Appearance
Agentforce Builder
Introducing Pronto
Pronto is a food delivery company on a mission to help restaurants run better and help customers get faster, more reliable deliveries. In Salesforce, that work spans merchants and storefronts, customer identities, orders (in an external system), and support interactions—so a great assistant needs to both understand natural language and execute repeatable business workflows (like verification and refunds) with precision.
Overview
In this chapter, you’ll build a Pronto-focused agent in the next-generation Agentforce Builder using Canvas and Agent Script. You’ll combine LLM flexibility with structured controls (topics, variables, and actions) to create more predictable behavior—this is the foundation of hybrid reasoning.
To achieve this vision, you’ll work through hands-on exercises that bring all of this together:
Build in the new Agentforce Builder: Create an agent in the new Builder, move between Canvas and Agent Script, and learn where to configure system messages, variables, and routing.
Design Topics and Routing: Add topics and wire them into the Topic Selector so the agent can reliably choose the right path based on the user’s request.
Extend with Agent Actions: Connect Apex/Flow/API actions to topics so the agent can search storefronts, look up orders in external systems, retrieve related cases, and perform operational tasks (like handling refunds).
Implement Verification Guardrails (Hybrid Reasoning): Add a dedicated Verify Customers topic, track verification state with variables, and use conditional logic + transitions to ensure sensitive actions only run for verified customers.
Test and Iterate: Use Preview and interaction details to inspect reasoning, validate inputs/outputs, and iterate until routing and actions behave as expected.
Continue with the exercises:
Start with Explore Pronto