Salesforce Agentforce Masterclass: Hands-on Guide

Explore the basics of ai, including natural language processing, large language models, and machine learning, and learn how prompts steer non-deterministic ai from predictive to generative to agentic applications.

Explore predictive AI, generative AI, and agentic AI through practical, hands-on demonstrations in the Salesforce Agentforce masterclass, linking concepts to real-world agent workflows.

Master the art of prompting by learning the structure, elements, and best practices to craft prompts for LLMs, including context, participants, grounding, constraints, and tone.

Explore zero-shot, few-shot, and chain-of-thought prompting techniques to enhance AI reasoning and decision making. Apply these methods to streamline sales workflows in a Salesforce context.

Explore prompt templates as reusable, scalable prompts with placeholders, and learn to build and test them in Salesforce using the prompt builder, data sources, and flows for b2c marketing communications.

Understand that AgentForce is a low-code platform to build, customize, and deploy agentic AI on the Salesforce platform. Serve both employees and customers with enterprise-grade, task-performing AI agents.

Explore how the agent force and data cloud, via data 360, unify structured and unstructured data to create a data-aware, context-aware AI agent with secure, governance-driven architecture.

Compare legacy agent force builder and the new agent script. Note legacy limitations like restricted context variable access, and the agent script's hybrid reasoning, variables, and visual or code-based options.

Discover the core building blocks of Salesforce agents - sub-agent, actions, data, channels, reasoning engine, and llm - and how to configure and deploy them with guardrails across channels.

Create and configure a Salesforce developer org to enable agent force, including signing up, resetting passwords, enabling data cloud and Einstein, and choosing between legacy agent builder or agent script.

Dive into the sub-agent building block, its name, instructions, reasoning, and actions, and learn how the agent router routes to the right sub-agent via llm-based or deterministic routing.

Explore how sub-agents drive conversations via the start agent and agent router, guided by the reasoning engine, deterministic logic, and hybrid reasoning to route, execute actions, and ground responses.

Explore actions as building blocks that let agentforce execute deterministic tasks and data interactions, including prompt template, flow, Apex code, named query, and custom actions.

Define actions with unique names, clear descriptions, precise inputs and outputs, and targets, while applying best practices like require user confirmation, progress indicators, and smart input binding for structured results.

Atlas reasoning engine powers the agent force system, enabling proactive thinking, planning, reasoning, and action through hybrid reasoning, react prompting, and transparent grounding.

Explore the Einstein trust layer, a secure prompt and response lifecycle that guards data privacy with dynamic grounding, zero data retention, toxicity detection, and secure prompt-to-LLM pathways.

Explore the legacy agent force builder to create agents, configure standard and custom actions, and use the conversation preview, plan tracer, and logs to test and validate behavior.

Configure an employee agent to summarize Salesforce cases using the legacy builder, create sub-agents and actions, and test case summaries with identify record by name and summarize record.

Activate the sub-agent, create and assign a permission set to the employee, then attach it to the agent and enable end-user access with testing via the agent force builder.

Use Agentforce flows as custom actions to enable agents with a low-code approach, employing auto launch flows with no trigger and required input and output variables for Salesforce data interactions.

Design and implement an employee agent for case management automation using custom actions in flow to update case priority, update case owner, and create follow-up tasks.

Create an auto launch flow to update a case priority by defining input variables for case id and priority, implementing fault and success messages, saving the flow, and testing it.

Create and activate a Salesforce flow that updates a case owner by fetching the user by username, validating the ID, and handling success and fault paths.

Build an auto launch flow to create a follow-up task for a case using input case id, due date, and subject, and assign the task to the case owner.

Expose auto launch flows by creating agent force assets in setup; create custom actions for the employee agent: update case priority, update case owner, create follow-up task.

this hands-on activity guides configuring a case management automation agent in Salesforce, using sub-agents, Atlas reasoning engine, and rules for priority, ownership, and task due dates.

Learn to build custom actions for Salesforce agents using Apex invocable methods. Apply bulkified, secure Apex with input/output wrappers and manage permissions and error handling.

Learn to build a custom Apex action that calculates an escalation score from multiple parameters using an invocable method, with case ID input and high risk classification.

Create and configure the agent force asset with a custom Apex action to calculate and display the escalation score for a case, including high-risk status and a score summary.

Learn to create and test prompt templates with the Salesforce prompt builder, grounding outputs in flows, apex, and data cloud for secure, custom actions in agent force.

Learn to generate a personalized welcome email with the sales email prompt template by grounding recipient, sender, and organization data in a Trailhead playground using Einstein for sales.

Explore how to create and expose a field generation prompt template, enable dynamic forms, and generate AI-powered descriptions for a specific field like opportunity description.

Create a record summary prompt template to generate a grounded, rich Salesforce case summary using the record snapshot, case data, and related lists.

Expose a prompt template as a custom action in AgentForce to generate case summaries, including identifying the customer record and invoking the action to produce the summary.

Explore the flex prompt template that consolidates data from multiple sources—account records, a file object, and related lists—to generate a complete 360-degree account overview with grounded insights.

Ground your prompt templates with flows to conditionally bind account summaries to open or escalated cases and closed opportunities, using the template triggered prompt flow and add prompt instruction element.

Ground Apex invocable methods, define capability types for sales email, field generation, or record summary, and align input/output parameter names.

