As more products ship their own AI agent, the AI support chat ends up in an awkward spot right next to it. Here are the best ways to get back to one conversation.
Your product already has an AI agent. Then your support team adds an AI support tool. Now customers see two chat bubbles and have to decide which one to ask.
The product agent can work with the user's account and take actions. The support agent may have documentation, internal runbooks, troubleshooting tools, and escalation workflows. The trouble starts when a task spans both conversations and the agents don't share context.
Say a user asks the product agent to invite a teammate. It can send the invitation, but when the user asks which role to choose, it can't explain the differences in permissions. The support widget can explain those roles, but it can't see the invitation the user is working on. The user has to switch conversations, explain what they're trying to do, and bring the answer back to the product agent.
Giving the existing agent access to that product knowledge would let the user ask the question and finish the task in one conversation.
Ways to handle two chat bubbles #
There are generally three ways to deal with this that we've come across in the wild:
- Hide one of the bubbles. Put support behind a help menu, or open it only when the agent gives up. The screen looks cleaner, but the support conversation still starts from zero. The widget doesn't know what the agent already suggested, so the user explains everything again. And when the widget is hard to find, the questions come back by email or Slack.
- Use the support tool as the agent. If the product doesn't have an agent yet, this is the simplest setup: one bubble that handles product questions and support. That's what the Frigade Assistant is for. It learns the product by using it, answers questions and takes actions for the user. It doesn't help a team that already built its own agent, because now there are two agents.
- Give the existing agent a support tool. The agent already calls tools to look things up and take actions. Creating a ticket via the support agent's API is typically the escape hatch here.
Building the support tool yourself #
Pointing the agent at the help center is a reasonable first step. It drifts out of date fast, though, because the docs are typically written by hand and the product oftentimes changes much faster.
Someone also has to own the answers. Every time a setting moves, the instructions need to change, and when the agent gives bad advice, support needs to find that conversation and correct it without filing an engineering ticket. Some questions also need the agent to walk the user through the steps on the page instead of describing them. All of that can be built, but it's a lot to add to a team that's trying to ship product features.
Adding in-app support and guidance as a tool #
Most teams we talk to don't want to own and maintain a homegrown support and onboarding tool. We first ran into this when demoing the Frigade Assistant, which comes with its own chat UI. A customer building an agent-first product told us on a call that adding another chat interaction felt iffy. They already had one in the product and didn't want a second, but they were super interested in connecting Frigade as a tool their own assistant could call for support questions and product guidance.
This is what led us to build the Frigade Assist API. Users keep interacting with your agent, in your UI and your voice. One tool integration gives it new skills: answering product questions, guiding users through tasks in the app, and helping them take action. Your agent decides when to use those skills and how to respond.
Frigade supplies and maintains the product knowledge behind those skills. It learns by using your product with a browser-based agent and re-learns when you ship changes. You can also connect your docs and help content. Frigade proactively flags content that's fallen out of date, so your team doesn't have to hunt for it.
Your CS and product teams can steer how the agent helps users without writing code. Through the Frigade dashboard, MCP, and Slack, they can review conversations, correct answers, and give the agent guidance on how to handle customer questions.
Picking a setup #
If the product doesn't have its own agent yet, a support tool that brings its own agent and chat UI keeps it to one bubble, and that's what the Frigade Assistant is built for. If the team has already built an agent, give it support and product guidance as a tool instead of putting a second bubble next to it. Building that tool in-house is a fair choice too, and it gives the team full control. It needs to understand the product itself: how the screens and workflows fit together, what the user is looking at, and how to help them finish a task. That takes more than retrieving the right passage from up-to-date docs. The cost is building that product understanding, keeping it current through every release, and giving support a way to fix bad answers, on top of maintaining the agent itself. It comes down to how much of that work the team wants to own.
The Assist API is live now for teams that want to try it with their own agent.