Over the past two years, I’ve been amazed by how far AzureChat has come. It’s one of the fastest ways to launch a secure, enterprise-ready chatbot in the Azure cloud—often in under an hour. What I love most about it: ✅ Enterprise-grade security (private VNet) ✅ Robust authentication & built-in roles ✅ Seamless tool integration ✅ Ability to process massive documents with ad-hoc RAG ✅ Accessibility features like voice interaction & expert personas But there was one missing piece: long-running, asynchronous tasks. Many real-world AI workflows—like deep research—can’t be answered instantly. That’s why I built the Think Extension for AzureChat. Here’s what it enables: 🔹 Start a complex query → get an immediate ID back 🔹 AzureChat polls results until the answer is ready 🔹 Behind the scenes, an agent orchestrates tools, runs analysis, and delivers the final response when complete This design keeps AzureChat clean and scalable, while adding the “deep thinking” many enterprise use cases need. 💡 Fun insight: During testing, GPT-4 reliably used tools when asked. GPT-5 produced the answer without actually calling the tool, even though the summary looked like the tool was called. A good reminder that with AI, trust but verify is key.