Azure Logic Apps supports LLM-backed agentic workflows

Azure Logic Apps supports agentic workflows that use agent loops with large language models to complete complex, multi-step tasks. The loops interpret prompts, process inputs, choose tools, and return results through autonomous or conversational experiences. Consumption agentic workflows are in preview, while both Consumption and Standard logic apps can use the pattern for predictable business processes that still require flexibility.

Agentic workflows differ from conventional workflows by handling unstructured data, responding to changing conditions, and selecting tools based on available information rather than fixed sequences. Tools are built from Logic Apps actions and connectors, giving agents access to services, APIs, email, data sources, calculations, and other tasks. The platform includes 1,400+ connectors for building tools across a wide range of scenarios.

Autonomous workflows can start from supported triggers and operate without human interaction, using Azure OpenAI Service, Microsoft Foundry, or LLM APIs depending on the logic app type. Conversational workflows start when a chat session begins and respond through an integrated or external chat interface. Production deployments require stronger authentication: Consumption uses OAuth 2.0 with Microsoft Entra ID, while Standard uses Easy Auth to validate callers and enforce access controls.