{"id":3133,"date":"2025-09-06T20:08:39","date_gmt":"2025-09-06T19:08:39","guid":{"rendered":"https:\/\/www.ituziast.com\/?p=3133"},"modified":"2026-01-28T10:30:23","modified_gmt":"2026-01-28T09:30:23","slug":"visual-studio-code-dev-days-2025-skopje","status":"publish","type":"post","link":"https:\/\/www.ituziast.com\/index.php\/2025\/09\/06\/visual-studio-code-dev-days-2025-skopje\/","title":{"rendered":"Visual Studio Code Dev Days 2025 Skopje"},"content":{"rendered":"\n

As part of the joint effort of the Macedonian .NET<\/a> and ITPro<\/a> User Groups, this year we took part in VS Code Dev Days<\/a>. This is a global initiative to help developers learn and up-skill in AI-assisted development using Visual Studio Code. After all, in September, there are over 60 events this year, all around the world. The events are organized by Microsoft, MVPs, Student Ambassadors, and GitHub communities.<\/p>\n\n\n\n

The ‘usual suspects’ for this event were Vlatko Ivanovski (MVP-DevOps, Moderator)<\/a>, Miodrag Cekikj (MVP-AI Platform)<\/a>, Miroslav Janeski(MVP-Microsoft Azure, Developer Technologies)<\/a> and Dimitar Grozdanov (MVP-Microsoft 365, ITPro User Group)<\/a>. The event was held at the hacker space Base42<\/a> and it was packed. <\/p>\n\n\n\n

One of the reasons being, that our guest speaker was Rade Trimceski, Corporate Vice President, Engineering Systems \u2013 Azure, Windows & Linux<\/a> at Microsoft. He went full demo mode in the session – showing how Microsoft uses AI-assisted tools in their daily operations, mono\/micro repos, some internal tailor-made systems and took questions from the attendees.<\/p>\n\n\n\n

\n
\"\"<\/figure>\n\n\n\n
\"\"
Vlatko Ivanovski<\/figcaption><\/figure>\n\n\n\n
\"\"
Opening<\/figcaption><\/figure>\n\n\n\n
\"\"
Rade Trimceski<\/figcaption><\/figure>\n\n\n\n
\"\"
IT Crowd (1)<\/figcaption><\/figure>\n\n\n\n
\"\"
IT Crowd (2)<\/figcaption><\/figure>\n<\/figure>\n\n\n\n
\n

Note: Due to the nature of the systems showed, no recording\/photos are available from the session content. Some reference materials are shared, which are publicly available.<\/p>\n<\/blockquote>\n\n\n\n

After the session we sat down with Rade, and asked him the ‘tough’ questions. \ud83d\ude42<\/p>\n\n\n\n

The Interview<\/h2>\n\n\n\n

Q: You’ve had a remarkable journey in software engineering. From your early days in the NT kernel team to leading engineering systems across Azure, Windows, and Linux. How has your perspective on developer tooling and productivity evolved over the years?<\/em><\/h5>\n\n\n\n

Developer tools that improve productivity have been advancing through the years at a pretty good pace in the 20 years I\u2019ve been working on large-scale software projects. But the dawn of generative AI has dramatically accelerated evolution in developer tools and their effects on productivity.<\/p>\n\n\n\n

So I would say that of the last 20 years, the first 18 we saw linear improvements to developer tools for productivity of developers. In the last 2 years we have seen exponential improvements. And we are just at the beginning of the AI revolution.<\/p>\n\n\n\n

Q: Looking back at your early career, what\u2019s one lesson or mindset that has stayed with you and continues to shape how you lead and innovate today?<\/em><\/h5>\n\n\n\n

It\u2019s the mindset of continuous improvement. Both as a person, as a team and as a company. If you invest every week in becoming better at something, after a few weeks you won\u2019t notice a big difference. But if you consistently do it for months and years, the \u201ccompounding\u201d effect will make you much better.<\/p>\n\n\n\n

This applies to skill improvements (e.g. how good am I at debugging a kernel mode crash), knowledge acquisition (I learn new thing every week), etc.<\/p>\n\n\n\n

Q: Your presentation explores customizing GitHub Copilot with Haystack Search. <\/em>What are some common misconceptions developers face when integrating Copilot into large code-bases, and how can they overcome them?<\/em><\/h5>\n\n\n\n

We are early in the AI journey. For some time we thought that building fine-tuned models for large code-bases is the best way to improve the quality of GitHub Copilot responses in large code-bases, think Windows OS code-base.  <\/p>\n\n\n\n

What we discussed in the talk is using GitHub\u2019s custom instructions and prompts capabilities to supercharge the relevance of GitHub\u2019s copilot\u2019s answers in large code-bases, like the Windows OS code-base.  <\/p>\n\n\n\n

In addition to custom instructions, we now know that providing a complete, even plain-text, index of the code-base to Copilot also improves the quality of the responses. I mentioned the Haystack Search VS Code extension<\/a> as one plugin that offers a full-codebase index. It\u2019s just one example.  <\/p>\n\n\n\n

Q: You mentioned generating custom instructions for Copilot. How important is it for developers to take ownership of their AI tooling setup? What best practices would you recommend for teams just starting out?<\/em><\/h5>\n\n\n\n

It\u2019s extremely important. AI systems like GitHub Copilot are a tool. And like with every tool you must learn how to use it and adapt it to your needs. Otherwise, you are not getting the best use out of the tool. <\/p>\n\n\n\n

At this talk we covered providing custom instructions<\/a> to GitHub Copilot. Best practices: <\/p>\n\n\n\n