HubSpot agents are AI tools configured to carry out particular jobs using instructions, relevant information and permitted actions. Depending on the agent, that might mean answering a customer, researching a prospective buyer or helping maintain CRM information.
For a business owner, the starting question is practical: which job needs help, and how will your team know the work has been done properly?
An agent needs a defined purpose and boundaries. Your business still decides what information it can use, what actions it may take and when someone should review or take over.
These tools serve different purposes:
HubSpot’s tool-selection guide explains when each approach fits.
For example, an enquiry with a known postcode may only need a routing rule. A customer asking an open-ended question about your services needs interpretation and a reliable information source. Start with the problem before choosing the tool.
HubSpot’s current guide lists these default agents. Availability and configuration differ; this is an orientation, not a promise that every agent is included in your subscription.
| Agent | Main job |
|---|---|
| Customer Agent | Handle customer support conversations. |
| Prospecting Agent | Research target businesses and prepare relevant outreach. |
| Data Agent | Support CRM enrichment and data quality. |
| Campaign Agent | Help plan and generate marketing campaigns. |
| Knowledge Base Agent — beta | Identify gaps in help content. |
| Revenue Agent — beta | Support invoice follow-up. |
Beta labels reflect the documentation checked on 2 November 2026. Check current availability before building a business process around a feature.
The Customer Agent can use supplied content to answer questions in supported customer channels, clarify a request or hand it to your team.
Imagine a customer asking what to prepare before an onboarding appointment. An agent could explain the steps from your current guide. An unusual situation or request for a person needs a clear human follow-up route. That is an illustrative use case, not a claimed customer result.
The business question to test is whether people receive accurate answers and useful follow-up. A count of automated replies alone will not tell you that.
The Prospecting Agent can look for relevant companies and contacts, use buying signals and prepare personalised outreach according to configured plays.
For example, a sales team could define a target segment and the signals it considers relevant, then assess the quality of the suggested contacts and messages. Before enabling automatic activity, decide what needs review and what the agent may do without it. Do not assume that every generated message waits for approval.
More outreach is not automatically better prospecting. Agree what a suitable prospect looks like, why the message is useful and who owns replies.
For data work, decide which missing or inconsistent information matters to a real decision. Establish how a reviewer will recognise an incorrect result before allowing important records to change.
For marketing work, agree the audience, offer, source material and approval process first. Generated campaign material still needs factual and brand review. An agent does not supply evidence for an unsupported customer claim.
Agent Builder provides a way to configure agents for more specific tasks. A sensible brief might be: review selected deal notes and produce a risk summary for the account owner to assess.
Keep the first version narrow. Specify its inputs, expected output, allowed actions and reviewer. Adding the ability to change records or contact customers is a separate decision from producing an internal recommendation.
HubSpot’s AI adoption guide describes account settings, permissions and data preparation. Requirements vary by feature. For example, current Customer Agent guidance lists eligible Professional and Enterprise subscriptions; Prospecting Agent guidance includes eligible Starter subscriptions as well. Seats, permissions, credits and individual feature limitations still need checking.
Ask your account owner to confirm:
Our HubSpot AI costs checklist covers the commercial questions. If the underlying records are unreliable, start with the CRM cleanup checklist. Team ownership also matters: see getting your team to use HubSpot.
Choose a repeated task your team understands. Record what happens today, what a good result looks like and what would make you stop the trial. Review accuracy, completed work and the time people spend checking or correcting it, alongside credit usage.
An agent is worth considering when it helps your team deliver a better result within clear boundaries. The right first step may be a small agent pilot, a simpler workflow or better source information.
Which job would you like help with? Talk it through with Foundry 11. We can help assess the process, account requirements and practical checks before recommending a rollout.