HubSpot Agent Hub: How to Build AI Agents Without Losing Control

AI agents inside your CRM are no longer some far-off idea.

HubSpot is putting the tools to build and deploy them directly into the platform with the new Agent Hub. And if you've spent any time looking at it, the first reaction is probably the same one we had:

This is really cool.

But there's another reaction that matters just as much:

Whoa. We need to be careful with this.

That's the part of HubSpot Agent Hub that doesn't always get enough attention. The technology is exciting, but AI agents aren't just another type of workflow. They can make decisions, access information, use tools, interact with other systems, and potentially take actions on your behalf.

So before you start building agents all over your HubSpot portal, it's worth understanding what you're actually building—and how to build it responsibly.

What Is HubSpot Agent Hub?

HubSpot has been incorporating AI agents into the platform for a while.

HubSpot already offers agents designed for specific jobs, including company research, customer health, deal loss analysis, video search, blog research, and AEO.

Historically, these were largely pre-built HubSpot experiences.

Agent Hub changes the equation.

Instead of simply using agents that HubSpot has created, you can start configuring and building your own.

That means you can create AI-powered processes around the way your specific business operates.

And that's where things get really interesting.

HubSpot's Featured Agents

One of the most interesting parts of the experience is the collection of featured agents built around different stages of the customer journey.

For example, an AEO agent can provide recommendations designed to help your business get found online.

A data agent can help enrich and score prospects, segment them, and help determine who should receive the lead.

A prospecting agent can analyze CRM activity, recommend follow-up actions, draft responses, and potentially send responses on your behalf.

Agents also focus on deal progression and customer-related activities.

Taken together, the direction is pretty clear.

AI agents are moving into virtually every part of the customer lifecycle.

Agent Hub Lets You Build Your Own

This is where Agent Hub really separates itself from traditional AI features.

You can start with an existing template and customize it, or build an agent from scratch.

You can define the instructions.

You can choose the actions it can take.

You can provide inputs.

You can connect other applications.

And perhaps most importantly, you can give the agent knowledge about your business.

That last part is easy to underestimate.

Your AI Agent Needs a Brain

We've all seen AI-generated content that sounds like it was written by absolutely nobody.

It's generic.

It's repetitive.

It has the same structure and the same tone as everything else AI is producing.

That's why knowledge matters.

If you want an AI agent to produce useful outputs—whether those are internal summaries, CRM updates, emails, website content, or other communications—you need to give it enough information to understand your business.

That includes things like:

• What your company does
• Who your customers are
• What products and services you offer
• What matters to your organization
• What the agent should and shouldn't do
• How you communicate
• What your preferred tone and style look like
• What information it should use when making decisions

Writing a few instructions takes minutes.

Actually teaching an AI agent how your business works takes considerably more effort.

But that investment is what separates useful AI from AI slop.

Agentic Workflows Change the Automation Game

HubSpot's traditional workflow builder is already extremely powerful.

You choose a trigger, select the object you're working with, and then define the actions that should happen.

For example, a contact submits a form. That triggers a workflow. The workflow can create records, update properties, send notifications, create tasks, or perform other predefined actions.

Agentic workflows introduce another layer.

You can still use triggers and actions, but you also have options to let AI reason through a process and work with more information.

You can run processes on schedules.

You can respond to company signals.

You can work with connected applications.

You can use AI agents you've already created.

And you can build much more sophisticated processes around what is happening inside your CRM and connected systems.

The possibilities are enormous.

Which brings us to the part that deserves just as much attention.

AI Agents Come With Responsibility

There's a reason to slow down before you start building dozens of agents.

AI agents can consume resources and perform actions at scale.

In the video, CJ shares a real example from another HubSpot partner who built an AI agent, let it run overnight, and woke up to find it had used all his available credits because it ran against every deal in his pipeline.

That's a pretty good reminder that "let's see what this does" isn't necessarily a great AI governance strategy.

Before deploying an agent, you need to understand what it will receive, what it is instructed to do, what tools it can access, and what it can actually change.

The Four Things Every AI Agent Needs

A useful mental model is to think of every agent as having four fundamental components.

1. Inputs

What information is the agent receiving?

That could be a contact, company, deal, form submission, question, conversation, or something else.

2. Instructions

What are you telling the agent to do?

Clear instructions matter enormously.

3. Tools

How does the agent actually take action?

It might browse the web, create an email, update a CRM record, use a connector, or perform another action.

4. Knowledge

What does the agent actually know about your business?

This is what helps determine whether its output is useful, accurate, and aligned with how your organization operates.

Think through these four components before an agent goes into production.

Start With the End in Mind

You don't need to design every AI agent your company will ever use before you build your first one.

But you should have a sense of where you're going.

Ask:

What problem are we trying to solve?

What manual work are we trying to eliminate?

What impact could this have on our team?

What happens if the agent makes a mistake?

And perhaps most importantly:

Are we automating something that actually needs automating?

The goal shouldn't be to replace every human interaction with an AI agent.

A great sales rep isn't great because they're excellent at filling out administrative forms.

They're great because of their judgment, experience, intuition, product knowledge, and ability to work with people.

If AI can eliminate some of the administrative layer surrounding that work, that's a very different proposition.

You're not necessarily replacing the person.

You're giving them more time to do the work that actually requires a person.

Start With One Simple Agent

If you're just getting started with Agent Hub, don't try to automate your entire business on day one.

Start with something small, well-defined, and relatively easy to evaluate.

For The Gist, one example is the process that happens after an initial prospect meeting.

We call that meeting a "fit check."

By the time the meeting happens, HubSpot already has structured information about the contact, company, and deal.

There's also unstructured information: notes, email conversations, and the meeting transcript or summary.

After the meeting, several manual steps move that information forward.

That's exactly the kind of process where an AI agent could be useful.

An agent could potentially read the meeting transcript, review the existing CRM information, update the deal, create notes, summarize the conversation, and pass the appropriate information to the next person involved.

That's a much better starting point than simply saying:

"Let's build an AI agent."

Instead, you start with a specific business problem and ask whether an agent is the right solution.

The Future of HubSpot Is Increasingly Agentic

It's hard to look at what HubSpot is building and not see where this is heading.

AI-supported agents are likely to become increasingly common throughout the customer lifecycle, from marketing and prospecting to sales, operations, customer success, and service.

That's incredibly exciting.

But the organizations that get the most value from these tools won't necessarily be the ones that build the most agents.

They'll be the ones that build the right agents.

Start with the business process.

Define the outcome.

Understand the inputs.

Write clear instructions.

Give the agent the right tools.

Teach it about your business.

Test it.

Monitor it.

Then iterate.

HubSpot Agent Hub gives businesses impressive power.

The next challenge is learning how to use that power well.