When to create a second AI agent: three warning signs
Your AI agent is working, but something is stuck. You're wondering whether to create a second one or simply adjust the first. This article gives you the three concrete signals that indicate a single agent is no longer enough, and how to organize the transition.
- A token reservoir depleted too early can signal a need for specialization or a higher-tier plan, not always a second agent
- An agent that mixes incompatible roles (sales and administrative, customer-facing and internal) benefits from being split into two specialized agents
- Isolating low-risk tasks on a dedicated agent allows you to switch to automatic validation and reduce your workload
An AI agent that works well is reassuring. It sends emails, follows up with clients, prepares quotes. Then one day, you notice it stops before the end of the month. Or it mixes two topics that have nothing to do with each other. Or you spend too much time validating its messages. These three situations have one thing in common: they signal that a single agent is no longer enough. But creating a second agent is not always the right answer. Sometimes you need to first review the mission of the first one, or change your plan. Here's how to tell the difference.
IFirst signal: the token reservoir runs out before the end of the month
Each ABYS plan provides a token reservoir for the month. An email received, the agent thinks, it consumes tokens. When the reservoir is empty, all agents on the account stop until the first of the following month. If you regularly receive the alert at four-fifths of the reservoir as early as the third week, that's a signal. But this signal has two possible causes, and the solution is not the same.
The monthly thinking budget of your agents, shared among all.
Tokens measure the agent's work: reading an email, understanding an attachment, drafting a response. The more complex the task, the more it consumes. The reservoir is shared by all agents on the account, regardless of the plan.
First cause: your agent is doing too many different things. A carpenter had a single agent to respond to quote requests, follow up on unpaid invoices, and send updates to former clients. Three professions, three ways of thinking. The agent consumed many tokens because it constantly had to change context. By separating these missions across two agents, total consumption dropped, because each agent works faster on their specialty.
Second reason: your business has grown. You receive more emails, you have more customers to follow up with. In this case, creating a second agent won't solve anything: the pool is shared. The real solution is to upgrade to the higher plan, which gives more tokens. Before deciding, look at the dashboard: if the agent does one thing and does it well, it's the plan you need to change. If the agent does three different things, it's specialization you should consider.
IISecond signal: an agent must master two incompatible professions
An AI agent works better when its mission is clear and consistent. Responding to prospects and chasing unpaid invoices are two opposite roles. One is sales-oriented, warm, it seeks to convince. The other is administrative, firm, it reminds of an obligation. Asking a single agent to juggle between the two is like asking a salesperson to do collections: they'll end up adopting an average tone that suits no one.
A consulting firm had a single agent to welcome incoming requests and manage internal follow-ups with consultants. Emails to clients were too dry, emails to consultants too polite. Specialization changed everything: one customer agent, trained in the firm's vocabulary and offerings, and one internal agent, more direct, who knows the consultants' names and current projects.
The simple rule: if you have to give the agent contradictory instructions depending on the email type, that means there are two agents to create. A sales agent and an administrative agent. A customer agent and an internal agent. An agent for individuals and an agent for businesses.
IIIThird signal: validation workload becomes too heavy
At first, you validate each email before sending. That's normal, that's prudent. After three validations, ABYS suggests switching the agent to automatic validation. But if you don't dare take that step, and you spend an hour a day rereading your agent's emails, something is wrong. Either the mission is poorly written, or the agent is doing too many different things.
A restaurant owner used a single agent to handle reservations, privatization requests, and server applications. Three topics, three risk levels. He validated everything, out of caution. By creating a dedicated agent for simple reservations, he was able to put it on automatic after a week. Privatizations and hiring remain with a second agent, which he validates by hand, but the volume is much lower.
IVHow to decide: the three-question grid
- 1Is the token reservoir running out?
If yes, check whether the agent does several different things. If so, specialization can reduce consumption. Otherwise, it's the plan you need to change.
- 2Is the agent mixing incompatible roles?
Sales and administrative, customer and internal, individual and business. If you need to give it contradictory instructions, separate the missions.
- 3Are you spending too much time validating?
If you don't dare switch to automatic because the agent is doing too many varied things, isolate simple tasks on a dedicated agent that you can release.
If you answer yes to just one of these questions, a second agent might be the solution. If you answer yes to all three, it's certain. But always start by reviewing the mission of your first agent before creating a new one. Sometimes a better-written mission is all it takes to unlock everything.
VOrganizing two agents: who does what
Once the decision is made, the next question is distribution. Two approaches work well. The first: separate by contactOne agent talks to customers, another talks to your team or suppliers. The second: separate by task type. One agent handles prospecting, another handles administrative follow-up.
An online store chose the first approach: one agent answers buyer questions (delivery, returns, products), another manages exchanges with carriers and suppliers. The documents are different, the tone is different, the email inboxes are different. Each agent has their specialty, and the workload is better distributed.
Check what applies to you: the list remembers your answers on this device.
VIWhat changes in your plan
From the €19 excl. VAT per month plan, you can have multiple agents working. The 100,000 token pool is shared among all. If your two agents are well specialized, they will consume less per email than the old generalist agent. But if your work volume increases, you'll need to upgrade to the higher plan for more tokens. The €99 plan gives 700,000 tokens, the €199 plan gives 1.8 million.
The transition happens without losing history. You create the second agent, give it its mission and documents, test it with a dry run. Meanwhile, the first agent keeps working. Once the second is ready, you adjust the first agent's mission so it focuses on what's left.
?Questions we're asked
Can I have two agents on the 19 euro plan?
Yes, from the first plan. The 100,000 token pool is shared among all your agents. If your two agents are well specialized and work on modest volumes, this plan may be enough. Watch the alert at four-fifths of the pool to know if you need to upgrade to the higher plan.
Can my two agents monitor the same email inbox?
Technically yes, but it's not recommended. Two agents on the same inbox risk treating the same email or getting in each other's way. It's better for each agent to have their own inbox, or for you to use sorting rules to direct emails to the right agent.
How do I know if my agent is consuming too many tokens?
Check the dashboard and compare consumption per email. A specialized agent generally consumes less than a generalist agent on the same task. If you receive the alert at four fifths before the third week of the month, that's a signal to analyze.
Can I pause one agent and keep the other active?
Yes. Each agent can be paused independently. The paused agent does not consume tokens and does not work. You can reactivate it at any time.
Should I rewrite the first agent's mission when I create a second one?
Often, yes. If you remove tasks from the first agent to give them to the second, its mission must reflect this change. A shorter and more precise mission will allow it to work better.
Check the plans to see which token reservoir matches your needs.
View the packagesRead more next
Mission too broad: how to break it down for your AI agent
An AI agent you ask to do ten things at once won't do any of them correctly. After this article, you'll know how to spot a mis…
Read the lessonRunning multiple AI agents at the same time: how-to guide
You can create multiple AI agents on the same account, each with its own mission, starting from the plan at $19 excl. VAT per month. This…
Read the lessonWhen to upgrade to a higher subscription plan?
Your agent is working well, but you're wondering if your current plan is still sufficient. This article gives you the three sig…
Read the lesson