Salesforce unveils observability tools to manage and optimize AI agents

Salesforce today unveiled new Agentforce 360 observability tools to give teams visibility into why AI agents behave the way they do, and which reasoning paths they follow to reach decisions.

Salesforce is providing the new tools as its agentic AI customers increasingly shift focus from building agents to maintaining them in production.

“We’ve had thousands of customers use the platform,” says Madhav Thattai, SVP and COO of Agentforce at Salesforce. “In the first 12 months, the focus has been, ‘How do we build good agents? How do we connect the right data? How do we control the agent behavior?’ It’s really been about build and design of the agents.”

But Thattai adds that an internal Salesforce survey now shows a three-fold increase in implementations that customers are now moving to production.

“Thousands of customers are now live with agents in production,” he says. “As customers move up the maturity curve, the problems shift from how to build a good agent to how to manage an agent at scale. That’s really what the observability tool stack is about.”

The new tools, he says, help customers understand whether agents deliver value, how they perform in interactions, and how to improve their performance.

Triple threat

The tools span three areas: Agent Analytics, Agent Optimization, and Agent Health Monitoring.

Agent Analytics is intended to help organizations continuously refine their AI agents by providing visibility into agent performance. Teams can use Agent Analytics to track agent usage and effectiveness metrics to understand how agents perform in real customer interactions. The tool surfaces KPI trends over time and highlights ineffective topics, actions, or flows.

Agent Optimization traces session flows to reveal how agents make decisions and help teams diagnose issues that arise. Teams can use the tool to see how agents respond, step-by-step, across complex reasoning chains. The tool groups similar requests into clusters to uncover patterns, friction points and quality trends. It scores agent responses using intent, topic, and quality metrics, and teams can also use the tool to identify configuration issues that affect performance and pinpoint the need for tuning, retraining, or guardrails.

Finally, Agent Health Monitoring focuses on uptime, reliability, and responsiveness, and tracks key health metrics in near real-time. It also provides alerts on critical errors, latency spikes, and escalations so teams can detect, investigate, and resolve problems while minimizing downtime.

Case in point

1-800Accountant, a virtual accounting firm for small businesses, has been a beta customer of the new observability tools. Because of the nature of its business, the small firm is busiest February through April. Ryan Teeples, the company’s chief strategy officer, explains that automation and agentic AI have been essential to help it service customers in that period. It already has more than 20 agents in production. These agents work together to address some of 1-800Accountant’s biggest pain points.

For instance, when one of the company’s salespeople meets with a new lead for the first time, they take the lead through an assessment to understand specific needs. That process is supported by a slate of agents that records and transcribes the conversation, analyzes it, summarizes it, and then dynamically generates an agenda, based on data from the conversation, for a potential follow-up onboarding appointment if the lead opts to become a client.

“Observability allows us to have much faster throughput,” Teeples says, adding that the tools allow his team to iterate and improve agents at a much faster pace. “Now you have tools that allow you to monitor the interactions, get feedback, and get better prompting and better data to the agent. It accelerates the process of the agent and gives us more time on the development side to focus on better client experience, rather than on monitoring and ensuring the AI agents aren’t going off the rails.”

Agent Analytics and Agent Optimization are both part of Agentforce Studio and are available today; regional rollout for APAC customers will happen on Friday. Agent Health Monitoring will be generally available in spring 2026.