Model Cards

Anthropic produce model cards for all their new models. They are well worth a read.

Claude Haiku 5.5 (October 2026) is formalised here. The card reports that no new RSP thresholds have been crossed.

Humanity's Last Exam is referenced in the cards.

The Azure.AI.Projects Client Library

The Azure.AI.Projects client library simplifies management and deployment of AI models and agents within the Azure ecosystem. It is fundamental to Microsoft Foundry.

 

If not already added, adding this library will add the following references:

System.IO.Hashing  (for non cryptographic hash values)

System.Security.AccessControl (for ACLs etc)

System.Security.Principal.Windows (for interacting with Windows users and groups)

Plus a load more.

Microsoft Visual Studio October 2026 Update

The October 29th update includes a number of security updates.

This includes a security fix for the Brotli decoder. 

This refers to Google's open source Brotli compression library, which implements a lossless compression algorithm, that uses, among other things, LZ77 (Lempel Ziv coding). The compressed data format is also a formal, Internet RFC - RFC 7932 - Brotli Compressed Data Format. It is used to make websites load faster by compression http content.

Google Antigravity

Google Antigravity is Google's solution for agentic development.

Rough Guide to Microsoft Teams

A few features in Microsoft Teams that are worth noting.

If the meeting is recorded/transcripted then it appears in chat - where it can be downloaded.

Microphone can be tested from audio settings.

To get joining info click on Meeting Info.

Interacting in the meeting:

Click on the hand, or Control-Shift-K, to raise or drop your hand.

You can use reactions (Like, Love) by clicking the Smiley Face.

Meeting host:

Can set a countdown timer in the meeting by clicking on Timer (defaults to 5 minutes)

So what is ONNX?

ONNX (Open Neural Network Exchange), pronounced "Onyx",  is the closest thing the AI industry has to being a "portable executable" for machine learning models.

ONNX Runtime is a success story of ONNX, and GitHub code is here.

You will see references to ONNX in disucssions on MAF, so it's important to know what it is.

Microsoft Agent Framework - Complete

So here it is .. Microsoft Agents

Microsoft Agent Framework can be studied here.

You can also go deeper into Agents and Providers.

Bear in mind AI agentic frameworks are subject to evolution and change. What you learn today, you may have to re-learn in future. What you write today, you may have to re-code in future. If you are ok with that, proceed.

What you get when you install MAF

If you nuget in the Microsoft.Agents.AI framework into your project, you'll get a load of references to stuff like:

Microsoft.Bcl.HashCode
Microsoft.Bcl.Numerics
Microsoft.Bcl.AsyncInterfaces

Microsoft.Extensions.Compliance.Abstractions

Microsoft.Extensions.FileSystemGlobbing

Microsoft.Extensions.AI

Microsoft.Extensions.AI.Abstractions

Google.Protobuf

Microsoft.Extensions.Primitives

System.Numerics.Vectors

System.Numerics.Tensors

System.Text.Json

System.IO.Pipelines

Microsoft.Agents.AI.Abstractions

Microsoft.ML.Tokenizers

A bunch of these require you to accept the MIT license. These include Microsoft.Bcl.HashCode, System.Numerics.Vectors, and Microsoft.Extensions.AI.