Stefano Demiliani
@demiliani.bsky.social
Microsoft MVP. Microsoft MCT Regional Lead. CTO @ EID Lodestar spa. #Azure #MSDyn365BC Solution Architect. International speaker & Author. Runner & Cyclist.
http://www.demiliani.com
http://www.demiliani.com
Are you ready for the new Dynamics 365 Business Central icon? And to refresh your tattoo too? 😂
#msdyn365bc #EMEA2025 #DirectionsEMEA
#msdyn365bc #EMEA2025 #DirectionsEMEA
November 4, 2025 at 8:29 AM
Are you ready for the new Dynamics 365 Business Central icon? And to refresh your tattoo too? 😂
#msdyn365bc #EMEA2025 #DirectionsEMEA
#msdyn365bc #EMEA2025 #DirectionsEMEA
We have a new king 👑 @cursor.com.web.brid.gy today released 𝗖𝗼𝗺𝗽𝗼𝘀𝗲𝗿 𝟭, their new agent model designed for software engineering intelligence and speed (4x times faster than similar models). And I can confirm this! A new gold star also in #AL development. Impressive! 🙌🏻🚀
#msdyn365bc
#msdyn365bc
October 29, 2025 at 6:31 PM
We have a new king 👑 @cursor.com.web.brid.gy today released 𝗖𝗼𝗺𝗽𝗼𝘀𝗲𝗿 𝟭, their new agent model designed for software engineering intelligence and speed (4x times faster than similar models). And I can confirm this! A new gold star also in #AL development. Impressive! 🙌🏻🚀
#msdyn365bc
#msdyn365bc
MiniMax just released 𝗠𝗶𝗻𝗶𝗺𝗮𝘅-𝗠𝟮, a compact open source MoE model (230B total params with 10 billion active) built for coding and agentic tasks. Around 10% of Sonnet costs, MIT license. Regarding #AL coding: good results, but too much verbose on token usage.
It can also run locally😉 #msdyn365bc
It can also run locally😉 #msdyn365bc
October 27, 2025 at 7:50 PM
MiniMax just released 𝗠𝗶𝗻𝗶𝗺𝗮𝘅-𝗠𝟮, a compact open source MoE model (230B total params with 10 billion active) built for coding and agentic tasks. Around 10% of Sonnet costs, MIT license. Regarding #AL coding: good results, but too much verbose on token usage.
It can also run locally😉 #msdyn365bc
It can also run locally😉 #msdyn365bc
My favorite #AI coding model ever just got even better. 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗼𝗻𝗻𝗲𝘁 𝟰.𝟱 is here to rule them all... yes, also on writing #AL code for #msdyn365bc 😉
Congrats @anthropic.com 👏🏻
Congrats @anthropic.com 👏🏻
September 29, 2025 at 6:31 PM
My favorite #AI coding model ever just got even better. 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗼𝗻𝗻𝗲𝘁 𝟰.𝟱 is here to rule them all... yes, also on writing #AL code for #msdyn365bc 😉
Congrats @anthropic.com 👏🏻
Congrats @anthropic.com 👏🏻
𝗴𝗽𝘁-𝟱-𝗰𝗼𝗱𝗲𝘅 writing #AL code... honestly it don't impress me a lot. But I will give it more chances. Very good for planning.
#msdyn365bc
#msdyn365bc
September 23, 2025 at 7:12 PM
𝗴𝗽𝘁-𝟱-𝗰𝗼𝗱𝗲𝘅 writing #AL code... honestly it don't impress me a lot. But I will give it more chances. Very good for planning.
