Ask your team to list every AI agent running in your HubSpot portal right now.
Most can't. Not from carelessness; until this week there was no single place to look. That's what Agent Hub fixes, and Agent Builder is where you build the ones you actually need.
GA for Professional and Enterprise. One place to see, activate and manage every agent: HubSpot-built agents (Customer, Prospecting, Data), AEO, Deal Progression, anything from Agent Builder, and the Agent Marketplace. Built for ops and admins rather than engineers. Agents can be called directly from Breeze Assistant, so nobody has to navigate back here to run one.
Before you build anything, run the inventory. Which agents are active, who turned them on, what has each produced in the last 30 days, and what are you paying in credits for agents nobody reviews. An agent with no human owner is the thing that gets blamed when something goes wrong.
Describe what the agent should do in plain language and Breeze configures it, using the customer, deal and call data already in the account. Growth Context loads by default, so the agent knows the business before you write an instruction. You decide which actions run automatically and which need sign-off.
Two updates matter most.
Custom event triggers. A workflow used to start only when something happened to a record inside HubSpot. Now workflows and agents can start from events outside HubSpot (a Slack message, a spreadsheet update) or on a schedule: every Monday at 9am. Scheduled execution is the unlock for recurring operational work. Weekly pipeline inspection, hygiene sweeps, a Monday briefing in the inbox before standup: jobs that used to need a person or a third-party scheduler.
One-click agentic workflow conversion. HubSpot identifies which existing workflows can become agentic and converts them in a click. Everything the workflow does today keeps working; it gains the ability to start from more places and pull in connected information. A portal with four years of workflow sprawl is suddenly the most interesting portal you own.
Tier boundary: Starter can now build and deploy custom agents and simple automation. Professional and Enterprise get the full workflow builder with event-based triggers and advanced automation.
Custom agents run on credits, measured in action units: each unit covers up to 2,000 tokens of input and output, and one action unit is one credit. All custom agents, including Marketplace templates, draw from the same credit pool.
Two controls to use from day one. Simulation mode reads production data and estimates cost without consuming credits; run every agent in it before publishing. Feature-level caps and per-agent monthly run limits stop a runaway agent from becoming a runaway invoice. Usage logs live on the Agent Details page.
Before you convert a single workflow, classify all of them. Pull the active list with last-edited dates. For each: what does it do, who owns it, when did it last fire on a real record. Then three buckets.
Convert: anything that would benefit from an AI action, a scheduled run or an external trigger. The Monday pipeline report someone assembles by hand. Routing logic that needs a judgement call. A handoff that should read the last five emails before it assigns.
Keep: simple, correct, boring. A date stamp on stage change. A notification. Leave it alone.
Retire: nobody can explain it, it duplicates another, or it hasn't fired in six months. Every workflow you keep is one more thing an agent might interact with.
CRM hygiene. Always. An agent that finds incomplete records, duplicates, deprecated properties and inconsistent formatting, generates a cleanup plan, and applies changes only after review. Every agent you build afterwards performs better because of it, and it requires no new purchase and carries no rebuild risk.
After that, the patterns worth productising: recurring pipeline inspection (scheduled deal reviews flagging stalls), pre-meeting briefs (deal history, tickets, sentiment in one summary), support triage, CS risk and renewal monitoring, segmentation from plain-language definitions, workflow and schema administration, lead qualification scoring against ICP criteria before a rep sees the contact.
Build one ad hoc first, on a single record, to prove value in the room. Then embed it in a workflow so it runs for every rep automatically. The number of agents embedded in workflows rather than run manually is the clearest sign a team has moved past the demo stage.
Take the single task the first agent will absorb. Get how often it happens and how long it takes from the people who do it. Multiply. Write it down. That's the baseline you measure against in 60 days, and without it you'll never be able to say whether the agent did anything.
If you've got more workflows than anyone can explain, our AI-ready audit starts with exactly that inventory; details on the pricing page.
Part of our guide to the HubSpot Unbound 2026 release: everything HubSpot announced, on one page.