Subscribe, if this was forwarded to you

Hey there,

This week’s newsletter brings a sharp focus on practical tools and cutting-edge progress across the AI landscape. We’re unpacking how to use Claude 4 effectively with tips straight from Anthropic, and we take a look at Google DeepMind’s new Veo 3, a major step forward in AI-generated video.

Meanwhile, Mistral is moving fast—launching new tools like Devstral, Document AI, and an Agents API—proving that Europe is picking up serious speed in the AI race. Plus: Google experiments with diffusion models for text, Anthropic releases tools for tracing LLM internals, and Odyssey rethinks video as something you can interact with.

Lots to learn, lots to build—enjoy the read!

Cheers, Andreas & Sascha

In today's edition:

📚 Tutorial of the Week: Get hands-on with Claude 4

🛠️ Tool Spotlight: Veo 3 – Google DeepMind’s latest leap in high-quality, audio-synced video generation

📰 Top News: Mistral ramps up – Europe joins the AI fast lane

💡 Tip: NLWeb – Turn any website into a conversational interface

Let's get started!

Find this Newsletter helpful?
Please forward it to your colleagues and friends - it helps us tremendously.

Tutorial of the week

How to Get the Most Out of Claude 4 – Tools, Practices & Daily Use

Anthropic recently released Claude 4, the company’s most capable assistant to date—comprising the two flagship models Opus 4 and Sonnet 4. While much of the spotlight has been on performance benchmarks, Anthropic has also released a set of practical resources to help users get immediate value from Claude in real-world tasks.

The focus? Getting better outputs by using better inputs.

🔧 What You’ll Learn:

In their official walkthrough video, Anthropic engineers demonstrate how they use Claude 4 in their daily work—from code reviews and project planning to writing, research, and even team retros. They show how the assistant isn’t just good—it’s useful, when prompted well.

To help you do that, Anthropic also published a detailed Prompt Engineering Guide, which includes best practices like:

• ✅ Give Claude context and intent, not just tasks

• ✍️ Provide structured examples that match your desired format or tone

• 🧠 Use chain-of-thought prompting to encourage deeper reasoning

• 🧩 Format with XML tags if you need consistent structured output

• 🎯 Set expectations clearly—what should Claude do, and why?

And if you're curious about what powers today’s top models, the conversation in this presentation reflects on the evolution of LLMs over the past year. It gives useful insight into how architectural shifts and training choices affect what models like Claude can actually do—and why prompt quality matters more than ever.

🧰 Resources to Get Started:

• 📘 Claude 4 Prompt Engineering Best Practices

• 🎥 Claude 4 Official Product Walkthrough

• 🎥 What Powers State-of-the-Art LLMs

If you want to get real work done with Claude 4, this material is a great starting point—straight from the team that built it.

Tool of the week

Veo 3 — Google DeepMind’s Leap in AI Video Generation

Google DeepMind has unveiled Veo 3, a state-of-the-art AI video generation model that transforms simple text or image prompts into high-quality videos with synchronized audio, including dialogue, sound effects, and ambient noise.

🎬 Key Features:

• 4K Video Output: Generates ultra-high-definition videos with realistic physics and motion.

• Native Audio Generation: Creates synchronized audio tracks, enhancing the storytelling experience.

• Enhanced Prompt Adherence: Improved accuracy in following user instructions for more consistent results.

• Creative Control: Offers tools for adjusting lighting, adding special effects, and fine-tuning character movements.

Veo 3 is accessible through the Gemini app for users subscribed to the Google AI Ultra plan.

Top News of the week

Mistral AI Accelerates Europe's AI Momentum with a Suite of Innovative Tools

French AI startup Mistral AI is rapidly closing the gap with U.S. tech giants by launching a series of groundbreaking tools that position Europe as a formidable player in the AI landscape.

🔍 Document AI Platform

Mistral's new Document AI platform offers enterprise-grade document processing with over 99% OCR accuracy. It efficiently handles complex documents, including handwritten notes and intricate layouts, transforming them into structured formats like JSON.

🧠 Agents API

The Agents API enables developers to build autonomous AI agents equipped with persistent memory and integrated tools for tasks like code execution, web search, and image generation. This framework simplifies the creation of multi-agent workflows, enhancing AI's practical applications in business environments.

💻 Codestral Embed

With Codestral Embed, Mistral introduces a state-of-the-art code embedding model that outperforms competitors like OpenAI and Cohere in real-world retrieval tasks. This tool enhances code search and retrieval-augmented generation (RAG) applications, streamlining software development processes.

🛠️ Devstral

In collaboration with All Hands AI, Mistral released Devstral, an open-source large language model tailored for software engineering. Devstral excels in automating complex coding tasks and outperforms several existing models on benchmarks like SWE-Bench Verified.

These developments underscore Mistral AI's commitment to advancing AI capabilities and highlight Europe's growing influence in the global AI arena.

Anthropic Open-Sources Circuit Tracing Tools

Anthropic has released a set of open-source tools designed to help researchers and developers better understand how specific neurons and circuits influence language model behavior. These tools make it easier to trace and visualize how models arrive at their outputs—bringing a new level of transparency to LLM internals.

Google DeepMind Releases Gemini Diffusion

Gemini Diffusion is an experimental new language model based on diffusion techniques—typically used in image generation—now applied to text. The approach aims to offer users greater control, faster iteration, and more creativity in how text is generated.

DeepSeek R1 Gets Smarter and More Reliable

DeepSeek has released an upgraded version of its R1 model with significant improvements in reasoning, especially in coding and math. The update reduces hallucinations and boosts accuracy, making it more suitable for production use in complex technical workflows.

Odyssey Debuts Interactive AI-Generated Video

Odyssey has launched a new video format that combines storytelling with real-time interactivity. Viewers can pause, ask questions, or change the direction of the story—all powered by an underlying AI system that adapts as you engage.

Google Introduces Gemma 3n – A Mobile-First AI Model

Google DeepMind has unveiled Gemma 3n, a compact, high-performance language model designed specifically for mobile and edge devices. Now in early preview, it brings powerful on-device AI capabilities with state-of-the-art efficiency and low latency.

Tip of the week

NLWeb – Turn Any Website into a Conversational Interface

Microsoft just introduced NLWeb, a new open project that brings conversational interfaces directly to the web—no custom UI, wrappers, or complicated infrastructure needed. With NLWeb, any site can become queryable via natural language, making your content discoverable and usable by LLMs and AI agents.

It’s a big move toward what Microsoft calls “the agentic web,” and it’s all built around open standards and the Model Context Protocol (MCP).

💡 Why it matters:

• You can turn your website into an AI-native experience—where users (or agents) ask questions and get structured answers.

• NLWeb is platform-agnostic and supports all major models and vector databases.

• It serves as an MCP server, so your site becomes immediately usable by any assistant or agent that speaks MCP.

🧰 How it works:

• Uses existing web standards (like Schema.org & RSS) + structured APIs to surface content

• Can return structured responses (JSON) or rich documents, depending on what agents need

• Works out-of-the-box with sites like TripAdvisor, Shopify, and Eventbrite already testing it

If you're building agent-facing tools or just want to future-proof your content, this is worth exploring.

🔗 Read the full announcement

The web is getting conversational—now’s the time to make your site part of it.

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