Building a Multi-Agent System in .NET using the Microsoft Agent Framework, GitHub Copilot SDK, and Entra Agent ID Two agents, two frameworks, one identity layer: how MAF, GitHub Copilot SDK, and Entra Agent ID work together in .NET 10.

New blog post! I've built a multi-agent system into my personal health platform: A conversational Q&A agent over my health data using the Microsoft Agent Framework, and a report generator using GitHub Copilot SDK, with Entra Agent ID handling auth: open.substack.com/pub/willveli...

Finally found a place that sells Heaps Normal through the draft! Lincoln Hotel on Cardigan street

Introducing the Agent Governance Toolkit: Open-source runtime security for AI agents | Microsoft Open Source Blog Discover how the Microsoft Agent Governance Toolkit brings policy, identity, and reliability to autonomous AI agent systems.

Microsoft just launched a new open security and governance framework for AI agents. When I first heard about this project, I immediately saw how impactful it could be. Looking forward to hearing the community’s feedback.
opensource.microsoft.com/blog/2026/04...

Why are Auckland playing a Championship team? 😅

Sack Voss. Sack him now.

Thank you! I've restricted the packages that can run on the Copilot sidecar, and through GHCP I can use Opus to do the report generation, ML analysis, etc.

I use the consumption tier, so for the amount of traffic that actually hits it, it costs nothing to run 🙂

biotrackr/src/Biotrackr.Reporting.Api at main · willvelida/biotrackr My personal health platform. Contribute to willvelida/biotrackr development by creating an account on GitHub.

I wanted to keep these agents separate to avoid giving my Chat agent the ability to execute code (ASI05), and to play around with the GitHub Copilot SDK. If you want to take a look at the moving parts, check out the code on GitHub: github.com/willvelida/b...

Once the report is generated, it's validated by a reviewer agent that either rejects the report due to inaccuracies, or it's passed back to the user. I can then download the report directly from the UI:

Essentially, in the UI I ask the chat agent to generate a report, which will then use the MCP server to retrieve the data to pass to the reporting agent. The reporting agent will then kick off the process of executing Python code to generate graphs and produce a summary report like so:

I've been playing with multi-agentic architectures in my side project. I built a reporting agent that uses the GitHub Copilot SDK to execute python code that generates reports and charts based off the data that gets passed to the agent:

The Emu war isn't looking so bad by comparison now is it?

Baker's Dozen is on in Fed Square and I've never been so full in my life 😅

Flying out of Perth last night had me feeling like this

Dutch comfort food

Flight got delayed by 90 minutes 👍

Visiting Perth has its challenges. Its a 3 hour time difference, so wake up at 6am usually? You're getting up at 3am. Heading back at 7pm? Thats a 2am arrival time for you 😅😅

Wikipedia Bans AI-Generated Content “In recent months, more and more administrative reports centered on LLM-related issues, and editors were being overwhelmed.”

NEW: Wikipedia has banned AI-generated content.

Fixing Claude with Claude: Anthropic reports on AI SRE QCon London: It's still a job for humans, even though bots can search logs at the speed of I/O

Why AI still doesn't replace #SRE engineers (no matter what all those fancy startups tell you) - from a guy who is actually keeping Claude up at #Anthropic:
www.theregister.com/2026/03/19/a...

Zero Trust for AI is here

Start exploring:
→ aka.ms/zerotrust/ass...

Workshop + roadmap:
→ aka.ms/ztworkshop

Bookmark this thread!
8/8

This is a Baklava ice cream sandwhich. Pure diabetes

Preventing Token Mismanagement and Secret Exposure in MCP Servers Implementing OWASP MCP01 mitigations against Token Mismanagement and Secret Exposure in a .NET 10 MCP server built with the Model Context Protocol SDK.

Now that I can use my MCP server in my agent, I'm going to do another OWASP Top 10 series, this time focusing on MCP servers! This first article is about Preventing Token Mismanagement and Secret Exposure in MCP Servers: open.substack.com/pub/willveli...

Kubernetes the (Very) Hard Way (course) | iximiuz Labs A hands-on, step-by-step guide to assembling a Kubernetes cluster from the ground up, (without using any automation) while deeply exploring each component's role and functionality along the way.

🚨 KUBERNETES THE VERY HARD WAY IS COMPLETE!

The complete course is now available on iximiuz Labs: labs.iximiuz.com/courses/kub...

If you want to learn how Kubernetes works under the hood, this course is for you.

From Function Tools to MCP: Migrating an AI Agent to the Model Context Protocol When I first built an AI Agent for my personal health platform Biotrackr, I used the Microsoft Agent Framework’s function tool pattern with 12 tools decorated with [Description] attributes, registered...

The Microsoft.Agents.AI.Anthropic 1.0.0-rc4, you can now use tools hosted in your MCP servers rather than defining local tools. In this blog, I talk about the process of migrating my agent to use MCP tools, and the various debugging adventures I went on! willvelida.substack.com/p/from-funct...

GitHub - indutny/no-ai-in-nodejs-core: A petition to disallow acceptance of LLM generated Pull Requests in Node.js core A petition to disallow acceptance of LLM generated Pull Requests in Node.js core - indutny/no-ai-in-nodejs-core

Will be interesting to see where this goes. I'm not entirely sure that they will be successful, but maybe this will refocus the conversation on quality over vibes: github.com/indutny/no-a...

Probably the best lego set I've gotten to build. Looks very cool!

they literally renamed the entire company after this

Turns out it didn't have legs

Will Velida's Technical Deep Dive | Substack Technical deep dives, in-depth tutorials, and going under the hood on technology. Click to read Will Velida's Technical Deep Dive, a Substack publication. Launched 15 minutes ago.

Finally got around to sorting out my substack! I do deep dives on Microsoft technologies and AI, so if you're interested in that, subscribe here: willvelida.substack.com?r=rbqyl&utm_...

I would 8-2 make fun of Chelsea right now......

I may need to refactor my Weight API so that it can call the API, so if/when I figure that out, I'll let you know.