How to Set Up a Team of AI Agents for Your Agency

What is a team of AI agents in an agency CRM?

A team of AI agents is a group of automated roles, each with a specific job, that split the work of handling, qualifying and following up on an agency's chats. In Draco, Skyline's AI CRM, that team has exactly five agents: Vega, Kai, Axel, Leo and Iris. It's not one generic bot that answers everything: it's a division of labor, similar to how you'd organize a human sales and support team.

For an agency getting leads through WhatsApp, Instagram, Messenger and TikTok, the question isn't "how do I turn on the AI" but "which agent do I activate first, and when do I add the rest?". This guide answers that, agent by agent, in the order that makes sense to turn them on.

What are the five agents and what does each one do?

Before touching any settings, it's worth understanding each role. Think of them as five distinct jobs, not five copies of the same thing.

This table already shows something important: only two of the five agents (Vega and Leo) talk directly to the traveler. The other three work behind the scenes, organizing and measuring. That changes how activation should be thought through: it's not "turn on the AI" in one move, but adding layers one at a time.

Where to start? Vega first

Vega is the agent that receives, pre-qualifies and closes in real time: it recommends from the catalog, quotes, builds packages and carries the sale through to payment. In practice, it's the front-line seller.

For an agency that has never used AI in its chats, Vega is the starting point for a simple reason: it's the only agent whose work can be validated message by message, live, while the human team still watches the conversation.

How to activate Vega without losing control

Draco starts in copilot mode: the AI drafts the reply and a person approves or corrects it before sending; it moves to autonomous when the team trusts it, and can be reverted at any time. This matters for an agency: nobody has to hand the chat over to an unknown AI on day one.

A travel agency in Cartagena, for example, can leave Vega in copilot mode for the first few weeks. The team reviews every suggested reply, approves the good ones and corrects the ones that don't match the brand's tone. Every correction feeds the AI, so the review work isn't wasted: it becomes learning. Once the team starts approving almost everything without changes, that's the natural moment to move Vega to autonomous mode in that funnel.

Adjusting Vega's intensity

Vega doesn't have a single way of talking: its intensity (low, medium, high) is set per funnel, with a hard cap of messages per reply and a fixed stance of never chasing the customer. This matters because an agency usually has more than one type of client: a lead asking about a destination wedding package isn't the same as one looking for a one-day tour.

At low intensity, Vega replies calmly, without pushing. At high intensity, it pushes harder toward closing within that same cap of messages per reply. The practical recommendation is to start at low or medium while the team learns to read how Vega responds, and only raise the intensity in funnels where lead volume is already high and there's room to test.

When should you add Kai?

Once Vega is working and the team trusts its replies, the next logical step is Kai. Kai does lead scoring: it assigns temperature, applies automatic tags and moves leads between funnel stages.

Kai doesn't talk to anyone. Its job is purely organizational: looking at each conversation and deciding whether it's a cold, warm or hot lead, and what stage of the sales process it's in. For an agency with several human agents and several active funnels, Kai solves a very concrete problem: knowing who to prioritize without having to read every chat manually.

It makes sense to activate it after Vega and not before, because Kai's scoring becomes more useful once there's a real volume of conversations, already handled by Vega, to classify. Turning it on from day one, with few chats, gives less signal.

And Leo, the one that reactivates cold leads?

Leo handles proactive follow-ups: it reactivates cold leads, sends reminders and asks for reviews. It's the agent that pursues what Vega didn't close on first contact, without a human having to remember to write back three days later.

Just like Vega, Leo has a configurable intensity (soft, medium, hard) per funnel, with the same cap of messages per reply and the same stance of never chasing the customer. This matters for the agency's tone: a follow-up in "hard" mode doesn't mean spam, it means Leo pushes a bit more within the configured limits, not that it breaks the rule of never chasing the customer.

The recommended order is to activate Leo after Kai, because Leo works better once the temperature classification already exists: reactivating a "cold" lead makes more sense once that tag has been applied consistently by Kai. A tour agency operating in Medellín handling hiking tour inquiries, for example, can start Leo in soft mode, only for leads marked as cold, and adjust based on how much people respond to those messages.

Setting Leo's intensity per funnel

Just as with Vega, it's worth having different funnels for different types of clients and assigning different intensities to each one. A corporate inquiries funnel can have Leo in soft mode, while a low-season promotions funnel can have Leo in medium or hard mode, always respecting the cap of messages per reply.

What do Axel and Iris do, and why do they come last?

Axel and Iris are the two agents that never talk to the customer. That's why it makes sense to activate them last: they don't change the traveler's experience, they change the visibility the team has over its own operation.

Axel runs a nightly analysis of how the team handled chats, both the humans and Vega, and turns that into metrics and coaching. It doesn't talk to customers. It's the agent that tells the agency's owner, for example, whether response times slipped on a Tuesday or whether a human agent is correcting Vega too much in a specific funnel.

Iris does the board triage: it reviews quiet conversations and decides whether a lead is still alive, whether it's in the wrong funnel, or whether it needs a human. Everything Iris does is auditable and reversible, and it doesn't message customers either. Iris is especially useful once conversation volume grows enough that some chats can slip through the cracks without anyone noticing.

Activating Axel and Iris after Vega, Kai and Leo has a simple logic: these two agents analyze and organize what the other three are already doing. Without data from Vega, Kai and Leo running, Axel doesn't have much to analyze and Iris doesn't have much to reorganize.

Recommended activation order, summarized

For an agency starting from scratch with Draco, this order lowers the risk of losing brand tone or control over the conversation:

1. Vega in copilot mode, reviewing every reply before it's sent.
2. Vega in autonomous mode, funnel by funnel, once the team trusts its replies.
3. Kai, to start classifying temperature and stages once there's real chat volume.
4. Leo, set to soft intensity, relying on the tags Kai has already applied.
5. Axel and Iris, to get daily visibility into the operation and catch abandoned leads.

This order isn't a rigid recipe: an agency with a very small human team might want to activate Iris earlier, if its main problem is losing track of conversations. What matters is understanding what each agent does before turning it on, and not activating all five at once on day one.

What if something doesn't work as expected?

Draco lets you go back to copilot mode at any time, for any agent that talks to customers. If Vega starts sounding off-tone in a new funnel, or if Leo pushes harder than the brand wants, the fix isn't to shut everything down: it's to lower the intensity or return to copilot mode in that specific funnel, while the rest of the agent team keeps working in the others.

This is part of why it's worth thinking in separate funnels from the start: each one can have its own intensity setting, so an adjustment in one funnel doesn't affect the others.

This topic is part of the AI CRM for Travel Agencies and Tour Operators on AI CRM for agencies and tour operators, which explains how the unified inbox and the rest of Draco's setup work.

What to do now

Before activating anything, map out the funnels the agency already has (for example, domestic tours, international packages, corporate inquiries) and decide which one will be the first to test Vega in copilot mode for one or two weeks.