Session Summary

The summary below provides a structured breakdown of the key moments from the session. It highlights the main concepts, configuration steps, and product capabilities demonstrated during the live demo, making it easy to understand how Customer Agent works and how teams can apply it in real-world scenarios.

Introducing HubSpot’s Customer Agent and Its Expanding Role

Graham opens the session by introducing himself and providing context on the evolution of HubSpot’s Customer Agent. Originally launched at INBOUND as a customer service–focused AI tool, the agent was designed to handle repetitive support questions, allowing human agents to focus on more complex, high-value interactions. The core goal was efficiency without sacrificing customer experience.

Over time, usage patterns revealed that Customer Agent was not only supporting existing customers but also engaging with prospects asking about pricing, features, and product fit. This broader adoption led HubSpot to expand availability beyond Service Hub, making Customer Agent accessible across all Pro+ hubs. As a result, the tool is now positioned as a cross-functional AI assistant relevant to service, sales, and marketing teams alike.

Creating and Configuring a Customer Agent

The live demo begins with the creation of a new agent, showing how quick and accessible the setup process is. Graham demonstrates naming the agent, assigning it a role, and selecting a personality, such as friendly or professional. He also highlights the option to apply an existing brand voice configured elsewhere in HubSpot, ensuring consistent tone across customer interactions.

This configuration stage emphasises flexibility and brand alignment. While pre-set personality options are available out of the box, teams can tailor the agent to reflect their company’s voice and communication style. The process reinforces that setting up an AI agent does not require technical expertise and can be completed in minutes.

Training the Agent with Knowledge and Content

Training is presented as the most critical factor in determining the agent’s success. Graham explains that the quality of responses depends entirely on the content provided, encouraging teams to gather strong source material before deployment. Knowledge bases, help centres, websites, blogs, landing pages, and uploaded documents can all be used to train the agent.

The demo shows how content can be selected directly from HubSpot or added via external files and public URLs. Once the content is chosen, the agent is trained automatically and notifies users when it is ready. This step illustrates how Customer Agent transforms existing documentation into a conversational support experience without duplicating work.

Testing, Deployment, and Human Handoff

Once trained, the agent can be tested using a built-in preview environment. Graham demonstrates how the agent responds to common customer questions and cites its sources, increasing transparency and trust. When ready, the agent is deployed through an assignment flow that determines which channels it operates on, including live chat, WhatsApp, Facebook Messenger, and email.

A key focus is human handoff configuration. Graham shows how the agent escalates conversations when it cannot confidently answer a question or when a user requests human support. These handoffs can route to available agents or unassigned queues, ensuring no conversations are lost while maintaining a seamless customer experience across synchronous and asynchronous channels.

Improving Performance and Introducing Actions

The session then shifts to ongoing optimisation through knowledge gap tracking. Customer Agent automatically flags unanswered or poorly handled questions, allowing teams to review conversations and close gaps by adding short answers, linking articles, or expanding content. This continuous improvement loop ensures the agent becomes more effective over time.

Graham also introduces “actions”, a powerful new capability that allows the agent to complete tasks on a customer’s behalf. By connecting to external APIs, actions enable the agent to perform operations such as resetting passwords rather than simply providing instructions. These actions are triggered by specific phrases, require user confirmation, and represent a move towards more autonomous, task-completing AI experiences.