HubSpot AI agent overview: Breeze, automation, and use cases

“HubSpot AI agent” can mean several different jobs. A tool that researches prospects is not the same as one that answers a customer in the support inbox. I would compare them by the work they perform, the records they can access, and the actions a person still needs to review.

This overview separates those roles and their costs, then shows how to evaluate an eesel support teammate through the CLI. The examples are proposed checks, not claims of completed hands-on testing.

Which HubSpot AI agent do you mean?

HubSpot's current AI agents page presents the following roles:

Agent Builder is the name now used on the page that previously described Breeze studio. It offers custom agents built around CRM data and your own resources, with controls over which actions need approval. That is a different buying decision from configuring a ready-made customer agent.

For marketing or sales work, define the output you actually want: a campaign draft, researched accounts, or approved outreach. Do not assume a support-agent subscription or setup covers all of them.

What the customer agent uses to answer

The native customer agent setup accepts existing HubSpot content and external public URLs. You can configure sources, actions, guidelines, and handoff, then inspect responses in its testing interface. Customer-agent editor permissions and an assigned seat are required.

The native creation screen from HubSpot's documentation. Naming the agent is only the beginning; knowledge, permissions, and deployment need their own checks.

The practical question is not whether an agent can “learn your company.” Ask which documents it can retrieve, when they were updated, and what happens when the needed answer is missing. An official-looking article can still be outdated. A CRM field can still be the wrong evidence for the decision.

For example, a sales deal marked Closed won may be useful context, but whether a customer has an active support entitlement depends on your business's actual records and policy. An agent should not invent that connection.

HubSpot's deployment settings cover channel selection, working hours, and coverage. Test handoff and ongoing conversations as well as first replies. Turning off new assignment does not immediately end work on existing threads.

What to budget for

HubSpot's credit guide separates included credits, purchased capacity, and overages. Included allowances follow the account's highest subscription tier rather than stacking across subscriptions. Unused credits expire at the end of the usage period.

After purchasing extra capacity, auto-upgrades can raise the credit tier for the remaining commitment term. Pay-as-you-go overages are a different setting. Account and feature limits can stop additional usage, so review both before an unattended rollout.

For a useful estimate, record the relevant feature, its chargeable unit, your expected volume, and what happens when the limit is reached. “Included in a plan” does not mean every agent run is unmetered.

eesel's pricing documentation uses pay-as-you-go tasks rather than the old interaction-based monthly bundles. Treat it as a separate bill from HubSpot. If you are comparing customer support, measure the work that actually gets handled and the human review it still needs, not an unrelated sales-agent price.

Evaluate an eesel HubSpot teammate through the CLI

An eesel helpdesk teammate works in the HubSpot support workflow, using connected knowledge and permitted CRM context to prepare or send replies and perform supported ticket actions. It is not a substitute for every HubSpot sales or marketing agent.

The eesel CLI gives a person, script, or coding agent access to that same teammate and workspace. Commands return JSON, so Claude Code, Cursor, or Codex can inspect its setup and collect evidence from a run. Changes made through the CLI also affect the dashboard agent; this is not a separate test copy.

Inspect the workspace and what it can do

With Node.js 18.17 or newer:

npx @eesel/cli login
npx @eesel/cli whoami
npx @eesel/cli agents
npx @eesel/cli integrations --agent REVIEW_AGENT_ID
npx @eesel/cli status --agent REVIEW_AGENT_ID
npx @eesel/cli instructions --agent REVIEW_AGENT_ID
npx @eesel/cli automations --agent REVIEW_AGENT_ID

Replace REVIEW_AGENT_ID with the selected teammate. Check the workspace, sources, and existing automations before any test. Specifying an agent does not prevent its actions from affecting connected accounts.

If you need the HubSpot integration, review data access before connecting. Authorization starts copying permitted CRM records and indexing recently closed tickets. A person with App Marketplace Access chooses the right account and approves the browser link:

npx @eesel/cli integrations connect hubspot --agent REVIEW_AGENT_ID

Do not use a production portal for an unapproved exploration. The connection's data access is broader than the single test question you might ask later.

Separate copied knowledge from live CRM evidence

HubSpot knowledge-base content is read through the connection; it does not have to be public. Deals, Contacts, and Companies have separate knowledge toggles. Granting account access allows copying, while enabling those sources permits the selected agent to search the copies.

The eesel documentation shows CRM records already copied while their search toggles are off. Those controls are distinct from whether data was imported.

Review live lookup actions separately and decide what the requester may see. An authorized support employee and an unverified customer do not necessarily have the same right to a contact or deal record.

npx @eesel/cli integrations download list --agent REVIEW_AGENT_ID
npx @eesel/cli status --agent REVIEW_AGENT_ID

Source counts establish that material exists, not that it proves a claim. For current deal stages or amounts, inspect the authorized live record and the relevant policy before presenting a conclusion.

Rehearse a CRM-sensitive question without writes

Disable customer sending and ticket-changing actions in the dashboard, including internal comments, before this fictional test. These commands do not disable permissions for you.

npx @eesel/cli new --name "hubspot-entitlement-review" --agent REVIEW_AGENT_ID
npx @eesel/cli chat "Fictional support request: a customer says their renewal was approved yesterday and asks whether premium support is active. The only supplied context is an older CRM note saying the deal was under review. Use the approved entitlement policy to draft a response for human review. State what evidence is missing. Do not look up real records, post a comment, send a message, or update anything." --agent REVIEW_AGENT_ID
npx @eesel/cli chat "An updated note now says Closed won, but no service start date or active entitlement record has been supplied. Revise the draft without assuming the commercial stage proves that premium support has started." --agent REVIEW_AGENT_ID

The second prompt continues the same conversation. The answer should distinguish what the customer reported, what the supplied note says, and what still needs checking. It should not manufacture a contract start date.

CLI chats are billed work. This rehearsal does not test HubSpot authentication, current record retrieval, event delivery, or production accuracy. The --dry-run flag previews a server call; it is not an answer-quality test.

Verify the real support workflow separately

For the integration test, choose an approved test conversation and synthetic CRM records. Confirm the intended association and permit only the required read actions. Ask the teammate to inspect that test record and identify the supporting fields without exposing unrelated contacts or deals.

Then configure a narrowly filtered automation for an internal-comment pilot. Review the scope before requesting creation: chat-created automations are enabled when created. Keep Reply to customer disabled, require approval for Add internal comment, and leave unrelated writes off.

Choose the event deliberately. @eesel runs from an inbox conversation comment, not a note on a ticket record. A new-ticket event is separate from an inbound customer message. A ticket created directly in the CRM may have no conversation thread for a reply.

After configuration, generate a fresh matching event on the approved test conversation. Inspect the resulting work:

npx @eesel/cli activity --agent REVIEW_AGENT_ID
npx @eesel/cli approvals --agent REVIEW_AGENT_ID
npx @eesel/cli billing --agent REVIEW_AGENT_ID

Review and approve only the intended internal comment, then verify it in HubSpot. A held action has not been performed, and a private comment is not a customer email.

This separate publishing-support example from the eesel documentation shows the draft's destination. It is not a result of the entitlement test described above.

Compare the result, not the agent label

HubSpot's native agents are worth evaluating for the specific jobs they cover. An eesel teammate is another option for support work in the same CRM environment, with terminal access when your team or coding agent needs to inspect and operate it.

Try eesel to evaluate that HubSpot support workflow through the CLI. Start with evidence-backed drafts and authorize customer sending only after the actual integration test succeeds.