Salesforce has officially unveiled Agentforce 3, a major enhancement to its flagship enterprise AI solution, introducing innovative methods for monitoring and managing AI agents across its platform.
The Agentforce platform empowers organizations to design, customize, and implement generative AI agents capable of autonomously enhancing workforce productivity. These goal-oriented software components can execute tasks with minimal or no human intervention. Sales, service, marketing, and business professionals can now tailor AI "workers" that leverage business logic and prebuilt automation to act on their behalf.
Today's release introduces a new command center offering full visibility and native Model Context Protocol (MCP) support for seamless plug-and-play compatibility across agents and services. The company also announced over 100 new pre-built industry actions to accelerate standardized AI agent deployment across sectors.
Centralized Command for AI Agents
As AI agents perform tasks behind the scenes and collaborate with human workers, the demand for enhanced observability has intensified. Salesforce addresses potential model safety and performance issues through its observability platform, but the upgraded Agentforce Command Center takes this further by unifying agent health, performance metrics, and optimization outcomes.
Integrated into Agentforce Studio - the AI agent customization platform - the command center enables teams to analyze every agent interaction, investigate specific events, and identify trends. It also delivers AI-driven recommendations for improving targeted conversation types, continuously refining Agentforce agent performance.
The command center serves as a unified interface for understanding AI agents through context-specific dashboards. For example, product delivery agent dashboards will display cancellation rates, shipment status, and other operational metrics. Online advertising agents will showcase click-through rates and campaign performance indicators through their visualizations.
Users can now employ natural language to generate topics, instructions, and test cases directly within Studio. The Test Center simulates agent behavior using data state injection and AI-powered evaluation, allowing pre-deployment stress testing of agents before going live.
Inter-Agent Connectivity via MCP Integration
Agentforce 3 now enables AI agents to natively connect with external services, tools, and agents using an open standard called Model Context Protocol. Developed by Anthropic PBC, this protocol functions like "USB-C for AI" by enabling plug-and-play connections without custom code while maintaining existing security policies.
Salesforce's integration platform MuleSoft can convert any existing API into an MCP connector equipped with security policies, activity tracking, and traffic controls. This will facilitate orchestration and management of multi-agent protocols.
For instance, users can rapidly build an AI agent that sends and receives PayPal invoices using natural language instructions through built-in MCP integrations. With this protocol, agents can be created, tested, and deployed directly within the Agentforce builder interface without requiring custom code development.
Customers will gain connectivity to numerous third-party tools including Amazon Web Services, PayPal, Box, Cisco Systems, Google Cloud, IBM, Notion Labs, Stripe, Teradata, and Writer through this integration framework.
