Salesforce Opens Agent Optimizer Beta for Continuous Agent Testing
The Agentforce tool connects production-session analysis with configuration, simulated conversations and reusable regression tests under adjustable human review.
Edited by Tyronne Panaino
Salesforce opened Agent Optimizer in beta on September 21, adding an Agentforce tool that links production-session analysis to agent configuration, simulated testing and staged improvements. Salesforce says access is available through an account contact, so the announcement establishes a beta rather than broad self-service availability.
The practical change is a shorter path from noticing a recurring production failure to testing a proposed fix. Instead of treating observability, configuration and regression testing as separate activities, Salesforce presents Agent Optimizer as one conversational workflow. The intended users are teams operating Agentforce agents that need to understand failures and improve behavior without handing every change directly to an autonomous system.
Business intent becomes testable configuration
In Agentforce Builder, the product starts from a business need expressed in natural language. The Salesforce product article says Agent Optimizer can translate that intent into agent configuration, including the subagents and actions needed for the requested workflow. A user can then refine the behavior through conversation while reviewing the resulting configuration.
Salesforce also says the tool can use patterns from earlier customer conversations with agents or human support representatives. Those transcripts can help identify the requests an agent needs to handle, providing a production-derived starting point instead of a blank specification. That design may reduce discovery work, but the source does not establish how transcript selection, privacy controls or data retention behave in a particular tenant. Administrators still need to verify those boundaries before using customer conversations as development material.
As configuration changes are produced, the beta can simulate conversations and inspect what happens next. When a test fails, Salesforce says the tool can investigate, change the configuration and run the scenario again. The resulting previews and tests can become reusable regression cases, giving teams a way to check whether a later instruction or capability change breaks behavior that had previously passed.
Production sessions feed the next test cycle
The other half of the workflow sits in Agentforce Observability. Salesforce says Agent Optimizer can analyze hundreds or thousands of sessions, identify recurring failure patterns and prioritize improvements against outcomes selected by the organization. The article names tool-call errors, knowledge gaps and off-topic requests as examples of failure groups.
That grouping is meant to turn a large session history into a smaller set of problems a team can inspect. A user can review relevant sessions, examine the evidence behind a recommendation and decide which issue to address. The proposed improvement can then move back into Builder for implementation and testing. Production behavior therefore becomes an input to the next controlled iteration rather than a separate reporting stream.
The announcement describes this as a product capability, not an independently measured result. This run did not test whether the grouping is accurate, whether it catches rare but serious failures or whether the proposed changes improve outcomes without creating new problems. Reusable regression cases can make change review more systematic, but their value still depends on representative scenarios and clear pass criteria.
Human review remains an explicit control
Salesforce describes an autonomy dial that lets teams set review points. At the restrictive end, a person can require sign-off at every step. With more delegation, the tool can investigate issues, suggest fixes, build, test and stage changes for deployment before stopping at chosen checkpoints. Users can watch the work in the browser or return later to review it.
That control is important because the source does not say that Agent Optimizer can safely deploy every proposed change without supervision. Staging a change is not the same as proving it is correct. Teams should keep production deployment behind explicit review until they have tested access boundaries, rollback behavior, audit trails and regression coverage in their own environment.
Availability and evidence limits
Agent Optimizer is available now in beta, with Salesforce directing interested customers to their account contact. The fetched article does not publish general-availability timing, public pricing, detailed plan eligibility or a region-by-region rollout. Those are the next practical checkpoints for buyers assessing whether the feature can enter a standard development process.
The evidence is one first-party Salesforce product article. It supports the beta status and described workflow, but it does not independently verify improvement rates, production safety or the quality of the tool's recommendations. Internal confidence is therefore medium.
Status
Confirmed beta. Salesforce has made Agent Optimizer available to eligible customers through account-led access; operational performance and broader availability remain unverified.
Sources
Update note: Last reviewed 2026-09-27. We will revise this post when Salesforce publishes general-availability details, public eligibility documentation or independently testable operating evidence.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.