How to Set Up HubSpot Breeze AI Agents | Markivis

Key Takeaways

• Breeze agents are HubSpot's AI agents that act on your CRM data to answer visitors, qualify leads, and draft content

• An agent is only as good as the content and data behind it, so cleanup comes before configuration

• Give each agent one narrow job and clear guardrails, then expand once it earns trust

• Keep a human reviewing answers and handoffs while the agent is new

• Measure accuracy and deflection, not just how many conversations the agent handled

What Are Breeze AI Agents and Why They Matter

Breeze is HubSpot's set of AI agents that work directly on your CRM. Instead of a chatbot bolted on the side, a Breeze agent draws on the content and records you already hold to answer visitors, qualify leads, and draft content, around the clock. The appeal is an agent that handles routine conversations and admin without adding headcount.

The risk is just as real. Point an agent at thin content or messy data and it will answer confidently and wrongly, in front of your prospects. That is why a good setup starts with the data and guardrails, not the configuration screen. Get those right and Breeze becomes a quiet, reliable extension of your team. Get them wrong and it becomes a liability talking to your buyers.

How to Set Up Breeze AI Agents: Step-by-Step

Breeze is evolving quickly, so confirm the exact menu names and availability against your own portal and tier before you build.

Step 1: Confirm Access and Plan

• Log into your HubSpot account

• Check that Breeze agents are included in your subscription, as access and limits vary by tier

• Identify the specific agent you want first (customer, prospecting, or content), rather than switching on everything

• Confirm in your portal which agents your plan supports before building around them

Step 2: Prepare the Knowledge Source

An agent answers from your content, so this step decides whether it helps or invents. In HubSpot, tidy the sources the agent will draw on:

• Review your knowledge base articles for accuracy and gaps

• Update key site pages the agent may reference

• Make sure the records the agent uses are clean and current

• Remove or fix outdated content that could produce wrong answers

Step 3: Create and Scope the Agent

• Open the Breeze or AI agents area in your settings
• Create a new agent and give it a single, well-defined job
• Set its tone and the audience it serves
• Resist adding extra responsibilities; one clear job keeps its behavior predictable

Step 4: Set Behavior and Guardrails

Define what the agent can and can't do before it ever goes live:

• Which topics it is allowed to answer
• When it must hand off to a human
• What it must never claim or promise (pricing, commitments, legal points)
• The escalation rules that bring a real person in at the right moment

Step 5: Test, Then Publish Narrow

• Run the agent against real past questions and edge cases in preview
• Fix anything it gets wrong before customers see it
• Publish to a single page or limited audience first
• Widen the rollout only once it is proven on real traffic

Best Practices for Breeze Agents

• Start With Clean Data: An agent grounded in good content answers well; one grounded in gaps invents. Clean the source before you configure anything.
• Give Each Agent One Job: Scope tightly to a single task so behavior is predictable and easy to judge. Expand only as the agent earns trust.
• Keep a Human in the Loop: Review answers and handoffs while the agent is new. Loosen supervision once accuracy is proven, not before.
• Test Against Real Questions: Use your actual past conversations and hard edge cases, not invented ones, so you see how it handles reality.
• Measure Accuracy, Not Volume: A busy agent giving wrong answers is worse than no agent. Sample responses for correctness regularly.

How to Use Breeze Agents in Your Workflow

• Let a customer agent deflect routine support questions so your team handles the complex ones
• Use a prospecting agent to qualify visitors and route the ready ones to sales with context attached
• Have a content agent draft first versions of emails and pages for your team to edit
• Set the agent to hand off to a human the moment a conversation needs one, with full history

Troubleshooting Common Breeze Issues

Problem: The agent gives confident, wrong answers

Solution: Its knowledge source is thin or outdated. Improve the content it draws on and tighten what it is allowed to answer.

Problem: It tries to handle questions it shouldn't

Solution: Your guardrails are too loose. Narrow the topics it can answer and set clearer escalation rules to a human.

Problem: Prospects get frustrated and drop off

Solution: The agent isn't handing off soon enough. Lower the threshold for escalation so a person steps in faster on hard conversations.

The Bottom Line

Breeze agents are powerful because they act on your real CRM data, and risky for exactly the same reason. Start narrow, with one job, clean source content, clear guardrails, and a human watching. Set an agent up to be accurate before you set it up to be busy, and it becomes a dependable part of your team. This careful, data-first approach is how we turned marketing into a scalable revenue engine for Maple Assist, building automation on a clean foundation so it could be trusted to run. See the Maple Assist case study.

The Markivis Approach

We treat AI agents as a data problem before we treat them as a configuration task:

• Audit the knowledge source first: We review and clean the content an agent will draw on before touching a setting, so it answers from something trustworthy.
• Scope one job per agent: We launch each agent with a single, narrow responsibility, so its behavior stays predictable and easy to judge.
• Keep a human in the loop: We review agent responses and handoffs while it's new, loosening supervision only once accuracy is proven.
• Measure accuracy, not activity: We track how often the agent is right, not just how many conversations it handled, so success means something real.

This careful, data-first approach is how we turned marketing into a scalable revenue engine for Maple Assist, building automation on a clean foundation so it could be trusted to run. See the Maple Assist case study.