Workflow Patterns

Agentforce Vibes has two modes: Chat and Plan Mode. In Chat, the agent figures out the right approach based on your prompt — whether that means answering a question, generating code, debugging an issue, or coordinating parallel work. Use Plan Mode when you want a structured plan to review and approve before the agent executes.

Use Chat for most tasks. Describe what you want and the agent determines how to handle it. Examples:

• Reading unfamiliar Apex code and asking what it does
• Querying org metadata to understand object relationships
• Generating a trigger + handler + test class from a description of the business logic
• Scaffolding an LWC with an Apex controller, event handling, and CSS
• Debugging a failing test class or tracing why a trigger isn’t firing
• Writing a batch Apex class with proper start/execute/finish methods

The agent uses your org context, active Skills, and Rules to produce accurate, org-aware output. Create a Checkpoint before any complex execution so you can roll back if needed.

What to verify after execution:

• Do generated Apex classes follow bulkification patterns?
• Are SOQL (Salesforce Object Query Language) queries selective and outside of loops?
• Do test classes cover positive, negative, and bulk scenarios?
• Are field-level security and object permissions handled correctly?
• Does the code reference real fields and objects from your org (not placeholder names)?

Use Plan Mode when the work is complex enough that you want to see and approve the approach before anything executes. Examples:

• Designing a trigger framework that spans a handler class, test class, and custom metadata configuration
• Scoping an LWC (Lightning Web Component) feature that requires an Apex controller, wire adapters, and Jest tests
• Breaking down a complex integration into sequenced tasks with dependencies
• Any multi-file feature where execution order matters

Plan Mode produces a structured plan with Goals, Non-Goals, Approach, and numbered Tasks. No files change until you click Approve Plan.

What to verify: Review the plan for completeness. Does it account for all the objects, permissions, and test scenarios you need? Are there dependencies it missed?

Some tasks benefit from more careful handling regardless of mode.

Security-sensitive logic that requires line-by-line review:

• Encryption/decryption implementations
• Authentication and authorization flows
• Code that handles PII (Personally Identifiable Information) or financial data
• Sharing rule and permission set configurations that control data access

Use Plan Mode to get a recommended approach, then review every generated change carefully before accepting.

Org-specific business rules the agent can’t infer:

• Complex validation logic based on undocumented business processes
• Integration mappings that depend on external system contracts not visible in your project
• Workflow behavior that exists only in production configuration (not code)

Provide these as explicit context or Rules before prompting. Don’t let the agent guess business logic.

Configuration-heavy work where Setup UI is faster:

• Building permission sets and profiles
• Creating page layouts and record types
• Configuring sharing rules and org-wide defaults

Use Chat to understand what you need, then configure manually in Setup.

• What Is Agentic Development
• Plan Mode
• Permissions and Safety