Promising AI

Decagon: AI agents that support customer service

Meet Decagon, an agent platform for service by chat, voice, and email, with operating rules, testing, and conversation monitoring.

Chat, telephone, and an email message connected to the same support case folder reviewed by a person.
Conceptual illustration of continuity in a support interaction across different channels. · Credit: Imagem gerada com inteligência artificial.

Explaining the same problem twice, switching channels, and starting all over again: few things make customer service so tiring. Decagon works with AI agents to organize this relationship across chat, voice, and email.

The idea is to let companies build assistants capable of talking with customers and taking part in service procedures. This brings an interesting question into focus: how do you turn an operation's rules into a service that really helps the person on the other side?

How the approach works

In its platform overview, Decagon describes Agent Operating Procedures, or AOPs. Put simply, these are operational instructions written in natural language that guide the agent's behavior.

A company can establish what to check, which information to consult, and how to handle a particular request. The platform also provides for connections to service systems, customer records, and knowledge bases.

There are features for maintaining context across sessions and channels. For the customer, the possible value is a conversation that takes previous events into account, provided the company has connected the appropriate information.

An imaginary service interaction

Imagine a store in an illustrative scenario. A person gets in touch to ask about a charge and later needs to continue the conversation through another channel. The desired workflow would retrieve the authorized history, consult the applicable policy, and indicate the next steps.

If there is an exception, the team needs to define how to route it. And any change that affects the order must comply with the established permissions. The example shows how to plan an implementation; the result depends on the connected systems and the store's rules.

The conversation is only one part of the work

One interesting part of the platform is its set of tools for testing simulated conversations, tracking decisions, and comparing versions of workflows, described in the same product overview.

This matters because a service interaction can sound friendly and still provide incomplete guidance. A careful evaluation looks at whether the request was resolved correctly, whether the customer understood the next step, and whether the team can explain what the agent did.

The ability to adjust these routines over time is what makes Decagon an interesting company to follow.

Where to start evaluating

I would begin by choosing a frequent, clearly defined request. Then I would include normal situations, missing information, and requests that fall outside the rules in the demonstration.

It is also worth checking how the handoff to a person works, which data is available to the agent, and which actions require confirmation. The experience improves when these decisions are clear from the outset. Anyone who wants to explore the approach can start with the official pages and bring a real process from their own operation to the conversation with the company.

Official information accessed on 21/10/2026.

EN

Ernani Nascimento

Developer, entrepreneur and business administrator. I build software and share what I learn along the way.

Meet the author ↗

Keep exploring.

FROM IDEA TO PRACTICE

Have something in mind?
Let’s build it.

Let’s talk ↗