Agents That Talk to Each Other
On June 23, 2025, Salesforce unveiled Agentforce 3 -- and buried within the feature list was something potentially more significant than any individual capability: native support for Model Context Protocol MCP, the open standard for AI agent interoperability originally developed by Anthropic.
This matters because it addresses one of the biggest unresolved problems in enterprise AI: agents from different vendors cannot communicate with each other. A Salesforce agent, a ServiceNow agent, and a custom-built agent have operated as isolated islands, unable to share context, coordinate workflows, or hand off tasks intelligently.
MCP changes that equation.
What Model Context Protocol Enables
MCP provides a standardized way for AI agents to:
- Share context: An agent handling a customer service issue can pass the full conversation context to a different agent managing the backend fulfillment -- regardless of which platform each runs on - Coordinate multi-step workflows: Complex business processes that span multiple systems can be orchestrated by agents that understand each other's capabilities and limitations - Discover capabilities: Agents can query other agents about what they can do, enabling dynamic workflow assembly rather than hard-coded integrations - Maintain security boundaries: MCP includes authentication and authorization mechanisms, ensuring that inter-agent communication respects enterprise access controls
The New Agentforce Command Center
Agentforce 3 also introduced a centralized Command Center that provides:
- Real-time visibility into agent performance, decision patterns, and escalation rates across all deployed agents - Policy management tools for defining guardrails, escalation rules, and autonomous action boundaries - A/B testing capabilities for comparing agent behaviors and optimizing performance - Compliance dashboards tracking regulatory adherence for AI-driven decisions
Why Open Standards Win
Salesforce's adoption of MCP is a strategic concession: no single vendor will own the agentic AI ecosystem. The enterprise world is too complex, too multi-vendor, and too regulated for any walled garden to prevail.
This mirrors the history of the internet itself. Proprietary networks gave way to TCP/IP. Proprietary messaging gave way to email. The vendors that embraced open standards thrived; those that resisted were marginalized.
For enterprises, the message is clear: invest in AI architectures built on open standards. Agents that can only operate within a single vendor's ecosystem will become liabilities as the interoperability era matures.
Implications for Custom AI Agent Development
The rise of MCP and open agent standards creates a powerful opportunity for custom AI agent development:
- Custom agents can now participate as first-class citizens alongside vendor-native agents - Specialized industry logic can be encapsulated in purpose-built agents that integrate seamlessly with platform a