Salesforce Unveils Agentforce 3 to Enhance AI Agent Visibility

Salesforce has introduced Agentforce 3, an innovation aimed at resolving a common business challenge: gaining visibility into the actions of their AI agents.

Since its launch in October 2024, Agentforce has proven beneficial across various industries. For instance, Engine successfully reduced customer service handling times by 15%, while 1-800Accountant efficiently redirected 70% of their administrative inquiries to AI during the hectic tax period.

However, the most compelling aspect of this upgrade is not just in the figures. Salesforce is addressing a significant issue: companies are rapidly deploying AI agents without fully understanding their operations or how to enhance them.

The core feature of Agentforce 3 is the Command Center—essentially a dashboard for monitoring AI 'employees'. It allows managers to examine performance trends, observe real-time health metrics (such as latency, escalation rates, and errors), and determine which areas are functioning well and which require adjustments.

For those who have launched AI capabilities and wondered, "What next?", this level of insight could be revolutionary. The system tracks all agent activity using the OpenTelemetry standard, ensuring compatibility with popular IT monitoring tools like Datadog and Splunk.

The adoption of AI is rapidly increasing. Upcoming data from the Slack Workflow Index indicates a 233% hike in AI agent utilization over six months, with approximately 8,000 organizations enrolling to use Agentforce during this period.

Ryan Teeples, CTO at 1-800Accountant, remarked: “During the peak of the most recent tax season, Agentforce autonomously resolved 70% of our administrative chat interactions—an impressive feat amidst our busiest time. However, that initial success is just the beginning.

“We’ve laid a solid deployment infrastructure and are focused on weekly advancements to roll out new AI-driven experiences and automations using Agentforce’s latest features. With enhanced visibility, we're able to see the effective elements, make real-time optimizations, and confidently scale our support.”

Salesforce Agentforce 3 transcends mere data provision; it suggests improvements actively. The AI self-monitors, recognizing conversation patterns and proposing modifications. It's a self-reflective yet potentially beneficial tool for teams stretched thin, with little time to manually audit numerous bot interactions.

Salesforce also addresses connectivity challenges. AI agents’ usefulness is contingent upon system access, but securely linking them to business tools can be difficult for many organizations.

Agentforce 3 delivers built-in support for Model Context Protocol (MCP), described by Salesforce as “USB-C for AI.” This means AI agents can seamlessly connect to any MCP-compliant server without custom coding while maintaining security protocols.

This is where MuleSoft, acquired by Salesforce a few years ago, becomes relevant by converting APIs and integrations into assets ready for agents. Heroku then handles the deployment and upkeep of customized MCP servers.

Mollie Bodensteiner, SVP of Operations at Engine, noted: “Salesforce’s open ecosystem strategy, particularly through its native support for open standards like MCP, is crucial for scaling our use of AI agents with complete confidence.

“We can securely link agents to our essential enterprise systems without custom coding or compromising governance. This interoperability gives us the agility to expedite adoption while maintaining complete control over agent operations within our infrastructure.”

Arguably, the most intriguing part of this announcement is not what Salesforce has created internally, but the growing ecosystem they are fostering. Over 30 partners, including AWS, Google Cloud, Box, PayPal, and Stripe, have developed MCP servers that integrate with Agentforce.

These integrations extend beyond basic data access. For example, AWS integration enables agents to analyze documents, extract information from images, transcribe audio, and even identify key video segments. Google Cloud connections integrate with Maps, databases, and AI models like Veo and Imagen.

The healthcare sector appears especially promising.

Tyler Bauer, VP for System Ambulatory Operations at UChicago Medicine, shares: “AI tools in healthcare need to adapt to the complex and unique requirements of both patients and care teams.

“Our goal is to support this by automating routine interactions in our patient access center that involve common inquiries and requests, freeing our team to focus on more sensitive or complicated needs.”

The pivotal question remains whether these developments will truly aid businesses in managing the increasing number of AI agents. Visibility into AI performance has been a challenge for many—companies usually know the general percentage of queries handled by AI but find it difficult to pinpoint specific weaknesses or opportunities for improvement.

Adam Evans, EVP & GM of Salesforce AI, asserts: “Agentforce 3 will revolutionize the collaboration between humans and AI agents, propelling productivity, efficiency, and business transformation to new heights.”

Whether this promise is fulfilled remains to be observed, but addressing the visibility and control issues is undoubtedly a positive move for businesses grappling with effective AI management.