📢Announcing agent loop: Build AI Agents in Azure Logic Apps 🤖

DivSwa

Microsoft

May 19, 2025

This post is written in collaboration with Kent Weare and Rohitha Hewawasam

The era of intelligent business processes has arrived! Today, we are excited to announce agent loop, a groundbreaking new capability in Azure Logic Apps to build AI agents into your enterprise workflows. With agent loop, you can embed advanced AI decision-making directly into your processes – enabling your apps and automation to not just follow predefined steps, but to reason, adapt, and act autonomously towards goals.

Agent loop becomes central to AI Agent development — it’s a new action type that brings together your AI model of choice, domain-specific tools, and enterprise knowledge sources. Whether you’re building an autonomous agent to process loan approvals, a conversational agent to support customers, or a multi-agent system that coordinates tasks such as Sales Report generation across agents, Agent Loop enables your workflows to go beyond static steps — making decisions, adapting to context, and delivering outcomes.

Agent loop is implemented using kernel object in the Semantic Kernel. The kernel object, along with an LLM, creates the plan for what needs to be done, while Logic Apps runtime handles execution of that plan. Agent Loop is highly configurable, enabling you to build agents with diverse capabilities:

Conversational or Autonomous Agents

With Logic Apps' extensive gallery of connectors, you can build fully autonomous agents that respond to real-time events — like new records in a database, files added to a share, or messages in a queue.

Agent Loop also supports conversational agents via Channels, allowing agents to interact with users through the Azure portal or custom chat clients.

Bring your own Model

Associate your AI agent with any Azure OpenAI model of your choice. As new models become available, you can easily switch or upgrade without re-architecting the solution.

Define Agent Goals and Guardrails

Specify your agent’s objective and behavioral boundaries through system prompts and user instructions. Using connectors like Outlook or Teams, you can easily introduce human-in-the-loop interactions for approvals or overrides — enabling safe, controlled autonomy.

Tools and Knowledge, Built In

Leverage hundreds of out-of-the-box connectors to equip agents with access to enterprise systems, APIs, and business data. Enrich their reasoning with knowledge from vector stores, structured databases, or unstructured files, and empower them to take meaningful actions across your environment.

AI Agents in Action

Here are some examples of AI Agents in Action that highlight the value and efficiencies of these agents across different domains and solution areas.

A product return agent verifies order details, return eligibility, and refund rules, then processes the return or requests additional info from the customer.
A loan approval agent evaluates credit score, income, and risk profile, applies business rules, and auto-approves or routes applications for review.
A recruiting agent screens resumes, summarizes qualifications, and drafts personalized outreach to top candidates, streamlining early hiring stages.
A sales report generation workflow uses a writer agent to draft content, a reviewer agent to verify accuracy, and a publisher agent to format and distribute the report.
An IT operations agent triages alerts, checks recent changes, and either resolves common issues or escalates to on-call engineers when needed.
A multi-agent retail supply chain solution combines inventory and logistics agents to ensure timely restocks and optimize fulfillment routes.

Why agent loop matters

Modern businesses thrive on agility and intelligence. Traditional workflows remain essential for deterministic tasks—especially those involving structured data or high-risk decisions. But when processes involve unstructured data, changing context, or require adaptive decision-making, AI agents excel. They can reason, act in real time, and dynamically sequence steps to meet goals. Agent Loop exactly serves this purpose.

What makes Agent Loop especially powerful is its deep integration with the Logic Apps ecosystem. Logic Apps comes with over 1,400+ connectors for Microsoft and third-party services – from databases and ERP systems to SaaS applications and custom APIs. They can also invoke custom code and scripts, making it easy to tap into homegrown capabilities. The agent isn’t limited to information in its prompt; it can actively retrieve knowledge, perform transactions, and effect change in the real world via these connectors. Logic Apps is uniquely positioned to enable customers to leverage their API and connector ecosystem cohesively across their workflows and AI Agents to build agentic applications.

Equally important, Agent Loop is designed for flexibility. You can orchestrate single-agent workflows or coordinate multiple agents working in tandem towards a common goal. Agent Loop can even involve humans in the loop when needed – for instance, pausing to get a manager’s approval or to ask for clarification – leveraging Logic Apps’ human workflow capabilities. All of this is handled within the familiar, visual Logic Apps designer, so you get a high-level view of the entire orchestration.

How agent loop works

At a high level, Agent Loop works by pairing the reasoning capabilities of large-scale AI models with the robust action framework of Logic Apps. Built on top of Semantic Kernel, the Agent loop operates in iterative cycles, allowing the agent to think, act, and learn from each step:

Reasoning (Think): The agent (powered by an LLM like Azure OpenAI Service under the hood) and on Semantic Kernel, examines its goal and the current context. It decides what needs to be done next – whether that’s gathering more information, calling a specific connector, or formulating an answer. This step is essentially the AI “planning” its next action based on the goal you’ve provided and the data it has so far.
Action (Act): The agent then carries out the decided action by invoking a tool or connector through Logic Apps. This could be anything from querying a database, calling a REST API, sending an email, to running a calculation. Thanks to Logic Apps’ extensive connector library, the agent has a rich toolbox at its disposal. Each action is executed as a Logic Apps step, meaning it’s secure, managed, and logged like any other workflow action.
Reflection (Learn): After the action, the agent receives the results (e.g. data retrieved, outcome of the API call, user input, etc.). It then evaluates: Did this bring it closer to the goal? Does the plan need adjusting? The agent updates its understanding based on new information. This reflection is what lets the agent handle complex, open-ended tasks – it can correct course if needed, try alternative approaches, or conclude if the goal has been satisfied.

