Agentforce Concepts and Architecture

Agentforce & AIAgentforce Concepts and ArchitectureWhat an Agentforce agent really is: the reasoning engine, subagents (formerly topics), actions, variables, knowledge, channels, running users and guardrails, plus the metadata that represents an agent in source control and the design principles that make agents reliable.Intermediate · about 75 minutesWhat you will learn

Name the building blocks of an Agentforce agent and explain how a single conversational turn is processed

Describe the Atlas Reasoning Engine loop and where hybrid reasoning changes it

Write effective subagent (topic) descriptions, scopes and instructions

Explain agent types, the agent user and the security context in which actions run

Map agents to metadata types (Bot, BotVersion, GenAiPlannerBundle, GenAiPlugin, GenAiFunction, AiAuthoringBundle) for source control

Apply design principles for subagent boundaries and determinism versus flexibility

Contents

What is an agent?

The Atlas Reasoning Engine

Subagents (topics)

Actions

Variables and context

Agent types and templates

The agent user and security context

Knowledge

Channels

Session lifecycle and memory

Guardrails

Agents in metadata

Design principles

Worked example: designing "Pronto Bikes Support"

Best practices

Common pitfalls

Hands-on exercises

Key takeaways