Updated October 8, 2026.

How does HubSpot Agent Hub work?

Agent Hub brings agent discovery, configuration, management, and outcomes into one HubSpot workspace. The Intro tab shows active capabilities and results, the Agents tab supports agent management, and the Context tab holds business information and knowledge vaults. Individual capabilities have their own configuration and eligibility requirements.

Source: HubSpot’s Agent Hub documentation.

Once your team starts using AI agents, the ongoing work becomes practical: reviewing outputs, maintaining instructions, resolving exceptions, and deciding what to expand.

This guide explains the workspace and how to build a manageable operating routine around it.

For product orientation, start with What Is Agent Hub?. For naming questions, read Agent Hub vs. Breeze.

Where do you find Agent Hub?

Navigate to More → Agents → Agent Hub, or directly to Agents → Agent Hub if More is absent.

On the Intro tab, active capability cards show outcomes; inactive cards describe capabilities. Use Manage for active features and Activate for inactive ones.

Source: Get started in Agent Hub.

Before activating something new, identify the process it should support. Give it an owner and a measurable objective.

For example, “improve account preparation” needs a definition: what information should be included, who uses the output, and how you will judge whether preparation has improved.

What should you check on the Intro tab?

Use the displayed outcomes as a starting point for investigation.

Ask whether the activity matches your intended use case. If output volume increases, check whether the results remain accurate and useful. If activity drops unexpectedly, investigate configuration, inputs, and usage before assuming adoption has stalled.

An operational review should connect three things:

• What work occurred.
• Whether the output met expectations.
• Whether the process improved.

A busy agent is not automatically a valuable agent.

How do you configure an agent?

Agent configuration covers inputs, instructions, tools, knowledge sources, and automation settings.

To create an agent, select Create → Agent from the Intro tab and continue through Agent Builder.

Source: Create and customize agents.

Write the task clearly before configuring it. Specify the intended output and how the agent should handle incomplete information.

Compare these hypothetical instructions:

Broad instruction: “Research this account.”

More useful instruction: “Identify the account’s current products and recent expansion announcements. Separate confirmed findings from uncertain information and identify missing evidence.”

The second gives reviewers a clearer basis for evaluating results. Confirm that the selected tools and sources support the instruction before testing it.

What role does CRM context play?

An agent’s output should be evaluated against the information it can actually use.

For an account-research process, ask whether the intended company record is correct. For a support process, review the source material. For a sales-follow-up process, check whether the relevant conversation and record information are available.

Avoid assuming every agent has identical access or interprets every kind of information the same way.

Prepare the data needed for the specific task. Duplicate records, ambiguous property definitions, and outdated documents deserve attention when they affect the workflow being tested.

Read Do I Need Clean CRM Data Before Using Agent Hub? for a readiness review.

What belongs in the Context tab?

The Context tab holds business information such as messaging, tone, brand, and ideal customer profiles. Knowledge vaults hold more specific material for targeted use cases.

Source: Manage AI context.

Assign responsibility for maintaining that information. Product changes, revised positioning, and updated internal policies should trigger a review of the relevant material.

Treat context maintenance as part of operating the agent. An output problem may originate in outdated guidance rather than the task instructions themselves.

When investigating an error, check the source material before adding more instructions to compensate for it.

How do you manage existing agents?

Use the Agents tab to manage agents. HubSpot documents Clone and Delete options in the three-dot menu. Deletion is permanent and removes associated agent data.

Source: Manage agents in Agent Hub.

Before making a substantial change, document the current purpose, configuration, and dependencies. Decide how you will test the revised behavior and who will review the results.

Check the controls available for the specific agent type rather than assuming every capability supports the same lifecycle actions.

How do agentic workflows fit in?

Agentic workflows combine triggers, actions, and agents in a multi-step process. HubSpot documents creating them through Create → Workflow and requires a Professional or Enterprise subscription.

Source: Agentic workflows in Agent Hub.

Plan the sequence before building it:

1. Define what starts the workflow.
2. Identify the information required.
3. Specify the output of each step.
4. Decide where review or exception handling belongs.
5. Define the final result.

For a hypothetical research workflow, a team might investigate selected account signals, evaluate the findings, and route useful results to an account owner. Confirm the supported actions and approval behavior before implementing that sequence.

Test individual steps before increasing volume. A correct first step does not establish that the entire process works as intended.

How should teams review agent output?

HubSpot provides an agent inbox for running agents and reviewing output.

Source: Run agents and review output in the agent inbox.

Give reviewers clear acceptance criteria:

Record recurring problems. A consistent pattern is easier to investigate than a collection of isolated complaints.

Separate a source-data problem from an instruction problem, a configuration problem, or an unrealistic expectation about the tool.

Where does human oversight belong?

Define oversight around the consequences of the work.

An internal research draft may need a different review process from a customer-facing message or a change to a forecasting field.

Check the actual approval controls for each capability. Do not assume every action always requires confirmation—or that everything runs automatically.

Document who reviews outputs, who resolves exceptions, and who can change the configuration.

For detailed guidance, read HubSpot AI Governance: How Data Access, Privacy, and Approvals Work.

A practical operating routine

Choose a review cadence that matches the volume and impact of the workflow.

Use a baseline to assess improvement. Track time recovered alongside the effort required to correct outputs.

If an agent creates more work than it removes, investigate before increasing volume. If results are useful, identify what conditions made the pilot successful and preserve those conditions as you expand.

Where to go for individual-agent setup

Use the resource that matches your task:

• How Does HubSpot’s Data Agent Work?
• How Does HubSpot’s Customer Agent Work?
• How to Set Up HubSpot’s Prospecting Agent
• What Is Deal Progression?

For a first deployment, read Getting Started With Agent Hub. For team rollout planning, use How to Deploy AI Agents in HubSpot.

Fast Slow Motion’s approach

At Fast Slow Motion, we start with the process: what needs to happen, what information supports it, and who owns the result.

Then we configure automation and establish visibility into its performance.

Operating agents well requires maintaining that foundation. Clear instructions, current information, accountable ownership, and regular review give your team a practical basis for improving and expanding the work.

Need help configuring or managing HubSpot agents? Talk with Fast Slow Motion about building a workflow your team can measure and maintain.