Build, Extend, and Troubleshoot Your Agent with AI Assistance (Beta)
Focus on what you want your agent to do, not how to configure or fix it. The Agentforce assistant is a built-in coding agent that extends, refines, and troubleshoots your agents directly in Agentforce Builder, at no additional cost. Explore use cases and considerations for working with the assistant.
Required Editions
| Available in: Lightning Experience |
| Available in: Enterprise, Performance, Unlimited, and Developer Editions. Required add-on licenses vary by agent type. |
Agentforce Development Agent is a beta service that is subject to the Beta Services Terms at Agreements - Salesforce.com or a written Unified Pilot Agreement if executed by Customer. Use of this beta service is at the Customer's sole discretion.
If you prefer the previous experience, you can opt out of the beta. In Setup, from the Einstein Setup page, turn on Disable Agentforce Development Agent (Beta).
| Use Cases | How Agentforce Can Help |
|---|---|
| Add new capabilities to your agent | You don't have to be an Agent Script developer to build out an agent. Describe what you want in plain language. For example, ask the assistant to add a subagent that looks up order details and requires user verification. Agentforce handles the rest. The assistant searches for relevant actions in your org and asset library, creates any subagents, and updates the Agent Router. If it can't find an existing action that fits your use case, it creates a placeholder with the metadata your agent needs and flags it for you to finish. When it's done, it explains what it plans to change and why, and you can accept or decline before anything is applied. |
| Refine or restrict your agent's behavior | Whether you're fine-tuning after an initial rollout or enforcing compliance guardrails, ask the assistant to update routing conditions and access rules without rebuilding Agent Script from scratch. For example, restrict a refund subagent to customers with a "delivered" order status, or limit escalation to customers above a certain entitlement tier. The assistant proposes targeted adjustments and surfaces any downstream issues that the change might introduce. |
| Troubleshoot unexpected agent behavior | When your agent behaves unexpectedly in testing or production, describe what it's doing instead of what you want, or ask the assistant to explain a specific error. You can also use it to fix Agent Script validation issues flagged on save or in the console. If you’re testing your agent’s behavior in a preview conversation, click Ask Agentforce in the Interaction Summary panel to add the conversation to the assistant’s context, and then send a message. The assistant surfaces likely root causes, such as missing instructions, ambiguous subagent routing, or empty variables. It proposes targeted fixes for you to review, so you don’t have to manually review your agent’s full script. |
| Validate and optimize your agent after upgrading from the legacy builder | When you migrate or upgrade an existing agent, you can use Agentforce or third-party AI coding tools to do most of the work. But metadata and syntax differences between the legacy and new builder can still cause initial validation errors. Use the Agentforce assistant to resolve those errors without memorizing Agent Script syntax. You can also ask the assistant to optimize your agent's script to improve quality and follow Agent Script best practices, so you get the most from your new agent. It reviews your natural language instructions, suggests improvements, and adds deterministic controls to improve reliability and performance. |
Considerations for Using the Agentforce Assistant
- The Agentforce assistant is included in all SKUs, licenses, and editions that grant access to Agentforce Builder at no additional cost. The assistant doesn’t incur billing or consume credits.
- The Agentforce assistant reasons in increments of five minutes. If your task is more complex, the assistant pauses and gives a summary of what it’s completed so far and the remaining steps. The assistant retains the context of what it was working on, so you can ask it to resume and it picks up where it left off.
- When you refresh Agentforce Builder during a conversation with the Agentforce assistant, a new session starts and context resets. Minimizing the assistant doesn’t affect an ongoing conversation.
- The Agentforce assistant uses the same reasoning model that you’ve specified for Agentforce agents overall. See Select Agentforce Model Option to learn more about available models for Agentforce agents (Salesforce default, AWS-hosted, and Gemini), and how to change your model selection.
