Subscribe, if this was forwarded to you

Hey there,

This week’s edition is all about practical AI engineering — from Microsoft’s new Agent Framework tutorial to Anthropic’s Skills and Cookbooks that make Claude more capable and customizable than ever. We’re also spotlighting RAG-Anything, a flexible framework for multimodal retrieval, plus several key developments shaping the AI ecosystem — from OpenAI’s Titan chips to Anthropic’s Microsoft 365 integration.

Plenty to explore — enjoy the read!

Cheers, Andreas & Sascha

In today's edition:

📚 Tutorial of the Week: Build your first agent with Microsoft’s new Agent Framework — a hands-on guide from the “AI Agents for Beginners” series.

🛠️ Tool Spotlight: RAG-Anything — an all-in-one framework for multimodal retrieval and document understanding.

📰 Top News: Anthropic launches Skills, modular expertise packs that turn Claude into a specialist agent.

💡 Tips: Get started with Claude Cookbooks — practical, open-source recipes for real-world AI development.

Let's get started!

Tutorial of the week

Hands-On with the Microsoft Agent Framework

Microsoft’s latest official tutorial, part of the “AI Agents for Beginners” series, takes you step-by-step through building, orchestrating, and deploying AI agents using the Agent Framework.

You’ll learn to:

• Set up your development environment with Python or .NET, Azure OpenAI, and tailor the Agent Framework to your stack.

• Construct multi-step, agent-based workflows by defining goals, sub-agents, sensors, actuators and memory layers (a core concept from the course).

• Deploy and monitor live agents with built-in governance, telemetry, and lifecycle controls — moving beyond prototyping into production-grade infrastructure.

Why it matters: This is your chance to go from reading about the Agent Framework’s announcement to hands-on building. Whether you’re a developer experimenting or a team looking to embed agentic workflows in business systems, the guide gives you the tools to do — not just theorize.

Start the tutorial →
Explore the full Agent Framework repo →

Tool of the week

RAG‑Anything — An All-in-One RAG Framework for Multimodal Content

If your data isn’t just text — but also images, tables, formulas, and slides — RAG-Anything is worth a close look. It provides a unified pipeline for document ingestion, parsing, retrieval, and query-response across formats, built with modular parser and controller components.

Key strengths include:

• Versatile format support: PDF, Office docs (.docx, .pptx, .xlsx), images, markdown, tables, equations.

• Unified multimodal retrieval: Combines vector-search, knowledge-graph links, and modality-aware ranking for seamless "mixed content" queries.

• Rapid setup & real-world examples: Install via pip install raganything[all] or clone examples for document-to-query workflows.

Why it matters:
Many retrieval-augmented systems still struggle with non-text elements (images, tables, math) — RAG-Anything bridges that gap. If you’re building a knowledge base from rich, heterogeneous documents, this tool could drastically reduce boilerplate and accelerate delivery of insight.

Check it out on GitHub →

Top News of the week

Anthropic introduces “Skills” — modular expertise for Claude

Anthropic unveiled Skills, a major new feature that lets Claude access modular, reusable knowledge packages for specialized tasks. Each skill is a structured bundle of instructions, examples, and resources that Claude can dynamically load — for instance, to write SQL queries, design branded documents, or format spreadsheets. This approach moves beyond static prompt templates toward composable expertise that adapts to context.

Skills can be created, shared, and managed across the Claude app, API, or enterprise environments, allowing teams to define consistent, domain-specific behaviors for their assistants. The feature is available to Claude Pro, Team, and Enterprise users and represents Anthropic’s biggest step yet toward scalable, maintainable AI workflows.

Read the full announcement →

OpenAI accelerates hardware push with in-house “Titan” XPU via Broadcom

OpenAI and Broadcom have entered a multi-year collaboration to co-design and deploy “Titan” XPUs: custom-built AI accelerators targeting 10 gigawatts of capacity, with roll-out beginning in late 2026.
The deal signals a strategic shift for OpenAI away from reliance on off-the-shelf GPUs and toward vertically integrated silicon that embeds model logic directly into hardware. For Broadcom, this agreement marks a major leap into custom AI infrastructure—jointly positioning both firms at the center of a reshaping AI-hardware landscape.

Read more →

How are developers using AI? Insights from the 2025 DORA report

Google’s newly released 2025 DORA report dives into developer workflows and how AI is shaping the way teams build software — from generating code snippets to participating in pull-request reviews. The data reveals rising trust in AI tools but also highlights key gaps around reliability, governance, and collaboration.

Read the full report →

Anthropic launches Claude Connector for Microsoft 365

Anthropic introduced a Claude Connector that brings its AI assistant directly into Microsoft 365 apps like Word, Excel, and Outlook. The integration allows users to summarize emails, draft documents, and analyze spreadsheets using Claude’s contextual understanding — offering an alternative to Copilot within the Microsoft ecosystem.

Read more on The Verge →
More details on TestingCatalog →

Andrej Karpathy revives nanoGPT for the modern LLM era

Andrej Karpathy’s nanoGPT project — a minimal, educational implementation of GPT training in PyTorch — is getting renewed attention as a go-to learning reference for LLM fundamentals. Compact yet complete, it walks through data loading, tokenization, transformer architecture, and training loops in under 1,000 lines of code — perfect for experimentation or teaching.

Explore nanoGPT on GitHub →

Tip of the week

💡 Anthropic Cookbooks — practical recipes for building with Claude

Anthropic has released the Claude Cookbooks, an open-source collection of practical guides, notebooks, and code samples for developers who want to get more out of Claude. Think of it as a living handbook for applied AI — full of small, focused “recipes” showing how to use Claude effectively across real-world tasks.

What it is:
A GitHub repository of runnable examples that demonstrate how to integrate Claude into workflows for retrieval-augmented generation (RAG), function calling, data extraction, evaluation, and more. Each notebook is self-contained and written to be easy to extend.

What to expect:

• Clear, minimal code that works out of the box — no heavy frameworks required.

• Step-by-step examples for creating custom tools and agentic behaviors.

• Guidance on prompt design, structured outputs, and evaluation strategies.

Why it matters:
The Cookbooks bridge the gap between documentation and production — helping teams move from concept to working prototypes quickly while keeping implementation transparent and flexible.

Explore the Claude Cookbooks on GitHub →

We hope you liked our newsletter and you stay tuned for the next edition. If you need help with your AI tasks and implementations - let us know. We are happy to help