Token pool: how much is your AI agent consuming?
You have a token pool, but you don't know how much each agent is drawing from it. This article gives you concrete benchmarks and a method to anticipate your needs before reaching the limit.
- Consumption depends on what the agent reads and what it writes: more documents, more tokens
- With 100,000 tokens, expect around 12 overdue payment reminders or 6 simple quotes per month
- An alert email arrives at four-fifths of the pool: it's time to decide or change your plan
The token pool is the shared resource for all your agents. Each subscription plan provides a monthly quantity: 100,000 tokens for the 19 euro plan excl. VAT, 700,000 for the 99 euro plan excl. VAT, and so on. When the pool runs out, the agents stop. They resume on the first day of the following month, or immediately if you upgrade to a higher plan.
The problem is that nobody really knows how much an agent consumes per action. You see the counter go down, but you don't understand why a quote costs more than a follow-up. This article provides concrete figures, explains what makes consumption vary, and helps you forecast your needs.
IWhat is a token and why does it matter?
A token is the unit of measurement for text processed by an AI agent.
One token corresponds to approximately four characters in French, or roughly three quarters of a word. A 500-word text represents approximately 650 tokens. Each action of an agent consumes tokens to read information, think and produce a response.
The token tank works like a phone plan. You have a monthly allowance, shared between all your agents. An agent drafting a three-line email consumes less than an agent analyzing a ten-page document before responding. Consumption depends on what the agent reads and what it writes.
IIHow many tokens per type of action?
Here are orders of magnitude observed on common tasks. These figures vary depending on the complexity of the context and the length of the documents provided, but they give you a basis for your calculations.
- 1Overdue payment follow-up
Approximately 8,000 tokens. The agent reads the invoice, checks the customer history, drafts a short personalized email. Few documents, brief message.
- 2Simple quote
Approximately 15,000 tokens. The agent analyzes the request received, consults your pricing grid, calculates and drafts a structured quote with terms.
- 3Milo prospecting email
Approximately 25,000 tokens. Milo reads the target company's website, analyzes its activity, personalizes the message based on what it understood. More reading, more consumption.
- 4Answer to a simple question
Approximately 3,000 tokens. An agent answering a factual question without consulting a large document consumes little.
- 5Long document analysis
Approximately 40,000 tokens or more. An agent that must read a twenty-page contract before drafting a summary consumes much more.
IIIWhy do some agents consume more than others?
Consumption depends on three factors. The first is the amount of input text: an agent that must read your fifty-page product catalog with each action consumes more than an agent that only needs a customer file. The second factor is task complexity: drafting a personalized email requires more thought than sending a standard confirmation.
The third factor is the length of the response produced. A detailed two-page quote consumes more than a three-line acknowledgment. If you notice that an agent consumes a lot, first check the documents it consults. Perhaps it doesn't need to reread everything each time.
IVHow to monitor your token counter?
ABYS sends an alert email when you reach the four fifths of your monthly tank. This is the signal to check where you stand and decide whether you need to slow down certain agents or upgrade to a higher plan. The counter is visible in your space, with details by agent.
Get into the habit of checking this counter once a week at first. You will quickly understand which agents consume the most and why. Some users discover that a single agent accounts for three-quarters of their consumption, simply because it works with documents that are too heavy.
Check what applies to you: the list remembers your answers on this device.
VWith 100,000 tokens, how many actions are possible?
The plan at 19 euros excl. VAT per month offers 100,000 tokens. Here is what this represents in practice, depending on the type of task. These estimates assume documents of reasonable size and responses of standard length.
By combining several types of actions, a tradesperson could for example handle 6 overdue reminders and 3 quotes per month with the basic plan. A consulting firm that does a lot of document analysis will need a higher plan. The key is to know your actual usage.
VIWhat to do when approaching the limit?
When the alert email arrives, you have several options. The first is to do nothing: your agents will stop when the tank runs empty and resume on the first of the month. This is acceptable if you can wait a few days.
The second option is to pause the least urgent agents to preserve tokens for priority tasks. An agent on pause consumes nothing. The third option is to upgrade to a higher plan, which immediately recharges your tank and increases your monthly capacity.
VIIHow to reduce an agent's consumption?
If an agent consumes too much, start by reducing the documents it consults. A hundred-page catalog can often be replaced by a ten-page summary sheet. The agent will work faster and consume less. Also check whether the agent really needs to reread everything with each action.
You can also simplify the agent's mission. An agent that must write a personalized email AND check three databases consumes more than an agent that does one thing well. Sometimes, creating two specialized agents costs less than a single multi-purpose agent.
VIIIPlan your token budget over several months
After one or two months of use, you will know your actual consumption. Note how many tokens each agent uses on average per week. Multiply by four to get a monthly estimate. If you are regularly close to the limit, the higher plan will prevent interruptions.
A business that uses one agent for quotes and another for follow-ups can expect about 20,000 tokens per week, or 80,000 per month. The 19 euro plan excl. VAT is sufficient, with a safety margin. A firm that adds a document analysis agent will likely need to upgrade to 99 euros excl. VAT per month.
?Questions we're asked
Is the token tank shared between all my agents?
Yes, the tank is common to all agents on your account. Each action by each agent draws from the same monthly allowance. This is why monitoring each agent's consumption allows you to better allocate your resources.
Can I purchase additional tokens during the month?
You can upgrade to a higher plan at any time. This immediately recharges your tank and increases your monthly capacity. The change takes effect right away.
How do you know which agent consumes the most?
Your space displays consumption details by agent. Check it regularly to identify resource-hungry agents and adjust their documents or mission if necessary.
Does an agent on pause consume tokens?
No, an agent on pause consumes nothing. Pausing an underused agent allows you to preserve tokens for priority agents, especially at the end of the month.
Check the plans to find the tank suited to your needs.
View the packagesRead more next
What happens when the token reservoir is empty?
Your AI agent works, then one day it stops and alerts you that the tank is empty. This article explains exactly…
Read the lessonWhich subscription plan should you choose for your AI agent?
You are hesitating between the 19 euro plan and the 99 euro plan. By the end of this article, you will know which one matches your act…
Read the lessonAI tokens: what an agent consumes and how to predict it
Tokens are the unit of work for AI: everything an agent reads and everything it writes is counted in tokens. After this course, you…
Read the lesson