#msdyn365bc
#msdyn365bc
🔥PREVIEW🔥: The misterious 𝗦𝗼𝗻𝗼𝗺𝗮 𝗦𝗸𝘆 𝗔𝗹𝗽𝗵𝗮 (foundation for the upcoming 𝘎𝘳𝘰𝘬 4.2-𝘵𝘩𝘪𝘯𝘬𝘪𝘯𝘨) with 2M context window, impressive in speed 🚀 on writing #AL code and extremely efficient on token usage! 🤩 xAI is still a super top player on writing code for #msdyn365bc
September 11, 2025 at 7:04 PM
🔥PREVIEW🔥: The misterious 𝗦𝗼𝗻𝗼𝗺𝗮 𝗦𝗸𝘆 𝗔𝗹𝗽𝗵𝗮 (foundation for the upcoming 𝘎𝘳𝘰𝘬 4.2-𝘵𝘩𝘪𝘯𝘬𝘪𝘯𝘨) with 2M context window, impressive in speed 🚀 on writing #AL code and extremely efficient on token usage! 🤩 xAI is still a super top player on writing code for #msdyn365bc
And here is the latest requested speed test🚀: 𝗚𝗿𝗼𝗸-𝗰𝗼𝗱𝗲-𝗳𝗮𝘀𝘁-𝟭 (top) vs 𝗚𝗣𝗧-𝟱 (bottom) writing #AL code. Same result (generated code is perfect!), quite the same token usage, but the winner in speed is still 𝗚𝗿𝗼𝗸-𝗰𝗼𝗱𝗲-𝗳𝗮𝘀𝘁-𝟭.👏🏻
#msdyn365bc #AI
#msdyn365bc #AI
September 3, 2025 at 2:08 PM
And here is the latest requested speed test🚀: 𝗚𝗿𝗼𝗸-𝗰𝗼𝗱𝗲-𝗳𝗮𝘀𝘁-𝟭 (top) vs 𝗚𝗣𝗧-𝟱 (bottom) writing #AL code. Same result (generated code is perfect!), quite the same token usage, but the winner in speed is still 𝗚𝗿𝗼𝗸-𝗰𝗼𝗱𝗲-𝗳𝗮𝘀𝘁-𝟭.👏🏻
#msdyn365bc #AI
#msdyn365bc #AI
𝗖𝗹𝗮𝘂𝗱𝗲-𝗦𝗼𝗻𝗻𝗲𝘁-𝟰 (on the bottom) vs 𝗚𝗿𝗼𝗸-𝗰𝗼𝗱𝗲-𝗳𝗮𝘀𝘁-𝟭 (on the top).
Same #AL dev task, same perfect result for both. But the first delivered the result in less than 26 seconds 🚀🚀 (with less token usage), while the seconds takes around 40 (and with more tokens).
#msdyn365bc
Same #AL dev task, same perfect result for both. But the first delivered the result in less than 26 seconds 🚀🚀 (with less token usage), while the seconds takes around 40 (and with more tokens).
#msdyn365bc
September 2, 2025 at 7:22 PM
𝗖𝗹𝗮𝘂𝗱𝗲-𝗦𝗼𝗻𝗻𝗲𝘁-𝟰 (on the bottom) vs 𝗚𝗿𝗼𝗸-𝗰𝗼𝗱𝗲-𝗳𝗮𝘀𝘁-𝟭 (on the top).
Same #AL dev task, same perfect result for both. But the first delivered the result in less than 26 seconds 🚀🚀 (with less token usage), while the seconds takes around 40 (and with more tokens).
#msdyn365bc
Same #AL dev task, same perfect result for both. But the first delivered the result in less than 26 seconds 🚀🚀 (with less token usage), while the seconds takes around 40 (and with more tokens).
#msdyn365bc
Congrats 𝗚𝗿𝗼𝗸-𝗖𝗼𝗱𝗲-𝗙𝗮𝘀𝘁-𝟭 🤩 . You're the new winner among the online #LLMs in writing #AL code. Incredibly accurate and super super fast 🚀. Just chapeau! 🙌
#msdyn365bc #AI
#msdyn365bc #AI
August 27, 2025 at 5:39 PM
Congrats 𝗚𝗿𝗼𝗸-𝗖𝗼𝗱𝗲-𝗙𝗮𝘀𝘁-𝟭 🤩 . You're the new winner among the online #LLMs in writing #AL code. Incredibly accurate and super super fast 🚀. Just chapeau! 🙌
#msdyn365bc #AI
#msdyn365bc #AI
#OpenAI GPT-5 sometimes surprises me positively, sometimes not. Here's a case where it shines🤩: multi-agents working together (in parallel) on different tasks in the same #AL project.
