HubSpot Work AI agents are automations you build inside a HubSpot Work project to handle recurring work on a trigger or a schedule. You can describe the agent's job in plain language, and the AI assistant builds it. Each agent reads the project's tables and your CRM data, then creates tasks, updates rows, flags delays, and notifies owners. Only its AI steps use HubSpot Credits.
HubSpot now offers agents in more than one place, and the product names don't make the lines between them obvious. Picking the wrong place leads to duplicate automations and credit bills nobody planned for. If you're mapping out your first agents, our HubSpot AI agent implementation team can help you choose the right jobs and guardrails before launch.
HubSpot Work holds the work, and AI agents do the repeatable parts of it. A workspace groups projects, and each project holds its own tables, documents, forms, and AI agents. The agents act on the data in those tables plus your CRM records.
HubSpot Work gives every project two kinds of AI help.
The assistant is something you talk to, and an agent is something you can set up once and leave running. Both depend on the same context. Professional and Enterprise customers using AI with high-quality context create 3.6x more MQLs and win 3.2x more deals. HubSpot compares them with customers on the same tiers that don't use AI. Treat those numbers as vendor benchmarks, but the practical point still holds that an agent reading clean records earns your team's trust faster than one reading duplicates.
An AI agent in a HubSpot Work project can review rows, compare them with CRM data, and act on what it finds. It runs the moment a record appears, on a set schedule, or both. Its actions include creating subitems with due dates, updating a status, adding notes, creating follow-up tasks, and notifying owners.
Here's an example onboarding agent prompt that tells the agent what to do in a client onboarding project.
Agents can also draft handoff documents from deal history. They can route requests based on CRM properties, like sending a deal desk request to the right owner.
We recommend that agents do one job. An onboarding agent that also chases invoices and writes recaps is harder to test, and harder to trust when something goes wrong. Build each agent prompt around a trigger, context, actions, escalation, and limits, and keep it to a single goal.
HubSpot Work agents live inside one project and act on that project's tables. Agent Hub agents are account-wide tools you build, publish, and reuse across HubSpot. Agent Hub is the place to activate, configure, and monitor pre-built agents, custom agents, and agentic workflows across your account.
|
HubSpot Work Agent |
Agent Hub Custom Agent |
|
|
Where you build it |
The prompt bar inside a project |
The agent builder in Agents > Agent Hub |
|
How you set it up |
Describe the job in plain language, then refine with more prompts |
Set instructions, actions, knowledge, and inputs in the builder |
|
What it works on |
One project's tables, plus CRM data |
Records and processes across the account, plus knowledge vaults |
|
How it runs |
New rows and schedules defined in its prompt |
Agentic workflows, other automations, or on demand |
|
Credit use |
AI workflow actions only |
Thinking, actions, and generated content in each run |
|
Plans |
Starter, Professional, or Enterprise during the beta |
Starter and up, with Professional or Enterprise for agentic workflows and some HubSpot-built agents |
Custom agents in Agent Hub use credits when they think, take actions, or generate content, and you can run tests without credits before you publish. Breeze Assistant is a third way in. You tell it what you need, and it assigns the right agents to get it done.
Use a HubSpot Work agent when the job starts and ends inside one project, like chasing onboarding milestones. Use Agent Hub when the same logic should run across many records or teams, or when an agent talks to customers. Keep one shared list of every agent, its owner, and its trigger, because the two don't share a build screen.
You can add an AI agent by describing it in the project's prompt bar, either from a template or with your own prompt. Before you start, a Super Admin needs to enroll the account in the public beta and turn on generative AI. Each user also needs a Core, Sales, or Service seat.
Follow these steps.
Start with a template for your first agent, then rewrite its limits in your own words. Templates give you a working structure, but only your team knows the records an agent should never touch.
An agent is also only as reliable as the table it reads, so set clear column types and dropdowns before you add one.
HubSpot Work agents use HubSpot Credits only for AI workflow actions, marked with an icon inside the agent. Rule-based steps, like updating a table value or sending an email notification, don't use credits.
Take an onboarding status agent that runs every weekday morning. The AI steps that assess ownership and spot at-risk onboardings use credits on every run. The status updates and owner emails that follow cost nothing.
You can cap spending with account-level credit spending limits, which HubSpot turned on by default in October 2026. The default limit sits up to 50% above your monthly credits, and you can adjust it at any time. Super Admins and Billing Admins get usage alerts at 75%, 85%, and 90% of the limit. You can also set limits for single features, swap seats for credits mid-contract, and buy credit packs at a flat monthly rate.
Our advice is to keep AI actions for judgment calls and run everything else as rules. Schedule frequency counts too, because every run repeats the agent's AI steps. A weekday morning schedule runs 5 times a week, so check that one daily review covers the job before you add more runs.
The best first agents handle internal follow-up work, where a missed step costs time but never reaches a customer directly. These four are good starting points.
A closed deal creates a row, and the agent builds the standard steps, watches due dates, and flags stalled clients. Client onboarding is one of the pre-defined workspace names in HubSpot Work setup, so the structure is quick to start.
Reps submit discount or contract requests through a form, and the agent assigns each request based on deal properties. It works well because the routing logic already lives in your CRM.
An agent can check a launch timeline each morning and notify the owner of any task due within two days. Landing pages, emails, and decks stop slipping without anyone noticing.
An agent can watch a content table and nudge reviewers when a draft sits in review too long. Pair it with AI-drafted HubSpot Work documents, and briefs stay right next to the tracker.
Skip agents that message customers, edit client-submitted data, or run on a messy CRM. HubSpot data shared at UNBOUND 2026 shows that AI running on poor or missing context performs worse than no AI at all. Clean up duplicates and fill in owner fields before the first agent goes live. Customer conversations belong with HubSpot Customer Agent, and that agent has its own knowledge sources and handoff rules.
HubSpot Work agents are part of a public beta, so expect gaps and frequent changes. Plan around these limits.
Start with one process, run it for a few weeks, and expand once the agent's output matches how your team works.
Agent problems usually start before the agent exists, with unclear ownership, messy CRM fields, and no rule for when a human steps in. Our team scopes each agent's job, cleans up the data it reads, and writes its guardrails and handoff rules. We use the same six-step framework from our HubSpot AI agent deployments. It covers discovery, planning, preparation, configuration, testing, and scaling.
We've been a HubSpot Elite Solutions Partner since 2022, so we know how these pieces fit together across Hubs. If you're planning your first agents, talk with our HubSpot AI agent implementation team.
HubSpot Work AI agents are project-scoped automations that read your tables and CRM data, then handle recurring follow-up work on a trigger or schedule. Agent Hub covers account-wide and customer-facing agents, and Breeze Assistant assigns agents when you ask for an outcome. Start with one internal process, keep AI steps for judgment calls, and set a credit limit before you expand.
Check how ready your portal is with the free HubSpot AI Readiness Scorecard.