Azure Logic Apps Introduces "Agent Loop" for Building AI Agents in Enterprise Workflows

At the annual Build conference, Microsoft announced Agent Loop, a new feature in Azure Logic Apps that allows developers to directly integrate AI agents into enterprise workflows.

Agent Loop is a core component for AI agent development in Logic Apps. It is a new type of operation that integrates selected AI models (such as Azure OpenAI), domain-specific tools (via Logic Apps connectors), and enterprise knowledge sources. Through this component, developers can create various types of AI agents, including autonomous agents for loan approvals, conversational agents for customer support, and multi-agent systems for coordinating activities like sales report generation.

Built upon the Kernel object in Semantic Kernel, Agent Loop utilizes LLMs to determine necessary steps. Meanwhile, the Azure Logic Apps runtime handles the execution of these plans. This approach provides significant flexibility, enabling the creation of agents that can both converse and act autonomously, responding to real-time events through Logic Apps' extensive library of connectors.

Divya Swarnkar, a project manager at Microsoft, stated:

With over 1,400 connectors, Logic Apps has a unique advantage in providing rich context and seamless access to enterprise systems and APIs for AI agents, allowing them to reason and act reliably.

Agent Loop operates through an iterative "think, act, and learn loop." The AI agent reasons based on its goals and context, takes actions by invoking connectors, and then reflects on results to adjust its plans when needed. Azure Logic Apps automatically manages this cycle.

Microsoft highlighted several potential use cases for AI agents built with Agent Loop, including:

• Product return agents: Verifying order details, return eligibility, and processing refunds or requesting further information.
• Loan approval agents: Evaluating credit scores, income, and risk profiles to automatically approve or forward applications.
• Recruitment agents: Screening resumes, summarizing qualifications, and drafting personalized outreach messages.
• Sales report generation workflow: Utilizing multiple agents to draft, review, and publish reports.
• IT operations agents: Triage alerts, check changes, and resolve common issues or escalate when necessary.
• Multi-agent retail supply chain solutions: Coordinating inventory and logistics agents for timely replenishment and optimized fulfillment.

Furthermore, key advantages of building AI agents using Agent Loop in Logic Apps include declarative orchestration, code extensibility, access to a vast library of integration tools, observability with full traceability, enterprise-grade governance inheriting Azure Logic Apps security and compliance, simple human-agent collaboration and multi-agent coordination, and faster time-to-value by abstracting away boilerplate code of agent architecture.

Kent Weare, Principal Project Manager of Microsoft Logic Apps, expressed:

Building agents or workflows isn't a binary choice. The most effective solutions often combine both - this is where Logic Apps excels. With Agent Loop, customers have complete control to adjust the level of agent automation that suits their needs. Logic Apps is where traditional workflows and AI agents come together, combining forces to solve complex business problems - all within a trusted enterprise-grade platform. I don't think any other platform offers such capabilities!

Further emphasizing its potential, Cameron McKay, Azure Application Architect, summarized the Agent Loop feature in Logic Apps in a LinkedIn blog post:

This feature holds great potential; the number of use cases depends on the business and implementers; some use cases include responding to error events and executing processes based on conversations with humans. I'm looking forward to seeing how this feature's use cases evolve; without a doubt, it's a highly transformative and useful addition to the Azure Logic Apps toolkit.

Agent Loop is available in Azure Logic Apps Standard, and the company has provided documentation and demos to help developers get started. It also outlined future plans, including multi-agent handoff support, A2A (Agent-to-Agent) protocol support, and OBO authentication for Logic Apps agents.