Salesforce Unveils Agentforce 3: AI Agents Get Real-Time Monitoring and Secure Interoperability Boost

On June 23, 2025, Salesforce launched significant enhancements to its AI agent platform with the roll-out of Agentforce 3. This release aims to address key challenges enterprises face when deploying digital workers at scale, particularly the need for real-time visibility into AI agent performance and secure integration across corporate systems. Agentforce 3 introduces a "Command Center" that provides executives with comprehensive, real-time insights into how AI agents are functioning. This feature includes detailed analytics, health monitoring with real-time alerts, and AI-powered recommendations for optimizing agent performance. These enhancements are crucial for enterprises that are moving beyond experimental phases and integrating AI agents deeply into their operations. According to Salesforce, AI agent usage has surged 233% in the past six months, with over 8,000 customers deploying the technology. Early adopters, such as Engine, have seen a 15% reduction in customer case handling time, while 1-800Accountant achieved a 70% autonomous resolution of administrative chat requests during peak tax season. One notable early adopter is PepsiCo, which is leveraging Agentforce 3 as part of its broader AI-driven transformation. Athina Kanioura, PepsiCo’s chief strategy and transformation officer, highlighted the company's longstanding partnership with Salesforce, emphasizing the importance of integrating AI platforms to meet evolving customer demands and integrate data more effectively. PepsiCo sees AI agents as vital for engaging customers and improving backend efficiency. The seven-year relationship with Salesforce has enabled PepsiCo to implement AI agents swiftly, enhancing the experience for field sellers through streamlined workflows and real-time insights. The Command Center is designed to tackle the "day two problems" of enterprise AI, which include operational challenges and the need to monitor and manage AI agents effectively. Jayesh Govindarajan, EVP of Salesforce AI, explained that while creating an AI agent demo is relatively easy, deploying a robust system requires tools to understand when agents are struggling and when to involve human intervention. The Command Center captures all agent activity in Salesforce’s Data Cloud using the OpenTelemetry standard, allowing seamless integration with existing monitoring tools such as Datadog. Another major advancement is Salesforce’s support for the Model Context Protocol (MCP), an emerging open standard for AI agent interoperability. This capability enables Agentforce agents to connect with any MCP-compliant server without the need for custom coding, expanding the integrations available through the AgentExchange marketplace. Gary Lerhaupt, VP of Product Architecture at Salesforce, emphasized the importance of enterprise-grade interoperability, which includes robust governance and control mechanisms. The platform supports integrations with over 30 partners, including Amazon Web Services, Box, Google Cloud, IBM, PayPal, and Stripe, with more MCP servers on the horizon. Underlying these new features is an enhanced "Atlas" architecture, which significantly boosts performance and security. Agentforce 3 now offers 50% lower latency compared to January 2025, along with response streaming for real-time user experiences and automatic failover between AI model providers to ensure continuous operation. For regulated industries, Salesforce hosts Anthropic’s Claude models directly within its infrastructure via Amazon Bedrock, keeping sensitive data within a secure perimeter. The company plans to add Google’s Gemini models later in the year, providing enterprises with more governance options. To accelerate deployment, Salesforce has developed over 200 pre-configured industry actions, with more than 100 added this summer. These actions span various sectors, from healthcare (e.g., patient scheduling) to media (e.g., advertising proposal generation), allowing companies to deploy functional AI agents quickly. Case studies from OpenTable and Grupo Falabella showcase impressive results: OpenTable handles 73% of all restaurant web queries with its agents, and Grupo Falabella saw a 71% reduction in phone call traffic within three weeks of deployment. Salesforce has also introduced more flexible pricing options, including unlimited usage licenses for employee-facing agents and per-action pricing that scales based on the AI work performed, not just conversation volume. The platform is now available globally, with support in Canada, the UK, India, Japan, and Brazil, and includes six additional languages: French, German, Spanish, Italian, Japanese, and Portuguese. Industry insiders laud these advancements as a significant step forward in enterprise AI. Gary Lerhaupt sees this moment as transformative, comparing the shift to previous technological revolutions like the personal computer and the Internet. He believes the "multi-agent revolution" will enable increasingly complex and effective workflows. PepsiCo’s commitment to Agentforce 3 underscores the technology’s readiness for widespread enterprise adoption, as companies recognize the competitive advantages of AI agents that they can monitor, measure, and control effectively. Salesforce’s competitors, such as Microsoft, Google, and Amazon, are also racing to establish AI agent platforms. However, Govindarajan argues that Salesforce stands out due to its comprehensive integration capabilities and the ability to orchestrate AI actions within existing enterprise ecosystems. The general availability of Agentforce 3, along with the upcoming features, marks a pivotal point in the maturation of enterprise AI, signaling a shift towards more reliable and controllable digital workers.