HubSpot Agent Builder: Build AI Agents Tailored to Your Business Processes

Artificial Intelligence is progressively taking on a more operational role within organisations. After an initial phase focused mainly on content generation, research and team assistance, the focus is now shifting towards agents capable of analysing information and executing tasks within business processes.

HubSpot takes another step in this direction with Agent Builder, a new feature, currently in beta, that allows companies to create custom AI agents directly within the platform.

In practice, organisations can define what they want an agent to do, what information it should access, what knowledge it should use and which actions it is allowed to perform.

It is another step in HubSpot's evolution into an intelligent go-to-market orchestration platform, moving beyond its role as a CRM system of record to coordinate operations through data, automation and AI.

What is HubSpot Agent Builder?

Agent Builder is HubSpot’s tool for creating and customizing AI agents tailored to each organization’s processes.

The configuration is based on four components:

• Instructions: define the agent’s role, objective, and expected outcome;
• Actions: determine what the agent can do;
• Knowledge: provides the necessary context and knowledge;
• Inputs: define the specific information received during each execution.

In this way, the agent no longer relies solely on an isolated prompt and instead works according to a predefined process.

The instructions, of course, continue to be critical to the quality of the results. This is a topic we’ve already explored in our Prompt Engineering guide for HubSpot: providing more data to AI doesn’t necessarily lead to better results without context, structure, and clear objectives.

Agents that work with CRM context

One of the main advantages of Agent Builder is the ability to use information already available within the HubSpot ecosystem.

Agents can work with data from CRM records and utilize different knowledge sources, such as documents and Knowledge Vaults. They can also access the web or connect to external systems via the Model Context Protocol (MCP).

A sales agent can, for example, analyze Contacts, Companies, Deals, and activities before preparing a briefing for a meeting. A customer service agent can consolidate a customer’s history before an interaction. In RevOps, an agent can analyze information and identify missing data or inconsistencies based on predefined criteria.

This evolution continues the path begun with Breeze Agents and AI Workspaces, bringing AI ever closer to organizations’ actual data and processes.

It also reinforces an important reality: the better the quality and consistency of data in the CRM, the more context will be available to these agents.

Workflows + agents: a new layer of automation

Workflows remain essential for automating processes within HubSpot.

If X happens, execute Y.

But there are tasks that don’t depend solely on an objective condition. Analyzing an opportunity’s history, identifying missing information, or preparing the handoff for a new client requires interpreting different elements before producing a result.

This is where agents add a new capability.

HubSpot itself provides the example of a Customer Onboarding Review Agent, which analyzes information from Contacts, Companies, Deals, and recent activity following a sale. The agent can identify the customer’s goals, purchased products, expectations, potential risks, and any missing information prior to onboarding.

Once published, agents can also be used in workflows.

Thus, a workflow can identify the moment when an action should occur, while the agent is responsible for the component that requires context and analysis.

For example: when a Deal moves to Closed Won, a workflow can trigger an agent to analyze all sales information and automatically prepare the handoff to Customer Success.

This is a significant advancement over traditional automation: not just executing tasks automatically, but being able to analyze information before executing them.

Where can they be used?

The flexibility of Agent Builder opens up various possibilities throughout the customer journey.

In Sales, it can support meeting preparation, opportunity analysis, account research, or the identification of missing information.

In Marketing, it can support research, the preparation of briefs, or the generation of outputs based on the organization’s knowledge and guidelines.

In Customer Service, it can consolidate context about customers and prepare information before an interaction.

In RevOps, it can support processes related to data analysis, quality, and consistency.

The common denominator lies in tasks that combine volume, repetition, and the need for interpretation— processes that, even with automation, still rely on manual work today.

Greater autonomy requires greater governance

As agents gain access to data and the ability to take actions, governance becomes even more important.

In Agent Builder, you can control who can run or edit each agent while maintaining existing permissions in HubSpot. Before publication, agents can also be tested to validate inputs, knowledge sources, and results.

You can also track estimated usage and set monthly limits on HubSpot Credits.

For organizations with multiple teams, complex processes, or high security requirements, this feature will be essential. As we’ve already discussed in the new HubSpot CRM governance tools, greater technological capability also means greater control over data, access, and processes.

The next step in automation on HubSpot?

Agent Builder reinforces a clear trend in the platform’s evolution.

After centralizing data in the CRM and automating processes through workflows, HubSpot is now beginning to introduce agents capable of interpreting context and performing tasks based on both.

The potential lies in reducing manual tasks, ensuring greater consistency in process execution, and making better use of the information already available in the CRM.

But, as with any automation project, the starting point shouldn’t be the technology.

Not all tasks require an agent. The most relevant use cases will be those involving repetition, high volume, available data, and a real need for analysis prior to execution.

Agent Builder is currently in beta, and using custom agents requires HubSpot Credits.

The technology now allows you to create agents tailored to each organization’s processes. The next challenge will be to identify where these agents can have a real impact on the business.

If you’d like to explore how Artificial Intelligence and HubSpot agents can be applied to your organization’s Marketing, Sales, Service, or RevOps processes, speak with a specialized consultant at YouLead.