š Summary: Agentforce Best Practices
This document provides best practices for implementing Agentforce, Salesforceās conversational AI platform. It covers how Agentforce functions, the role of filters and instructions, and guidelines for defining topics and actions.
š How Agentforce Works
Instructions: Interpreted by the LLM (Large Language Model) and guide the agentās behavior.
Filters: Deterministically modify the decision space, allowing precise control over what the agent can and cannot access.
š§ Best Practices
š§© Topic Classification
Include varied customer phrasings and common keywords.
Cover the entire range of questions related to a topic.
Write in natural language.
Make each topic description distinct and non-overlapping.
š Topic Scope
Write as if explaining a job: āYour job is to⦠You cannotā¦ā
Be explicit about permissions and boundaries.
āļø Topic Instructions
Guide action selection.
Set conversation patterns.
Provide business context.
Use clear definitions and consistent terminology.
Avoid absolutes unless necessary.
Pair specific actions with clear situations.
āļø Action Instructions
1ā3 sentence summary of the actionās purpose.
Specify:
When to use the action.
What input is required.
Edge cases.
Example customer phrases that should trigger it.
š” Action Input Instructions
Describe what input is needed and why.
Define expected format, including:
Length
Character rules
Validation and formatting criteria
š£ Final Note
Developers are encouraged to join the Salesforce Developer Community for support and updates:
š salesforce.com/devcommunity