Salesforce Launches New Agentforce Observability Tools to Solve the AI Black Box

Salesforce is introducing new tools in its Agentforce 360 platform to provide greater transparency, giving enterprises clearer insight into how AI agents behave. The goal: crack open the proverbial black box that has slowed AI adoption and undermined trust. As organizations push to build truly agentic enterprises, the ability to observe and understand these bots has become essential.

“As AI adoption accelerates, the biggest enterprise challenge will no longer be about building an organization’s first agent, it will become how to best manage a fleet of agents that are making real-world business decisions,” Adam Evans, Salesforce’s executive vice president and general manager, remarks in a statement. “You can’t scale what you can’t see.”

New Agentforce Observability Tools

Nine new features are being announced, each covering one of three key areas: analytics, optimization, and health monitoring.

Agent Analytics

This set of tools promises to give organizations a detailed view of how an agent is performing, translating actions into data, trends, and insights.

• Track Agent Performance: Monitor usage and effectiveness metrics for all active agents, giving teams insight into how effective bots perform in real customer scenarios.
• KPI Trends: Tracks KPI trends over time, helping teams identify where performance is improving, declining, or requires their attention.
• Actionable Insights: Surfaces ineffective topics, actions, or flows, and proactively informs teams of steps needed to optimize agent performance.

Agent Optimization

The following three capabilities should give organizations visibility into how an agent interacts with others. They’re intended to surface performance gaps, trace session flows, and provide clarity as to why an agent behaved a certain way.

• Observe Every Interaction: Provides complete visibility of every Agentforce interaction to understand the exact step-by-step response taken by an agent.
• Cluster and Analyze Sessions: Compile similar requests to identify patterns, friction points, and quality trends, and evaluate agent responses based on intent, topic, and quality metrics.
• Optimize Agent Configuration: Identify configuration issues affecting an agent’s performance and determine where fine-tuning, retraining, or additional guardrails are needed.

Agent Health Monitoring

The last group aims to ensure continuous agent uptime, reliability, and responsiveness. These features provide real-time tracking of all bots, along with actionable trust signals. Salesforce describes these three as essential for ensuring agents are at their best even during peak loads.

• Monitor Agent Status Continuously: This continuously monitors critical health metrics in “near real-time,” ensuring the information in the Agentforce dashboard is accurate, and proactively warns of any issues that may arise.
• Resolve Failures Proactively: This sounds the alarm on critical errors, latency spikes, and escalations to agents, giving teams time to mitigate issues that could lead to downtime.
• Maintain Enterprise Reliability: A tool to ensure the system stays reliable even as organizations add more agents. It will continuously monitor all agents easily, giving organizations confidence that the platform scales safely and all operations are running smoothly.

“We’re giving IT leaders the critical tools to continuously track performance, debug issues, and prove the ROI of their AI investments, ensuring every intelligent agent performs reliably, securely, and with total transparency,” Evans states.

From Pilot to Program

These nine features aren’t the only observability features available on Agentforce. The platform also offers capabilities such as deep session-level tracing, per-interaction quality scores, and optimization. And not all these features are home-grown either—Salesforce will soon incorporate analytic tools from Spindle AI, a startup it recently says it’s acquiring. However, the ones announced today are intended to bring deeper visibility, monitoring, and optimization to all agents than before.

To make this possible, Salesforce is leveraging two foundational building blocks: its new Session Tracing Data Model and MuleSoft Agent Fabric. The former catalogs every interaction, including user inputs, agent responses, reasoning steps, LLM calls, and guardrail checks, storing them within Data 360. By doing so, organizations have a standard approach to visibility, enabling them to understand precisely how agents are behaving and responding. The latter was introduced in September and transforms unmanaged agents into a secure and intelligent network. In other words, this capability creates a single place to register, orchestrate, govern, and observe an agent regardless of where it was built.

The push for enhanced observability comes as Salesforce seeks to capitalize on the growing number of organizations looking to transform their business using AI. Back when Agentforce was unveiled, it was an era of experimentation and pilot programs. However, in the aftermath of the debated MIT study, Salesforce has shifted its approach from cheerleading the agentic era to providing better guidance and tools to help its customers see value from the technology.

In its latest CIO study, Salesforce reports a 282 percent surge in AI implementation over the past year. Company leaders are focused on building agentic workflows, better integrating AI into their core platforms, and scaling AI safely across the entire business. And amid concerns about data and security, CIOs are sticking with the tried-and-trusted platforms and systems for AI investments. So, to keep these organizations as customers and to prove Agentforce’s worth, Salesforce is debuting more features so IT leaders aren’t scratching their heads about why something happened.

After all, it’s not a scenario teams would let get away with human workers, so why should things be any different in a human-agent environment?

So, when will these capabilities be generally available?

Salesforce says deep observability in Agentforce Studio, including Agent Analytics and Agent Optimization, is available today, though it’s slightly different for those in EMEA (November 20) and APAC (November 21). As for Agent Health Monitoring, that won’t be publicly available until Spring 2026.

Featured Image: Credit: Salesforce