many model providers, with budgets and rate limits on keys, teams, tags and end users.
The catch. On March 24, 2026, two of its PyPI releases carried a credential stealer, and LiteLLM says anyone who installed between 10:39 and 16:00 UTC may be affected. The official Docker image wasn't. A gateway holds every key you have, so pin versions.
Product, Cloudflare
A managed gateway for AI traffic with analytics, caching, rate limiting, model fallback, and spend limits that block requests once a budget is spent, per user when the gateway sits behind Cloudflare Access.
The catch. Spend limits are in beta. Test what happens at the limit before you rely on it.
Product
Cost reporting for Anthropic spend, from the Claude API and Claude.ai Enterprise. The Admin API key it needs has read-write access, and Vantage says it only reads.
The catch. It reports spend and can't stop it. API-key costs are tagged by who created the key, which may not be who used it.