Meet the Reasoning Engine
Learning Objectives
After completing this unit, you’ll be able to:
- Define what the reasoning engine is.
- Explain how the reasoning engine works.
Putting It All Together
Topics and actions are the building blocks of an agent, and they’re pretty powerful, right? But how does an agent know when to launch these topics during conversations? Let’s meet the reasoning engine behind Agentforce.
What’s a Reasoning Engine?
Think of the reasoning engine as the conductor of an orchestra. Just like a conductor keeps time and guides musicians, the reasoning engine orchestrates how topics and actions handle a user’s request.
Here’s the breakdown.
- Interpret the user’s request. When a user starts a conversation, the reasoning engine interprets the request and classifies it into a specific topic. For example, if you ask, “When will my order be delivered?,” the engine classifies it as the Order Management topic.
- Build a plan for accomplishing the goal. The reasoning engine then processes each step to accomplish the goal. It considers the context, previous interactions, and available data when breaking the problem down.
- Use data to change behavior and respond to the user. When the agent finds and reviews the data from topics and actions, the results can provide a wealth of additional information. This empowers the agent to respond dynamically and make better decisions. The insights the agent gains can show potential pathways and significantly influence the next steps of the plan.
Remember that scenario at the beginning of the module with Alex and Agentforce? Let’s analyze it and see the reasoning engine in action.
Customer (Alex): “Hi, I just placed an order for a pair of custom sneakers. Can you tell me when it will be delivered?”
Agentforce: “Hello, Alex! Your order will be shipped within the next 2 to 3 business days and should arrive at your address within 5 to 7 business days after shipping. Is there anything else you’d like to know about your order?”
- Interpret the user’s request: The reasoning engine classifies this as the Order Management topic.
- Build a plan for accomplishing the goal: It queries the order database to confirm the shipping timeline and provides a clear response.
- Find and launch the right topics and actions: The agent retrieves the shipping information and provides a detailed response.
Alex: “Thanks! Do you have any stores in New York City?”
Agentforce: “We don’t have any physical stores in New York City, but we have a popup shop opening next month at 123 Fashion Avenue from October 1 to October 15. You can also shop online for the latest styles.”
- Interpret the user’s request: The reasoning engine classifies this as the Store Locations custom topic.
- Build a plan for accomplishing the goal: It takes the next step of searching the company’s store locations and upcoming events to solve requests.
- Find and launch the right topics and actions: Given the lack of stores, the agent provides alternative information about the popup shop and online shopping options.
Alex: “Awesome, thanks! One more thing—do you have any recommendations for a good running trail in New York City?”
Agentforce: "Certainly! For a great running trail, I recommend Central Park’s Reservoir Loop, which is 1.58 miles and offers beautiful views. For a longer run, try the Hudson River Greenway, which stretches for about 13 miles. Enjoy your run!"
- Interpret the user’s request: The reasoning engine classifies this as the Local Recommendations custom topic.
- Build a plan for accomplishing the goal: It accesses a database of local recommendations and selects the most relevant options.
- Finds and launches the right topics and actions: The agent provides detailed recommendations for running trails in New York City.
Alex: “That’s perfect, thanks so much for all the information!”
Agentforce: “You’re welcome, Alex! If you have any more questions, feel free to reach out. Have a great day!”
In this scenario, the reasoning engine handles the initial request about order delivery then seamlessly transitions to other topics like store locations and local recommendations. The reasoning engine can tackle topic changes and complex situations, ensuring a smooth and satisfying experience for the customer.
Time for Action
Linda’s feeling more confident now that she knows how Agentforce works, and she’s ready to take it for a spin. We suggest you do the same! Here are some suggested Trailhead badges to continue learning about Agentforce.
- Agentforce Builder Basics
- Quick Start: Build Your First Agent with Agentforce
- Connect Data Cloud to Agentforce and Prompt Builder
Now that you have a solid grasp of both standard and custom topics and actions and the reasoning engine, you’re ready to plan, build, and customize your own agents with Agentforce. Explore our resources to learn more about Agentforce and get hands on with Trailhead badges.