Blogs » Give Your Agent Its Own Badge
AI agents need clear identities and limited permissions. Many early deployments rely on a shared service account, API key or other common credential. That can work in a tightly supervised pilot. It becomes risky when several agents begin calling tools, other agents and Salesforce workflows.
If something goes wrong, the organization must be able to tell who initiated the request, which agent acted and what it was allowed to do.
The multi-agent enterprise is already here
Salesforce’s architecture is already designed for networks of agents. MuleSoft Agent Fabric includes a registry to catalog agents and tools, a broker to route work, a gateway to apply policies to agent-to-agent and agent-to-tool traffic, and a visualizer to trace how agents interact.
AIforce makes this more immediate by bringing Salesforce data and actions into interfaces such as Claude and Slack. A request can move through several agents, MCP servers and services before a record changes. At each hop, the system must preserve who initiated the action and give the downstream service only the access it requires. Salesforce’s architects describe trusted agent identity as foundational for this reason.
What Salesforce has documented
Salesforce Headless 360 makes Salesforce capabilities available to external agents and tools through MCP and APIs. Salesforce says those requests are still authenticated and authorized through its Trust Layer, and object- and field-level security continues to determine what data is returned.
Salesforce also describes different identity patterns for different types of work. When a request involves user-specific data, the user’s identity should be preserved across each hop with access scoped to the downstream service. When user context is not needed, a service identity may be appropriate. The point is not that every agent should borrow a person’s login. It is that every interaction should use the right identity and the narrowest permissions required.
In simple terms, Salesforce also plans to measure the work that third-party agents perform through its Hosted MCP Servers. It lists two usage types: Salesforce Record Operations for reading or changing records, and Salesforce Process Invocations for triggering Flow or Apex.
The pricing details are not final. Salesforce says the multipliers are still to be announced and these usage types are not currently metered. It will provide 30 days‘ notice before metering begins.
Why identity matters as agents multiply
An agent network should answer four basic questions: Who initiated the request? Which agent acted? What was it allowed to do? What happened at each step? A reliable identity chain makes those answers visible and allows the right security policy to be applied to each interaction.
With one supervised pilot, a team may be able to follow the activity manually. That breaks down quickly when agents call other agents and tools. A registry and a clear identity chain let the organization scale without losing accountability.
What Salesforce leaders should do now
Four practical steps:
As agents take on more work, identity becomes part of the operating model. Give each interaction the right badge: a clear identity, limited permissions and a traceable path.
IndiQube Fort, 3rd Floor Malviya Nagar, Jaipur India, 302017
AWFIS – 26th Floor, GIFT One Tower, Gujarat International Finance Tec-City, Gandhinagar, Gujarat – 382355