Salesforce Agentforce Masterclass hands-on shows building an apex grounded flex prompt template to compute an account health score using annual revenue, open cases, and case status for an LLM-generated summary.

Invoke prompt templates from a flow using a screen flow and asynchronous path. Ground prompts with Apex and pass account data to an LLM to generate and display AI summary.

Invoke a prompt template from Apex using the connect API to process an uploaded file via a lightning web component, generating answers with the Einstein prompt template and displaying results.

Create a lightning web component that invokes apex to call a prompt template, uploads a file, captures user questions, and displays the LLM output with toast feedback.

Invoke a prompt template from rest api via post request with a value map. Learn to set up Postman, authenticate with a connected app, and test account 360 summary.

Discover why Salesforce introduced the agent script to balance hybrid reasoning and deterministic control, enabling reliable operations with natural language and if-else logic in a single dot agent file.

Master agent script basics: key-value syntax, yaml-like indentation, and at the rate references; use pipes, template expressions, slot filling, and basic conditional logic.

Explore the building blocks of a Salesforce agent script, including system, config, variables, language, connection, and topic blocks, plus start agent, reasoning, and transitions.

Explore the execution flow of an agent script, detailing five steps from starting agent routing to after reasoning, including deterministic logic, prompt buffer, llm reasoning, and optional tool calls.

Create and customize Salesforce agent scripts in VS Code with the AgentForceDX extension, access the AI Authorizing Bundle, edit with AgentForceWipes, and maintain rules and agents.md for versioning in GitHub.

Master the agent script variable block, distinguishing regular and link variables, mutable vs non-mutable options, slot filling with three dots, and how to reference variables in scripts and instructions.

Explore how agent scripts use out-of-the-box utils to navigate to another topic, set variable values, and escalate to a human agent, including utils.transitionTo, utils.set variable, and utils.escalate.

configure actions in agent scripts by adding an action block against a topic, define action metadata with inputs, outputs, and target, and choose deterministic or llm-driven invocation depending on context.

Learn to build a balance transfer agent with agent script, validate inputs and accounts, enforce business rules, and log transactions with audit logs and custom balance and transaction log objects.

Create the financial account and transaction log objects with fields for account number, balance, amount, from account, to account, and status, then assign the agent a permission set.

Build and test balance transfer automation using flows. Fetch and validate accounts, update balances, log transactions, and return transaction IDs for agent script invocation.

Execute a hands-on activity to invoke an agent from a record-triggered flow in the Salesforce Agentforce masterclass, demonstrating practical workflow automation.

Learn to invoke an agent-first from a screen flow using a standard ai action, session ids, and structured outputs to perform a balance transfer with validation and consent.

Invoke Salesforce agents via the agent api using rest: authenticate with a connected app, start sessions, and exchange messages with synchronous or streaming endpoints.

Invoke agents from Apex using the invocable action API by creating a custom action, passing the user message and session ID, and retrieve the agent response for context.

Learn to deploy and invoke Salesforce agents from Slack by configuring the Slack platform connector, approving connections, mapping users, and exposing a legacy agent as a Slack app.

Learn how context variables pass customer data to agents via the messaging session, with read-only variables accessed by the dollar context prefix and automatic user detail retrieval.

Clone the customer community logged-in user, create test accounts, contacts, and cases, and configure experience cloud user access to enable the agent to list all cases without emails.

Enable the experience by applying customer login test profile, build a lightning web component to capture the logged-in user’s email, and embed it on the community page via head markup.

Define a custom parameter in messaging channel, name must match earlier, string length 50, enable pre-chat; map to flow input and publish to pass email to inbound flow.

Create a text field on the messaging session, grant the agent access, and update the flow to populate the customer email from the embedded messaging session, validating the record.

Create an invocable apex action to fetch cases by contact email. Return success, message, case count, and a formatted case summary, with permission set and sharing checks.

Learn how retrieval augmented generation grounds LLMs with company data—without retraining—by augmenting prompts with data from data 360 via keyword, vector, and hybrid searches.

Explore how the data cloud processes unstructured data with chunking, vectorization, and semantic search to enable rag in data360, using retrievers and prompts for intelligent agent responses.

Understand retriever concepts in agent force data library, including default and custom retrievers, individual versus ensemble retrievers, and their impact on data retrieval and credits.

Discover how the AgentForce data library automates rag setup by uploading data, creating data streams, lakes, retrievers, and data models, then using prompts to answer queries from indexed knowledge.

Shows how to create a data library, upload files, and use a prompt with semantic search to answer cancellation and damage policy questions.

Enable an AI agent to reference Salesforce knowledge articles as the trusted source for billing and payments, by configuring knowledge, creating a data library, and publishing articles.

Activate the data library and prompt builder to retrieve knowledge article content from the data cloud, testing with a car rental billing query.

Explore how to build custom agent UIs with lightning type overwrite in Salesforce, replacing generic UI with branded, structured inputs and rich outputs.

Define a lightning web component that captures user inputs via the lightning agent force input, using handleInputChange and the value change event to update state for the planner.

Build a lightning web component to render lead creation output and confirmation. Expose it to the agent force output and configure lead input and output lightning types.

Fix lead creation errors by adding no-argument constructors for the lead request and lead DTO, and switch the booklet component to custom lightning input and output types.