Salesforce launched Agentforce 3, which features Agentforce Command Center, support for Model Context Protocol (MCP), and updated Atlas architecture to speed up reasoning and performance.

The rollout maintains Salesforce's Agentforce cadence, which includes updates every few months as the company learns from enterprise use cases. In addition, Salesforce has added more than 30 partners to AgentExchange including AWS, Box, Cisco, Google Cloud, IBM and payments players such as PayPal and Stripe.

Salesforce said the Agentforce 3 updates address a big blocker to implementations--visibility into what agents are doing. Agentforce 3 adds an observability layer and tools to optimize agents. Agentforce launched late in September 2024 and Agentforce 2 followed in December with developer features in March.

According to Salesforce, 8,000 customers have signed up to deploy Agentforce. The Agentforce 3 release is based on feedback from thousands of Agentforce deployments so far. Recent moves from Salesforce include:

Here's a look at the Agentforce 3 updates:

Agentforce Command Center, an observability console, which features support for MCP and more than 100 prebuilt industry actions. Command Center is built into Agentforce Studio and includes:

  • Optimization tools to tweak agents based on visibility into interactions, usage trends and recommendations.
  • Live analytics on latency, escalation frequency, error rates and unexpected actions.
  • Dashboards on adoption, feedback, success rates, costs and topic performance.
  • Integration with Data Cloud, third party observability tools and Service Cloud, which will get a purpose-built version of Command Center.

A new Atlas architecture that will improve latency, accuracy, resiliency and support for native LLMs from providers such as Anthropic.

Industry actions from partners that include flexible pricing.

Salesforce didn't provide a time frame on when Agentforce 3 will be generally available.

Constellation Research's take

Martin Schneider, analyst at Constellation Research, said:

"The new Agentforce Command Center is a must-have as we continue to develop hybrid human/agent workforces. But will be interesting to see how well it can leverage and measure effectiveness of multi-agent flows that utilize agents from other platforms. Salesforce has made all the right partner announcements around helping their customers manage a multi-platform AI strategy but has not explicitly stated how users can access and leverage other product's agents. Perhaps we will hear more on that during Dreamforce, but just like with humans - digital agents need to leverage data and functions from multiple systems, not just the CRM to do their jobs. This must be addressed sooner rather than later. 

It is also good to see more tools for evaluating the effectiveness of Agentforce agents - while it is easy for almost anyone to build and deploy an agent, measuring the efficacy and value these agents are providing is important. Many customers will need to show value now that the pricing and total cost of using Agentforce is becoming more clear. So, if the agents are not pulling their weight, they may be hard to justify especially as a lot of organizations are not ready to downsize the human labor element just yet."

Holger Mueller, an analyst at Constellation Research, added:

"Salesforce keeps moving on Agentforce with never seen before speed, releasing Agentforce 3 quickly after Agentforce tdx. And the lead of Salesforce in agentic AI shows by the vendor tackling V3 challenges - better agent contol and agent testing as well - a first for all vendors - vertical agent automation options. As always - what is new, true agentic era AI innovation, and what is "AI washing" will have to be unpacked in the weeks and months to come."

Business Research Themes

Data to Decisions, Future of Work, Next-Generation Customer Experience

Roles

Chief Information Officer

Tags

salesforce, AI agents