The evaluation and monitoring stages of agentic AI may be the only ones left since businesses want to make the agents they're starting to use more visible.

Even though AI agent benchmarks might be deceptive, it is still very useful to determine if the agent is functioning as intended. In order to do this, businesses are starting to provide platforms that let users test the performance of AI agents or sandbox them.

On Wednesday, Salesforce launched a restricted test of Agentforce Testing Center, their agent assessment platform. December is when general availability is anticipated. Businesses may watch and test AI agents in the Testing Center to make sure they have access to the data and workflows they require.

AI-generated tests for Agentforce, Sandboxes for Agentforce and Data Cloud, and Agentforce monitoring and observability are among the new features offered by the Testing Center. AI-generated tests enable businesses to create "hundreds of synthetic interactions" using AI models to see if agents respond to questions in the manner that businesses desire. As the name implies, sandboxes provide a controlled setting for testing agents while simulating a company's data to better represent how the agent would function for them.

The Testing Center belongs to a new class of agents that Salesforce calls Agent Lifecycle Management, according to Patrick Stokes, executive vice president of product and industry marketing at Salesforce, who spoke with VentureBeat.

According to Stokes, they are setting up what they believe will be a significant new agent subcategory. The term "lifecycle" refers to the entire process, from conception to development to deployment, and then to subsequent iterations of your deployment. Additionally, the Testing Center still lacks workflow-specific insights that would allow developers to view the precise API, data, or model selections made by the agents. On the other hand, Salesforce gathers this type of information through its Einstein Trust Layer.

Additionally, he stated that they are creating developer tools to make that metadata available to their clients so they may utilize it to improve the development of their agents.