#msdyn365bc
#msdyn365bc
August 12, 2025 at 11:58 AM
#OpenAI GPT-5 sometimes surprises me positively, sometimes not. Here's a case where it shines🤩: multi-agents working together (in parallel) on different tasks in the same #AL project.
#msdyn365bc
#msdyn365bc
As usual, @bctechdays.com was a huge success and an amazing experience. Thanks to the entire staff who work behind this event and thanks to all the people attending my sessions or simply chatting with me during the event. See you soon somewhere.
#msdyn365bc #bctechdays
#msdyn365bc #bctechdays
June 19, 2025 at 12:48 PM
As usual, @bctechdays.com was a huge success and an amazing experience. Thanks to the entire staff who work behind this event and thanks to all the people attending my sessions or simply chatting with me during the event. See you soon somewhere.
#msdyn365bc #bctechdays
#msdyn365bc #bctechdays
Because a local LLM 🧠can always be useful…💻 also when landing to @bctechdays.com 🛬🥳
June 11, 2025 at 8:12 AM
Because a local LLM 🧠can always be useful…💻 also when landing to @bctechdays.com 🛬🥳
For all #AI geeks that will be at @bctechdays.com next week I’ve packed a pro-code session to learn everything about one of the most exciting AI frameworks: 𝐌𝐢𝐜𝐫𝐨𝐬𝐨𝐟𝐭 𝐒𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐊𝐞𝐫𝐧𝐞𝐥.
If you want to create enterprise-ready AI agents, probably this could be something interesting for you 😉
#msdyn365bc
If you want to create enterprise-ready AI agents, probably this could be something interesting for you 😉
#msdyn365bc
June 6, 2025 at 10:25 AM
For all #AI geeks that will be at @bctechdays.com next week I’ve packed a pro-code session to learn everything about one of the most exciting AI frameworks: 𝐌𝐢𝐜𝐫𝐨𝐬𝐨𝐟𝐭 𝐒𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐊𝐞𝐫𝐧𝐞𝐥.
If you want to create enterprise-ready AI agents, probably this could be something interesting for you 😉
#msdyn365bc
If you want to create enterprise-ready AI agents, probably this could be something interesting for you 😉
#msdyn365bc
Live picture of our staff working hard for Business Central Day Italy 🐥 🤩🇮🇹
See you next week for sharing a lot of knowledge and fun 🥳
#bcdayit #msdyn365bc @dtacconi.bsky.social
See you next week for sharing a lot of knowledge and fun 🥳
#bcdayit #msdyn365bc @dtacconi.bsky.social
May 13, 2025 at 7:16 AM
Live picture of our staff working hard for Business Central Day Italy 🐥 🤩🇮🇹
See you next week for sharing a lot of knowledge and fun 🥳
#bcdayit #msdyn365bc @dtacconi.bsky.social
See you next week for sharing a lot of knowledge and fun 🥳
#bcdayit #msdyn365bc @dtacconi.bsky.social
It’s official: we made the history 🙌🏻🙌🏻
Thanks everyone who helped on that achievement.
#AI #TogetherWeWin #AISkillsFest @microsoftlearn.bsky.social
Thanks everyone who helped on that achievement.
#AI #TogetherWeWin #AISkillsFest @microsoftlearn.bsky.social
May 1, 2025 at 5:10 PM
It’s official: we made the history 🙌🏻🙌🏻
Thanks everyone who helped on that achievement.
#AI #TogetherWeWin #AISkillsFest @microsoftlearn.bsky.social
Thanks everyone who helped on that achievement.
