The work Agentforce agents perform for Salesforce users is no longer a black box.
Salesforce on Tuesday released Agentforce 3, a bundle of features that create, test and manage the AI agents. Agentforce Studio, an agent testing and design environment, and Agentforce Command Center are the most significant pieces, but the capabilities won't be available until August.
Command Center features an agent observability dashboard that can show user-defined performance metric measurements such as agent latency, error rates and escalation rates. It can also break down agent performance to individual customer interactions, according to Salesforce.
MuleSoft support for Model Context Protocol (MCP) and Agent2Agent (A2A) agentic AI interoperability standards is slated for release in July. Native support for MCP on the Salesforce platform is also planned for July, with native A2A support coming later.
Salesforce and other vendors such as ServiceNow are working on agent interoperability to create environments where one agent can manage others, in theory, to bring together automations across platforms.
First, though, tools like Command Center are needed to give users control over what agents do, said Rebecca Wettemann, founder of independent research firm Valoir.
You've got to be able to understand, monitor and manage these guys before you're going to let them loose on customers, let alone on other agents.
Rebecca WettemannFounder, Valoir
"You've got to be able to understand, monitor and manage these guys before you're going to let them loose on customers, let alone on other agents," Wettemann said.
In an Agentforce 3 demo, Salesforce showed how the Command Center can zoom in on unsuccessful agent interactions with customers, see what went wrong and offer plain-language paths to optimize agent behavior for future customer service interactions.
"Agentic AI is no longer a science project," said Adam Evans, executive vice president and general manager of AI at Salesforce.
Tire manufacturer Goodyear is an early adopter of Agentforce. The company is tailoring its Agentforce agent to reinforce relationships with car manufacturers and resellers and build new relationships with consumers shopping at its Goodyear tire shops, said Mamatha Chamarthi, chief digital officer at Goodyear.
When a consumer, for example, is looking to replace tires, the Goodyear agent can take into consideration customer data, weather data from their ZIP code, and other points to make personalized recommendations -- a new feature for Goodyear.
"The tire industry is a low-margin business and it's crowded, so we need to create a differentiator," Chamarthi said. "[We are] moving from a transactional relationship with the customer to one that's focused on lifetime value, being very proactive and making it convenient."
MCP here, A2A not far behind
MCP servers, which expose an application's resources -- such as data -- to AI agents, are the heart of the interoperability protocol. AWS, Box, Google Cloud, PayPal and about two dozen other companies have already uploaded MCP server connections to AgentExchange, Salesforce's agent developer marketplace.
Salesforce tooling will support governance that ensures both MCP and A2A-equipped agents make data available for only the people and use cases for which they're allowed, said Gary Lerhaupt, vice president of product architecture at Salesforce. He added that users will be able to put business policies and security in place to manage these coming cross-platform agents.
"The builders are going to be able to connect and orchestrate [with governance and security] all sorts of new types of dynamic, multi-agent experiences," Lerhaupt said.
While interoperability might be a long-term goal, Salesforce currently blocks other AI companies from searching Slack messaging data, according to an update to its terms of service last month. Wettemann said that might not be the best long-term strategy for getting developers to use Salesforce Agentforce tools.
"A lot of this is going to be determined not just by the individual vendors' stuff, but by the ecosystems they're able to build -- which means [they] have to have the capabilities for people to interoperate," Wettemann said. "[They] have to bring more people on the boat."
Don Fluckinger is a senior news writer for Informa TechTarget. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.