These steps repeat in a loop. The Agent Loop action manages this cycle automatically – calling the AI model to reason, executing the chosen connector operations, feeding results back, and iterating.

Why Build AI Agents in Logic Apps?

Building AI agents is an emerging frontier in automation but doing it from the ground up can be daunting especially when organizations build them in large numbers. Agent Loop in Logic Apps makes this dramatically easier and more scalable for several reasons:

Declarative Orchestration: Logic Apps provides a visual workflow canvas and a serverless runtime. The Agent Loop action plugs into this and the platform handles the sequence of steps and iterations, so you can focus on defining the goal and selecting the connectors (tools) the agent can use.
Code extensibility: Logic Apps supports both declarative and code-first approaches to building agents. You can combine the two — using visual designer for orchestration and injecting code where needed through extensibility points. Write custom logic in C#, PowerShell, JavaScript, or use inline scripts for lightweight processing. Python support is coming soon, enabling even more flexibility.
1400+ Integrated Tools: With the rich connector ecosystem at its disposal, your agent can seamlessly tap into your enterprise systems and SaaS applications. Your entire ecosystem of connectors, APIs, custom code and agents can be used by deterministic workflows and agents to solve business problems
Observability: Logic Apps offers full traceability into each agent’s decisions and actions. Every run is logged in the workflow history, with data stored within the customer’s own network and storage boundaries. The Agent Chat view provides insights into the agent’s reasoning, tool invocations, and goal progress. Developers can easily revisit these logs for debugging, auditing, or analysis.

Enterprise-Grade Governance: Because it runs on Azure Logic Apps, agent loop inherits all the robust monitoring, logging, security and compliance capabilities of the platform You can secure connections with managed identities and leverage built-in rate limiting, retries, and exception handling. Your AI agents run with the same enterprise-ready guardrails as any mission-critical workflow.
Human-in-the-Loop & Multi-Agent Coordination: Logic Apps makes it straightforward to involve people at key decision points or to coordinate multiple agents. You can chain Agent Loop actions or have agents invoke other workflows, enabling collaborative problem-solving that would be difficult to implement from scratch. The result is a system where AI and humans can smoothly interact and complement each other.
Faster Time to Value: By eliminating the boilerplate work of building an agent architecture (managing memory, planning logic, connecting to services, etc.), Agent Loop lets developers and architects concentrate on high-value logic and business goals, accelerating how you bring AI-driven improvements to your business processes.

In short, agent loop combines the brains of generative AI with the brawn of Azure’s integration platform. It offers a turnkey way to build sophisticated AI-driven automation without reinventing the wheel. Companies no longer have to choose between the flexibility of custom AI solutions and the convenience of a managed workflow service – with Logic Apps and Agent Loop, you get both.

Getting Started

Agent Loop is available in Logic Apps Standard starting today! Here are some resources to help you begin:

Documentation: Explore the agent loop concepts and detailed guide with step-by-step instructions on how to configure and use Agent Loop.
Samples & Demos: Watch pre-recorded demos showcasing both conversational and autonomous agent scenarios built with Agent Loop. You'll also get a preview of exciting features coming soon.

Looking Ahead

Agent Loop opens up a new realm of possibilities for what you can achieve with Azure Logic Apps. It blurs the line between application integration and AI, allowing workflows to evolve from static sequences into adaptive, self-directed processes. We can’t wait to see what you will build with Agent Loop!

This is just the beginning. We’re actively investing in new capabilities that are planned for release soon

Multi-agent Hand-off Support – A multi-agent application with hand-off capabilities enables different agent-loops to collaborate by transferring tasks between one another based on expertise or context, which is crucial for building agentic applications that can dynamically adapt to complex, evolving goals and user needs.
A2A (Agent-to-Agent) protocol support – A2A is a communication standard that defines how autonomous agents exchange messages, share context, and coordinate actions in a secure and structured way. It’s especially important in building agentic applications because it ensures interoperability, enables seamless hand-offs between agents, and maintains context integrity across different agents working toward a shared goal. This will allow Logic Apps agents to seamlessly integrate with other agentic platforms.
OBO Auth for Logic Apps Agents: On Behalf Of Auth support for logic Apps agents would allow Logic Apps agents to use logged-in users identity for authentication when invoking Logic Apps connectors as part of agent-loop execution. This will enable building conversational applications to dynamically perform OAuth flows for fetching consent from log-in users to invoke Logic Apps connectors on logged-in user’s behalf.