Benefits of hybrid reasoning
Key Takeaway
Write anything that must happen the same way every time, such as a business rule or a required action, as deterministic logic. Leave judgment and conversation to natural language instructions that the LLM interprets.
Hybrid reasoning is how Agentforce agents mix two kinds of instructions in one agent. Deterministic logic runs like code, the same way every time. Natural language instructions are interpreted by a large language model (LLM). Together they give enterprises predictability and control, plus the flexibility and creativity that LLMs make possible.
Hybrid reasoning and Agent Script
Agent Script combines two kinds of instructions in one script. Programmatic expressions are the deterministic part. They include if/else conditions, transitions, setting and comparing variables, and running actions. Natural language instructions are the part the LLM interprets. Combining the two is hybrid reasoning.
The Atlas reasoning engine is graph-based. It works like a flowchart with nodes, variables, and transitions, so agents follow specific, predictable paths. Unlike strictly prompt-based reasoning engines, Atlas separates an agent's big-picture workflow from its conversational skills.
Hybrid reasoning is not one global switch. You make the choice node by node. A node with prompt instructions triggers an LLM call, and a node without them runs deterministically. Because that trigger is explicit in the script, the LLM boundary is visible and enforced instead of being inferred at runtime.
Beginning in April 2026, agent topics are called subagents. Functionality didn't change, so you may see both terms for the same building block.