Until now an AI agent working in Salesforce borrowed the login of the person it was helping. 🔐 Salesforce Agentic Identity changes that and gives every agent its own badge, its own permissions and its own paper trail.
For admins this is the security upgrade they quietly wanted. For finance teams it also brings a new line in the budget, so it pays to understand both sides before your first external agent goes live.
What is Salesforce Agentic Identity
Salesforce Agentic Identity is a registration system that gives each AI agent connecting to Salesforce its own OAuth credentials and its own scoped permission set, instead of acting under the login of the employee it serves. An agent gets issued its own badge rather than borrowing an employee's badge, and every action it takes is tied back to that badge alone.
This matters now because the quiet workaround is already everywhere. Claude, ChatGPT and custom built agents already read and write CRM data through MCP and direct API calls, almost always under a human user's credentials, which means an audit log cannot actually tell you whether a given record change came from the person or from the tool acting on their behalf.
What changes for admins
- Every external agent gets registered and scoped the way a new employee would be, under the principle of least privilege rather than inheriting a person's full access
- Every agent action becomes fully traceable, and access can be revoked for one agent in a single click without touching the human user it used to borrow credentials from
- MCP clients will eventually move to the OAuth credentials of the registered agent identity instead of a shared user login
How Flex Credits metering will work
Every successful call a registered agent makes to Salesforce, whether it arrives through MCP or a direct API call, will count as a Headless Platform Interaction and draw down Flex Credits, tracked through Digital Wallet. Traditional integrations keep today's pricing, and sandboxes, scratch orgs and Developer Edition orgs are not metered under this model, so building and testing an agent costs nothing extra before it goes live.
Salesforce has not yet published the Flex Credit multiplier for a Headless Platform Interaction, so the per call cost cannot be calculated yet. Salesforce has committed to 30 days of notice before metering actually begins, which gives every org a real window to measure expected agent volume before a single credit is charged.
A five point readiness checklist
Agents with their own identity are easier to trust, easier to audit and much easier to explain to a security team asking hard questions about what is touching customer data.
Agentic Identity builds directly on the access review we covered in AIforce and the permission cleanup it just made urgent. For a structured way to do that review, see our Salesforce permission set audit guide.
