Salesforce Agentforce 3: Unleashing Operational Insights for Your AI Agents

What is truly happening within AI agents? Salesforce has made an effort to clarify this by launching Agentforce 3, with visibility being the key focus.

Since its debut in October 2024, Agentforce has already demonstrated its value: Engine reduced customer query processing time by 15%, while 1-800Accountant confidently delegated **70%** of routine chat inquiries to AI. These statistics are remarkable, but the latest version goes beyond mere scaling; it emphasizes understanding.

The standout feature of Agentforce 3 is the Command Center. This is more than just a dashboard; it’s a mission of enlightenment:

It allows users to observe the **patterns** in agents’ operations—how they tackle tasks and where they encounter difficulties.

**Real-time metrics:** latency, the frequency of escalation of complex questions to humans, and errors—all available at a glance.

**Works/Doesn’t work:** a clear distinction between tasks that agents handle with ease and those that urgently require a «firmware update.»

For anyone puzzled after the rollout of AI tools with the question “So what?”, this transparency is a refreshing breath of air. The system records **all** agent activity following the standard [OpenTelemetry](https://opentelemetry.io/). This means that data can be seamlessly integrated into familiar monitoring tools like Datadog or Splunk—your IT gurus already know how to handle this.

The excitement around AI agents is palpable; fresh data from the Slack Workflow Index reveals an explosive 233% growth in just six months. During the same period, around 8,000 organizations have begun to implement Agentforce.

Ryan Tipples, CTO at 1-800Accountant, affirms the platform’s capability: *»Agentforce autonomously handled 70% of our admin chats—an incredible help during peak times. But this was just the beginning. Now, with enhanced visibility, we see what’s effective, optimize on the fly, and scale our support confidently.»*

This isn’t just data collection—Agentforce 3 possesses self-improvement capabilities: the AI analyzes dialogues, identifies recurring patterns, and actively suggests ways to enhance them. It might sound a bit meta, but for teams with no time to sift through thousands of chat logs, this could be a lifesaver.

Another critical issue is integration. An agent is only as effective as the systems it can connect to. However, securely linking a bot to all necessary business tools can be quite a headache.

Agentforce 3 introduces native support for the Model Context Protocol (MCP). Salesforce aptly compares it to “USB-C for AI.” The concept is straightforward: agents can connect to any MCP-compatible server without custom code while strictly adhering to your security policies.

This is where MuleSoft (acquired by Salesforce previously) comes into play, transforming your APIs and integrations into ready-made «building blocks» for agents. Meanwhile, **Heroku** manages the deployment and maintenance of custom MCP servers.

Molly Bodensteiner, SVP of operations at Engine, emphasizes the importance of openness: *»Salesforce’s open approach, especially the native support for standards like MCP, is critically important for confidently scaling agents. We can securely connect them to our enterprise systems without custom code and compromises in management. This interoperability provides the flexibility to accelerate implementation with full control.»*

Arguably, the most intriguing aspect of this announcement is not what Salesforce has created, but the ecosystem they are nurturing. Over 30 partners have already developed MCP servers for integration with Agentforce, including AWS, Google Cloud, Box, PayPal, and Stripe. For example, integration with **AWS** enables agents to analyze documents, extract text from images, transcribe audio, and even identify key points in videos. **Google Cloud** grants access to Maps, databases, and their leading models like **Veo** and **Imagen**.

A particular area of potential lies in **healthcare**. Tyler Bauer, VP of outpatient operations at UChicago Medicine, explains the need: *»AI tools in healthcare must adapt to the complex and individual needs of patients and providers. Our goal is to automate routine interactions at the patient access center (scheduling, general inquiries) to free up team time for addressing sensitive and complex tasks.»*

**The crucial question:** will this genuinely help businesses manage the growing army of AI agents? Many companies have been operating blindly: they knew the percentage of processed requests but did not understand **where exactly** improvements could be made for the agent.

Adam Evans, EVP & GM of Salesforce AI, confidently states: *»Agentforce 3 will redefine the interaction between people and AI agents, delivering breakthroughs in productivity, efficiency, and business transformation.»*

Interested in testing powerful neural networks for your tasks? Check out **BotHub**—a platform where you can access models like ChatGPT o3, DeepSeek R1–0528, Midjourney 7, and Google Veo 3. [Sign up for a bonus](https://bothub.chat/?invitedBy=m_aGCkuyTgqllHCK0dUc7).