Stefano Demiliani

@demiliani.bsky.social

Microsoft MVP. Microsoft MCT Regional Lead. CTO @ EID Lodestar spa. #Azure #MSDyn365BC Solution Architect. #AI engineer. International speaker & Author. Runner & Cyclist. http://www.demiliani.com

@bctechdays.com is calling... Waiting for you in #Antwerp, where passion meets expertise, and every conversation is a pull request worth merging. 🚀 #msdyn365bc #BCTechDays 👇 👇

BC TechDays@bctechdays.com · 4mo ago

#BCTechDays remains the benchmark event for #msdyn365bc developers. If you want to contribute to this year’s program with your expertise, now is the time. Submit your session proposal and be part of the BC TechDays experience! Call for Speakers closes 5 April! 🌐 mibu.so/cfs26

Testimonial of Stefano Demiliani regarding BC TechDays

With 𝗔𝘇𝘂𝗿𝗲 𝗟𝗼𝗴𝗶𝗰 𝗔𝗽𝗽𝘀 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱 you can now create #MCP servers for many apps using built-in connectors and you can convert existing workflows into discoverable MCP tools. Why not using it for exposing MCP servers for 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗖𝗲𝗻𝘁𝗿𝗮𝗹?🚀 demiliani.com/2026/03/03/c... #msdyn365bc #AI #LogicAppsAviators

Creating MCP servers for Dynamics 365 Business Central with Azure Logic Apps.

Model Context Protocol (MCP) servers standardize how AI agents interact with external systems and data and Azure Logic Apps is the enterprise-ready tool for creating low-code workflows in the Azure…

demiliani.com

𝗤𝘄𝗲𝗻 𝟯.𝟱, the latest series of AI models by Qwen are very interesting, expecially the Qwen3.5-Medium Series that can be used offline. What about #AL coding offline? Verdict: sometimes bad (low quantizations), sometimes acceptable (starting at least from 𝗤𝘄𝗲𝗻𝟯.𝟱-𝟯𝟱𝗕-𝗔𝟯𝗕 𝟴 𝗯𝗶𝘁). #msdyn365bc #AI

Bild

𝗠𝗶𝗻𝗶𝗠𝗮𝘅-𝗠𝟮.𝟱, the new open model by MiniMax. Impressive in speed and it costs just $1 to run the model continuously for an hour at 100 tok/sec. At 50 tok/sec the cost drops to $0.30. Fully comparable with Sonnet 4.5, but faster and 11x cheaper. Oh, very good in #AL coding too😉 #msdyn365bc #AI #LLM

The day I was waiting for Claude 5 to be released, I fell in love 😍 with the new 𝗤𝘄𝗲𝗻𝟯-𝗖𝗼𝗱𝗲𝗿-𝗡𝗲𝘅𝘁 released yesterday. Here is the 8 bit version running FULLY LOCAL and writing #AL code. With GPT-OSS-120B, it's absolutely the best AI model for writing #AL code fully offline. 🚀 #msdyn365bc #llm #AI

𝗞𝗶𝗺𝗶 𝗞𝟮.𝟱, the latest native multimodal agentic model by Moonshot AI (released some days ago) built through continual pretraining on approximately 15 trillion mixed visual and text tokens on top of Kimi-K2-Base, rocks on #AL development too... 👍 (and saves you money 💰 ) #msdyn365bc #AI

𝗚𝗟𝗠-𝟰.𝟳-𝗙𝗹𝗮𝘀𝗵-𝟴𝗯𝗶𝘁, the new 30B parameter MoE model by Z.AI. Here it's running local, trying to write #AL code. As you can see, not good for #AL dev🤭 For C# or Python coding, it's instead a very good offline model. 𝗚𝗟𝗠-𝟰.𝟳 full (online) is instead great on #AL, see my previous test. #msdyn365bc #AI