Three questions to ask about your AI bill
By Team Two · Last updated September 23, 2026 · 3 min read
In a pilot, nobody looks at the AI bill. Ten people try something new, the invoice is small, and the conversation is about whether it works at all.
Six months later, when a few hundred people use agents every day, the conversation changes. Finance wants to know what the money buys, which teams drive it, and whether it will double next quarter. The time to answer those questions is before you roll out, not after.
1. Can you see what each agent costs?
A single monthly total tells you nothing. To manage cost you need to see it where decisions are made: per agent, per team, per day.
Team Two tracks input and output tokens for every agent, rolls them up daily, and shows usage per user and per team, without anyone having to read conversations. When a cost spike happens, you can see which agent caused it and when.
2. Can you choose the model per job?
Using the most capable model for every request is the simplest setup and usually the most expensive. A short classification step or a routine lookup doesn't need the same model as reviewing a contract.
In Team Two you can set the model per agent, or let workspace policy decide. The step that decides which agent should handle a new request uses a small, fast model by default. You can also route specific agents to a model you host yourself, which turns per-token cost into infrastructure you already pay for.
3. What happens when something goes wrong?
Two things quietly drive AI cost: runaway requests and outages.
- Runaway requests: an agent that loops through tool calls, or delegates to many specialists, can burn through tokens fast. Team Two limits tool steps per request, how many specialists can run at once, how many a single request can start, and how long each can run.
- Outages: if your only model provider is down, your agents are down. Team Two can fall back to another provider you have allowed, so work continues.
How pricing works in Team Two
Our plans include a monthly token allowance. If you go over it, overage is billed at the rate shown on the pricing page, calculated daily and invoiced at the end of the month. Agents keep working. We don't cut them off mid-conversation.
On the Business and Enterprise plans you can bring your own Anthropic or OpenAI keys. Model usage is then billed to you directly by the provider, and we don't add a markup to it.
A simple rollout rule
Start with one or two agents on a clear use case. Look at usage per agent after two weeks. Move routine agents to a smaller or local model where quality holds up. Then roll out wider. It is much easier to set expectations with real numbers from your own workspace than with estimates from a vendor, including us.
Want to model it for your team? Try the ROI calculator or talk to us.
Frequently asked questions
What counts as a token?+
Tokens are the units models use to process text, roughly four characters of English each. Both what goes into the model and what comes out count.
Can we cap spend?+
Plans include a fixed token allowance, and usage is visible per agent and team so you can spot growth early. If you need a hard spending cap, talk to us about your requirements.
