With AI agents multiplying across enterprise IT systems, Salesforce is adding new features to its no-code agent building and deployment platform to help enterprises make and manage them. Agentforce 3 also includes support for agent interoperability frameworks such as MCP and A2A.
This is the fourth version of Salesforce Agentforce since its debut in September last year, with the newest, Agentforce 3, succeeding the previous ‘2dx’ release.
A new feature of the latest version is Agentforce Studio, which is also available as a separate application within Salesforce.
Studio brings together existing tools that were previously located in different parts of Agentforce, said Salesforce AI general manager Adam Evans.
“The launch of Agent Studio also reflects an evolution in what customers need: full agent lifecycle development is no longer just about tools to build agents. Agentic success now spans more stages beyond the initial build and involves more personas involved in the bettering of these agent capabilities. Studio is a major step forward in unifying those needs into one integrated environment,” Evans said.
Dion Hinchcliffe, lead of the CIO practice at The Futurum Group, noted that Salesforce’s move to transform a collection of disparate tools into a cohesive product mirrors how other vendors sometimes had to adopt platform engineering as a mainstream discipline because the tooling inside was made self-service.
This strategy, he said, empowers enterprise customers to scale digital labor beyond IT, enabling automation and intelligence to reach across various business functions.
A command center for agents
Central to enterprises’ ability to manage agents on Salesforce’s platform is the company’s addition of Command Center, an observability layer to monitor and optimize agent performance, inside Agentforce Studio. This is an evolution of existing capabilities, Evans said.
“What’s new is the level of visibility, control, and optimization it offers across the full agent lifecycle. It not only centralizes observability but also includes an embedded optimizer that continuously recommends improvements based on performance, driving more value for enterprises at every stage of agent deployment,” he said.
This could be a turning point for enterprise adoption, analysts suggested.
“Command Center may finally give enterprises the confidence to scale Agentforce into customer-facing environments,” said Valoir Research’s Rebecca Wettemann. The tool is especially relevant for CIOs and CX leaders, she said.
Developers also benefit from Command Center’s improvements to real-time debugging and telemetry, enabling smoother collaboration with non-technical teams, according to The Futurum Group’s Dion Hinchcliffe.
But Salesforce’s definition of observability leans more toward business outcomes than traditional system performance metrics, however, Jason Andersen of Moor Insights & Strategy cautioned.
“The CRM software provider is focusing on the business side of observability (the effectiveness of the agents) versus the more common view: are the agents running in a performant way?” Andersen said.
Salesforce’s Evans said Command Center includes an embedded optimizer that continuously recommends improvements to agents based on performance.
Cost limitations
But there is a limitation for now: It doesn’t optimize for cost, while software from other AI agent developers such as Snowflake and Databricks does.
Cost is a growing concern for enterprises as AI agent adoption increases, Hinchcliffe said, adding that this is a “white space” in the CRM segment and whoever builds real-time agent cost benchmarking first will have an edge.
Other features of the Command Center include the ability to track agent health in real time and to intervene if needed, and to simulate agent behavior at scale with state injection and AI-driven evaluations before deployment.
The agent health monitoring feature, the Command Center and the Agentforce Studio app will be made available in August, Salesforce said.
More Salesforce news:
>
Read More from This Article: Salesforce Agentforce 3 promises new ways to monitor and manage AI agents
Source: News