Topic instructions in Agent Builder
Key Takeaway
Instructions guide an agent, but an LLM interprets them, so they can't reliably enforce critical business logic. Build that logic into an agent action instead.
Subagent instructions are guidelines inside a subagent that tell an Agentforce agent how to handle a conversation. Beginning in April 2026, agent topics are called subagents, with no change in how they work, so you will see both terms. Admins write and refine these instructions, then test them by chatting with the agent in the Agentforce Builder Preview panel and reviewing how it reasoned.
What a subagent is
A subagent is one job an agent can do. It contains actions, which are the tools for the job, and instructions, which tell the agent how to make decisions.
The subagents assigned to an agent together define what the agent can handle. You can use standard subagents for common use cases or create custom ones.
In the legacy builder, a subagent also has a name, a classification description, and a scope. The name and classification description help the agent choose the subagent. The scope sets limits on what the agent can and can't do with the subagent, for example "Your job is only to... You cannot...".