# AI Agent

## What Is an AI Agent?

An **AI Agent** is a rule type that delegates decision logic to a large language model (LLM). Instead of defining explicit conditions and outcomes — as you would in a Decision Table or Decision Tree — you describe what you want the model to do via a natural language prompt and a structured output schema. The model then reasons over the input and returns a typed, structured JSON response every time the rule is called.

What makes AI Agents different from other rule types is their ability to handle **judgment, language, and ambiguity**. They can read a contract and extract its key terms, evaluate a supplier narrative and produce a risk score, validate a dataset against a policy, or write a plain-language summary of a complex decision, all without a single hard-coded condition.

Each AI Agent is best built around a single, clearly defined task. In practice, the most common patterns are:

* **Decision making:** evaluating input and producing a decision, score, or routing outcome
* **Deep analysis:** reading documents, records, and signals to extract structured intelligence
* **Report writing:** condensing complex input into a concise, decision-ready narrative
* **Policy validation:** checking data against rules and standards; fixing and normalizing non-conforming records

If your use case doesn't fit any of these patterns, you can define the role and task freely from scratch in the prompt.

{% hint style="info" %}
**When not to use an AI Agent:** \
If your logic can be fully expressed as explicit conditions and outcomes, a Decision Table or Decision Tree will be faster, more predictable, and easier to audit. Use an AI Agent when the complexity or ambiguity of the input makes explicit rules impractical.
{% endhint %}

## AI Agent vs. Other Rule Types

<table><thead><tr><th width="147.12890625">Parameters</th><th width="193.74609375">Decision Table</th><th width="192.421875">Decision Tree</th><th>AI Agent</th></tr></thead><tbody><tr><td><strong>Input type</strong></td><td>Structured values</td><td>Structured values</td><td>Structured or unstructured</td></tr><tr><td><strong>Logic definition</strong></td><td>Condition/result rows</td><td>Branching conditions</td><td>Natural language prompt</td></tr><tr><td><strong>Output</strong></td><td>Fixed result values</td><td>Fixed result values</td><td>Typed JSON generated by LLM</td></tr><tr><td><strong>Determinism</strong></td><td>Fully deterministic</td><td>Fully deterministic</td><td>Probabilistic (caching available)</td></tr><tr><td><strong>Auditability</strong></td><td>Full row-level trace</td><td>Full branch trace</td><td>Explainable AI output (optional)</td></tr><tr><td><strong>Best for</strong></td><td>Tabular business rules</td><td>Hierarchical decisions</td><td>Reasoning, scoring, extraction</td></tr></tbody></table>

## What You Will Find in This Section

{% content-ref url="ai-agent/ai-agent-designer" %}
[ai-agent-designer](https://docs.decisionrules.io/doc/rules/ai-agent/ai-agent-designer)
{% endcontent-ref %}

{% content-ref url="ai-agent/caching" %}
[caching](https://docs.decisionrules.io/doc/rules/ai-agent/caching)
{% endcontent-ref %}

{% content-ref url="ai-agent/explainable-ai" %}
[explainable-ai](https://docs.decisionrules.io/doc/rules/ai-agent/explainable-ai)
{% endcontent-ref %}