AI agent workflows in Azure Logic Apps

At this year’s INTEGRATE Summit, the Azure On Air podcast featured a timely and insightful conversation between Microsoft’s Kent Weare (Logic Apps product team) and Michael Stephenson, a well-known integration expert and Microsoft MVP. Their discussion offered a rare behind-the-scenes view of how AI is transforming the way businesses build integration solutions in Azure and where the technology is heading next.

Here are the major takeaways.

From Static Queries to Dynamic Agents

The conversation opened with a reflection on how far Logic Apps and AI have come in just 12 months.

In 2023, the focus was on retrieval-based AI (like RAG models), where users could ask a question, get an answer, and maybe extract some insight from their documents. That alone had promise, particularly for enhancing visibility into tickets, trends, and unstructured data.

But in 2024, the shift is clear: it’s not just about asking AI for information it’s about letting AI do the work.

With new AI agent capabilities in Logic Apps, teams can now create multi-turn, dynamic interactions between AI and business systems. Instead of triggering a single operation, agents can parse documents, make decisions, orchestrate tools, and handle processes from end to end.

Real-World Integration Scenarios

Much of the value of AI agents lies in solving challenges that were previously brittle, slow, or overly complex:

• Parsing unstructured or semi-structured data like emails or PDFs
• Routing based on policies or tribal knowledge that isn’t codified in a system
• Avoiding brittle parsing logic by letting AI impose structure dynamically

Kent highlighted how agents now allow teams to build workflows based on “knowledge documents” such as return policies, operational guides, or FAQs and use those documents to drive decisions rather than just respond to queries.

This opens up new possibilities for business process orchestration that would have been too fragile or time-intensive using traditional methods.

Agents, Tools, and the Composable Workflow Model

The podcast also outlined a pattern that’s beginning to emerge in modern integration architecture: decomposition and reuse.

Rather than building monolithic workflows, teams are encouraged to:

• Break down processes into smaller child Logic Apps
• Expose those workflows as tools to AI agents
• Use knowledge and context to dynamically compose orchestration on the fly

This model aligns well with Microsoft’s broader “plug-and-play” vision for MCP servers and agent-to-agent (A2A) interoperability, allowing seamless collaboration between AI agents across systems (e.g., Logic Apps calling a Workday agent).

Challenges: Mindset, Cost & Governance

While the potential is significant, both Kent and Michael were quick to point out that AI isn’t the answer to everything.

Key considerations include:

• Business Value First: AI agents should solve real problems, not be used just for buzz.
• Latency Sensitivity: For high-speed transactions, agents may introduce delays. They’re better suited for processes with human latency (approvals, reviews, returns, etc.).
• Governance Is Key: Non-deterministic systems require strict boundaries what AI can and can’t do needs to be clearly defined.
• Cost Awareness: AI workloads are metered by tokens. Understanding things like token usage, context windows, and temperature settings is vital for managing cost and reliability.

Retail Returns: A Practical Example

One of the strongest use cases shared was a retail product return scenario.

Here, an AI agent receives an image of a damaged item, evaluates the return policy, checks the customer’s history, and decides whether to auto-approve or escalate. It’s a scenario that blends automation, judgment, and customer experience making it a perfect example of AI adding business value without unnecessary overhead.

What’s Next:

Looking ahead to 2025, Kent expects:

• Maturation of the MCP server model (standardized plugin-like access to tools across vendors)
• Adoption of the agent-to-agent protocol, enabling platform-neutral AI collaboration
• More production use cases and measurable business impact

As experimentation gives way to scale, Logic Apps with its low-code tools, connector ecosystem, and Azure OpenAI integration could become a powerful launchpad for enterprise-grade AI solutions.

Final Thought

What makes this shift so compelling is its balance of innovation and practicality. Microsoft isn’t pushing AI agents as a silver bullet but rather as a tool to extend and enhance existing workflows, especially where the old rules-based approach falls short.