As AI Agent Usage Surges, Companies Need a Complete Observability Solution

Our forward-looking statement applies to this blog.

Back in March, we posed a simple question: How do you know if your agent is production ready? The post resonated with our Agentblazer community as practitioners sought to bring more predictability to agentic responses and better understand why agents behave the way they do. Now, less than three months later, we’re posing a decidedly bigger question: How do you manage a burgeoning digital labor force at scale?

If the delta between these two conversations creates a bit of whiplash, you’re probably not alone. It’s hard to overstate just how quickly the agentic lifecycle has evolved. Simply put, AI agents are no longer experimental. In the last 6 months, AI agent usage is up 233%, with 96% of employees reporting that AI helps them complete tasks they couldn’t before, according to new Slack research. But as a new hybrid workforce begins to emerge, most agent platforms still lack the necessary tooling, governance and observability to scale beyond POCs.

After thousands of Agentforce implementations — from Goodyear to Finnair — it’s clear that customers need better visibility, more granular metrics and actionable recommendations to improve the accuracy, quality, cost and latency of their agents. To equip our customers with the tools to tackle this challenge, we’re announcing Agentforce Command Center, a first-of-its-kind observability tool designed to measure AI agent activity, manage the partnership between humans and agents, and drive continuous improvement.

Flying blind

Many startups today offer agents that are easy to spin up, but difficult to optimize and scale. As we’re learning together with our customers, building a great agent that can reliably and efficiently deliver results at 95% accuracy takes time and iteration. But iterating can be hard when you don’t know where to start. Many customers understand that their agents need work, but struggle to define where to focus their efforts. In essence, they’re flying blind. “Is my agent delivering bad results because of a configuration mistake, or is there something wrong with the data I’m feeding it?” 

Agentforce provides a range of tools to address this challenge. Customers can troubleshoot individual utterances and observe how an agent identifies topics and executes actions using Plan Tracer. They can test agent responses at scale with Testing Center. These tools provide powerful insights, but as organizations go from making sure their agents are production ready to managing an entire workforce of agents working alongside humans, the need for a unified observability solution spanning the breadth of the hybrid work force becomes increasingly acute.       

Managing your hybrid workforce

Agentforce Command Center is the single source of truth for your digital labor force, enabling complete visibility across all of your production AI agents. Command Center rolls up all your agent activity, metrics and telemetry into a single unified dash, delivering a new observability layer for monitoring agent health, measuring outcomes and optimizing collaboration between humans and AI.

Going beyond basic metrics like total number of sessions, Command Center provides deep insight into agent performance, with detailed analytics for error rates, escalation frequency, agent latency and much more. Users can tailor their view to their unique use cases, surfacing only the metrics that are most helpful and relevant. 

Command Center is designed to answer your most burning Agentforce questions, such as:

• “How are my agents performing?”  
• “Are my agents following legal and regulatory requirements?”
• “How is adoption and usage trending?”
• “What are agents costing us over time?”
• “How are agents impacting the customer experience?”

It also gives managers the ability to set real-time alerts in case an agent isn’t behaving as expected. But scaling and optimizing your Agentforce means you need to connect broad trends to actions for improvement. That’s why Command Center provides visibility into what’s happening across your org and the ability to drill down into specific conversations — tracing every turn of an agentic interaction to see what went right, what went wrong, and what we can do to improve outcomes. 

In addition to topics, which are configured at runtime to describe the job an agent should be doing, Command Center also leverages new AI-derived tags that categorize actual conversations agents are having.  Below you can see the total number of conversations related to order management, and the latency and quality across all of these agent interactions.

Command Center not only helps surface shortfalls and identify trends, but provides recommended improvements, such as updating a topic description or drafting a new knowledge article to close a gap. It also allows you to inspect the session trace to see conversation-level details such as which topics and actions are triggered, latency and quality score.

These tools and several other exciting innovations are part of Agentforce 3, the newest version of Agentforce, delivering complete visibility and open integration to scale hybrid workforces. Learn more here.