Definition
An Agent Topic in Agentforce defines a specific subject area or domain of responsibility for an AI agent. Topics group related actions and instructions together, telling the agent what it should handle (for example, 'Order Management' or 'Returns'). Each topic includes a description, scope, instructions, and a set of associated actions that the agent can execute when a conversation falls within that topic.
Real-World Example
a data scientist at CognitiveTech recently implemented Agent Topic to automate a complex decision-making process that used to rely on gut instinct. By deploying Agent Topic, the organization now uses data-driven intelligence to guide actions, resulting in better customer outcomes and more efficient use of team resources.
Why Agent Topic Matters
Topics are the primary organizational unit for an Agentforce agent's knowledge. Each topic has a name, a classification description (used by the Atlas Reasoning Engine to decide when the topic applies), a scope, instructions that guide the agent's behavior within the topic, and a set of actions the agent can invoke while the topic is active. When a user message arrives, Atlas picks the best-matching topic and scopes the agent's responses and available tools to that topic.
Grouping actions into topics keeps agents manageable as they grow. Rather than giving an agent access to every possible action all the time, topics let you expose the right subset for each conversation context. Topics can also be reused across agents, so a well-designed 'Returns' topic built once can power the returns experience in a customer agent, an employee agent, and a partner agent.
How Organizations Use Agent Topic
- •Wanderlust Travel — Built a 'Booking Management' topic with actions for looking up reservations, modifying travel dates, and cancelling bookings. The topic's instructions tell the agent to always confirm the booking reference before making any change, which the agent follows consistently without explicit code.
- •DeepSight Analytics — Created an 'Access Requests' topic that handles internal employees asking for dataset permissions. The topic's scope is tightly limited to a small set of actions, which means the agent can't accidentally wander into answering security policy questions that require a human.
- •ShieldGuard Security — Reused their 'Account Lookup' topic across three different agents: a customer support agent, a field technician agent, and an internal admin agent. Each agent has the same actions available when that topic is active, which cut development time significantly.