#AI #TogetherWeWin #AISkillsFest @microsoftlearn.bsky.social
April 30, 2025 at 5:13 AM
The key to learn Microsoft Dynamics 365 licensing in 10x less time 😉
#msdyn365bc
#msdyn365bc
April 1, 2025 at 6:37 PM
The key to learn Microsoft Dynamics 365 licensing in 10x less time 😉
#msdyn365bc
#msdyn365bc
If you’re flooded with emails like this, please ignore them and press delete before raising alarms… 😵💫
#msdyn365bc
#msdyn365bc
March 19, 2025 at 10:12 PM
If you’re flooded with emails like this, please ignore them and press delete before raising alarms… 😵💫
#msdyn365bc
#msdyn365bc
Another funny #AI test: left to right 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗼𝗻𝗻𝗲𝘁 𝟯.𝟱 vs 𝗢𝟭 (𝗽𝗿𝗲𝘃𝗶𝗲𝘄) vs 𝗢𝟯-𝗺𝗶𝗻𝗶 vs 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸 𝗥𝟭 writing a p5.js script that generates 100 colored balls bouncing inside a rotating sphere ensuring collision detection. Results below. For me 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗼𝗻𝗻𝗲𝘁 𝟯.𝟱 is still the best for coding in general🥇
February 5, 2025 at 4:12 PM
Another funny #AI test: left to right 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗼𝗻𝗻𝗲𝘁 𝟯.𝟱 vs 𝗢𝟭 (𝗽𝗿𝗲𝘃𝗶𝗲𝘄) vs 𝗢𝟯-𝗺𝗶𝗻𝗶 vs 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸 𝗥𝟭 writing a p5.js script that generates 100 colored balls bouncing inside a rotating sphere ensuring collision detection. Results below. For me 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗼𝗻𝗻𝗲𝘁 𝟯.𝟱 is still the best for coding in general🥇
#AI model distillation explained visually😆: Full 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸 𝗥𝟭 𝘃𝘀 𝟯𝟮𝗕 result on the following task: writing an HTML JavaScript app that shows a ball bouncing inside a spinning hexagon, more realistically as possible.
Left DeepSeek R1, Right DeepSeek R1:32B response. Who is the winner?😜
Left DeepSeek R1, Right DeepSeek R1:32B response. Who is the winner?😜
February 1, 2025 at 2:20 PM
#AI model distillation explained visually😆: Full 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸 𝗥𝟭 𝘃𝘀 𝟯𝟮𝗕 result on the following task: writing an HTML JavaScript app that shows a ball bouncing inside a spinning hexagon, more realistically as possible.
Left DeepSeek R1, Right DeepSeek R1:32B response. Who is the winner?😜
Left DeepSeek R1, Right DeepSeek R1:32B response. Who is the winner?😜
No I mean that when you run it it uses the Mac GPU at max:
January 29, 2025 at 7:58 PM
No I mean that when you run it it uses the Mac GPU at max:
💣 🔥 Here it is: a fully local web research assistant that uses a local #LLM for deep search. Give it a topic, it generates a web search, summarize the results, reflect on them and improve the summary for a user-defined number of cycles.
Because #AI has no limits 🚀
#msdyn365bc
Because #AI has no limits 🚀
#msdyn365bc
January 27, 2025 at 3:23 PM
💣 🔥 Here it is: a fully local web research assistant that uses a local #LLM for deep search. Give it a topic, it generates a web search, summarize the results, reflect on them and improve the summary for a user-defined number of cycles.
Because #AI has no limits 🚀
#msdyn365bc
Because #AI has no limits 🚀
#msdyn365bc
#PrivateLLM: Microsoft Phi-4 (left) vs Llama 3.2 (right) writing #AL code fully offline.
Both lightweight, both powerful, Phi-4 is better on coding.
#msdyn365bc #AI
Both lightweight, both powerful, Phi-4 is better on coding.
#msdyn365bc #AI
January 20, 2025 at 2:43 PM
#PrivateLLM: Microsoft Phi-4 (left) vs Llama 3.2 (right) writing #AL code fully offline.
Both lightweight, both powerful, Phi-4 is better on coding.
#msdyn365bc #AI
Both lightweight, both powerful, Phi-4 is better on coding.
#msdyn365bc #AI