MuleSoft Agent Fabric: Multi-Agent Orchestration & Governance

Overview of MuleSoft Agent Fabric

MuleSoft Agent Fabric serves as a centralized solution for discovering, orchestrating, governing, and observing AI agents across heterogeneous environments. By addressing "agent sprawl"—the fragmentation that occurs when agents operate in silos—Agent Fabric provides enterprise architects with a unified view of the entire AI ecosystem, regardless of the underlying LLM or deployment platform.

New Capabilities in the Agent Fabric Expansion

Announced at TrailblazerDX, the latest updates focus on enhancing trust, operational control, and connectivity for AI-driven workflows.

Discovery and Asset Management

• Agent Scanners: Automates the discovery and cataloging of existing AI tools and agents within your network, including those deployed on third-party platforms like Amazon Bedrock or Azure AI Foundry.
• Visual Authoring Canvas: A drag-and-drop interface for mapping complex workflows. It enables developers to integrate human-in-the-loop checkpoints, ensuring critical business decisions remain supervised.

Connectivity and Integration

• MCP Bridge: Enables existing APIs to become agent-ready by implementing the Model Context Protocol (MCP) without requiring extensive code refactoring. This allows for standardized security and rate-limiting at the gateway level.
• Informatica-Hosted MCPs: Provides direct access to Informatica’s data quality and governance servers, which are integrated directly into the Agent Registry.

Governance and Security

• Trusted Agent Identity: Implements role-based access control (RBAC) for AI agents. High-risk actions—such as financial transactions—now require mobile-based authorization from a human manager, effectively preventing "rogue" agent behavior.
• AI Gateway: Standardizes token management, compliance, and observability across your multi-LLM stack, providing guardrails against cost overruns and data policy violations.

Multi-Agent Orchestration

Salesforce is doubling down on deterministic orchestration. By leveraging Agentforce Script for Agent Broker, developers can now define fixed, deterministic hand-off rules between agents, while allowing LLMs to manage the reasoning logic between these nodes. This hybrid approach ensures reliability while maintaining the flexibility of generative AI.

Deployment and Availability

• Regional Expansion: Available in Canada and Japan.
• Runtime Fabric: Supports deployment on private cloud infrastructure to ensure data residency and security for sensitive workloads.
• Roadmap: The beta for deterministic orchestration in Agent Broker starts in April 2026, with general availability for the visual authoring canvas and Salesforce model support arriving in June 2026.

Key Takeaways

• Unified Governance: Use the Agent Registry and AI Gateway to control token usage and security across multi-vendor LLMs.
• Deterministic Control: Leverage Agent Broker to insert hard-coded logic into generative workflows, reducing the risks associated with autonomous agent behavior.
• Simplified Integration: Utilize the MCP Bridge to make legacy APIs compatible with AI agents without manual code modifications.
• Human Oversight: Implement mobile authorization workflows for high-stakes agent decisions to ensure enterprise-grade security.