Workflows have been the backbone of HubSpot for years. If a contact fills in a form, set a property, send an email and create a task. Now there's a new option: AI agents that can interpret, judge and act. With Agent Hub and Agent Builder you can even turn an existing workflow into an agentic workflow. So the question many teams are asking is: when should we use which?
Agent Hub, Agent Builder and agentic workflows
During 2026, HubSpot has brought its AI automation features together in one place:
- Agent Hub (formerly Breeze Agents) gives you a single view of all agents in the portal across marketing, sales and service. Here you can see status and activity and switch agents on or off.
- Agent Builder (formerly Breeze Studio) is where you build your own agents by describing the task, giving instructions and choosing which tools and data the agent can use.
- Agentic workflows are built in Agent Hub's new workflow builder. They can start manually, on a schedule, from a CRM event or from external triggers like Slack and webhooks, and can run agents, custom code and actions in third-party apps in the same flow.
Unlike classic workflows, they don't require an enrolled record, they can pull data from multiple records and external sources, and the trigger can be changed afterwards. The builder is still in beta and requires Professional or Enterprise, so check what applies to your portal.
The rule we use: rules or judgment?
The simplest way to choose is to ask whether the task is governed by clear rules or requires judgment.
Use a classic workflow when
- the logic can be described as if X, do Y
- the result must always be exactly the same, for example lead routing, lifecycle updates and internal notifications
- you're dealing with high volumes where predictability and cost matter most
- compliance or reporting requires every step to be traceable
Use an agent when
- the task requires reading and understanding text, such as an incoming email, a call note or a free-text form answer
- the response needs to be tailored to the recipient, like a personal follow-up email or a summary ahead of a meeting
- someone currently makes a judgment call manually, such as deciding whether a lead is relevant or which rep is the best fit
- the information lives in several sources that need to be weighed together
Combine them when
Most of the value often lies in the combination. Let the workflow handle structure and rules, and let the agent make the judgment call in the step where it's needed. An example: a form is submitted, the workflow creates the contact and sets the basic data, an agent reads the free-text answer and assesses the need, and the workflow then routes the lead and creates a task based on the agent's assessment.
Four pitfalls to avoid
- Replacing working workflows for no reason. A rule-based flow that works shouldn't become an agent just because it can. Agents consume credits and don't always produce exactly the same result.
- Vague instructions. An agent needs a job description. If you can't describe the task in a few clear paragraphs, the process isn't ready for an agent.
- No oversight at the start. Let the agent suggest and a person approve until you've seen that the results hold up. It builds trust in the team and gives you input for refining the instructions.
- Poor context and data. Agents read from Context Home and the CRM. If your ICP, offers and properties are incomplete, so are the decisions.
How to get started
Go through your existing workflows and look for steps where someone currently makes a manual judgment, or where a workflow ends with "create task: check this". That's often where an agent makes the biggest difference. Pick one such flow, build an agent for that specific step in Agent Builder, test it on real records and measure both quality and time saved before you move on.
Want help reviewing your workflows and finding where agents add the most value? Get in touch with us at Structsales and we'll walk through your portal together.