AI agents for HubSpot and Salesforce: built-in agents, and where custom fits
By the Codivated team · Updated
HubSpot and Salesforce both sell AI agents now: HubSpot's Breeze agents and Salesforce's Agentforce. Use them first when the job lives inside the CRM and fits how the vendor built it. Add a custom AI agent when the work spans the CRM and other systems, like your inbox, your product, your billing or your ERP, or when it has to follow your own qualification rules and approvals before it touches a record.
What AI agents do HubSpot and Salesforce include?
| HubSpot | Salesforce | |
|---|---|---|
| Name | Breeze agents (in HubSpot's Agent Hub) | Agentforce |
| Common agents | Customer agent for support chat and email, prospecting agent for outreach, data and deal agents | Service agents, sales development and coaching agents, and agents you build in Agentforce's builder |
| Build your own | Agent builder inside HubSpot | Agentforce builder, with actions from Flows, Apex and APIs |
| How it's priced | Credits per conversation or action, on top of your plan | Per conversation, consumption credits or per-user licenses |
| Strongest when | Your data and team live in HubSpot | You already run complex processes in Salesforce |
Both vendors change their agent lineups and prices often. Check the current pages before you budget, and run a test on your own data before you trust any resolution-rate claim.
When is the built-in agent enough?
When the whole job happens inside the CRM and follows the vendor's way of working: answering support questions from your knowledge base, drafting first-touch emails to contacts already in the CRM, cleaning up duplicate records, summarizing a deal before a call. Turn those on, set them up well and measure them before you build anything.
When do you need a custom agent on top of your CRM?
| Job | Why the built-in agent falls short | What a custom agent adds |
|---|---|---|
| Lead follow-up by SMS and email | Your qualification rules and handoff points are your own | Works the conversation, qualifies by your rules, hands hot leads to a rep with context |
| Quotes and renewals | Pricing lives in a spreadsheet, CPQ or billing system | Pulls usage and contract data, drafts the quote, routes it for approval |
| Account health and churn risk | Signals live in your product, support desk and billing | Combines them, flags the accounts and drafts the next step for the owner |
| CRM hygiene from email and calls | The facts are in inboxes and call notes | Reads them, proposes field updates, a person approves |
| Order handoff to operations | The order has to land in the ERP, like NetSuite | Turns a closed deal into a clean order and checks it before it goes |
How does a custom agent connect to HubSpot or Salesforce?
Through the same APIs your integrations use today: HubSpot's CRM APIs with a private app, or Salesforce's REST APIs with a connected app. Both also offer connections for AI assistants built on the Model Context Protocol (MCP). The technical part is the easy part. What matters:
- Its own user and scopes. The agent gets a dedicated integration user with only the objects and fields its job needs. Never a system administrator login.
- Approvals before writes. At first, every record it would create or change goes to a person. See human in the loop.
- No mass updates without a check. Bulk changes go through a preview and a person, and the agent can't delete.
- A log of every action. What it read, what it changed and why, kept outside the CRM's own history.
Our team has connected agents to Salesforce and to custom CRMs in production, including lead-nurture agents that worked about 80,000 leads a month for 12 months.
Should you buy the CRM's agent or build your own?
| Choose the built-in agent when | Choose a custom agent when |
|---|---|
| The job lives inside one CRM | The job spans the CRM and other systems |
| The vendor's workflow matches yours | Your rules, tone or qualification are a real advantage |
| You want it live this month with little setup | You need control over approvals, logging and model choice |
| Per-conversation pricing works at your volume | Volume is high enough that credits get expensive |
Many companies do both: the CRM's agent for routine support and data cleanup, and a custom agent for the jobs that make money. See build with n8n or a custom agent for the same choice on the automation side.
What should you measure?
Pick the one number the job exists to move: time to first reply on new leads, the share of leads qualified, quotes sent per rep, hours of manual CRM updates. Measure it for two weeks before the agent and every week after. See how to measure AI ROI.
Common questions
Do we need Salesforce Data Cloud or a HubSpot upgrade first?
Not for a custom agent, which works through the standard APIs. Some of the vendors' own agents do need specific editions or add-ons, so check before you buy.
Can an AI agent update our CRM on its own?
It can, but it shouldn't at first. Start with a person approving every change, then let routine updates through below clear limits once they've been approved unchanged for a few hundred cases.
Will it work with a custom or older CRM?
Yes, if the CRM has an API or a database we can reach safely. The rules about scopes, approvals and logs are the same.
Is our CRM data used to train the AI model?
Not with business API plans from the major model providers, which don't train on your data by default. Check the terms of any product you buy.
Sources
- Agentforce, Salesforce: Salesforce's AI agent platform and pricing models.
- Spring 2026 Spotlight, HubSpot: the current Breeze agent lineup.
- Model Context Protocol: the open standard many CRMs and assistants use to connect.