Chorouk Malmoum

Two Anthropic engineers will teach you more about building AI Agents in 16 minutes than most paid courses ever will Barry Zhang and Mahesh Murag built Claude Skills at Anthropic In this talk, they lay out the full architecture and the thinking behind it 📌 The core insight is simple : Today's AI agents are like a genius with no work experience They can reason, write code, call APIs But they don't know YOUR processes, YOUR tools, YOUR domain Intelligence ≠ expertise So Anthropic stopped building separate agents for every use case They built one general agent + a skill system instead A skill is literally a folder Markdown instructions. Scripts. Assets. That's it 📌 Here's what they cover: → Why code is the universal interface for agents, not custom scaffolding → How skills are progressively loaded to protect the context window → The split between MCP (access to external tools) and Skills (expertise) → How Fortune 100 companies are encoding internal workflows as Skills → Why non-technical teams in finance, legal, and recruiting are building skills too The architecture they're converging on : Agent loop + Runtime + MCP servers for connectivity + Skills for expertise Deploying Claude to a new domain = the right MCP servers + the right skill library They already used this to launch Claude for financial services and life sciences 🔗 Full talk: https://lnkd.in/djx2SiPd Want more AI guides and updates like these ? 📩 Join 30K builders and execs who never miss an AI agent update : → Weekly AI agent briefings → Free templates & exclusive giveaways 👉https://lnkd.in/dvvUD3hA

Day 1&2 at VivaTech 2026. I'm still processing it... I walked into Hall 7 yesterday's morning not really knowing what to expect By noon my feet hurt, my phone was at 30%, and my notebook was full Here's the thing nobody prepares you for: the scale. The travelator just keeps going. The ceilings are impossibly high! You turn a corner expecting one more booth and there's an entire second world behind it But that's not what I'll remember... What I'll remember is the conversations  The founder who explained his entire product to me in 90 seconds because he only had 90 seconds The engineer demoing something he'd clearly been up until 3am fixing The stranger in the coffee line who turned out to be building the exact thing I'd been thinking about for months This is the part of tech that doesn't fit in a screenshot The energy of thousands of people who actually build things in the same room, all slightly sleep-deprived, all a little obsessed I came here thinking I'd be observing. Instead I spent the whole day talking. Two more days to go and I already don't want it to end ! If you're here too, come say hi. I'm the one with the dying phone and the full notebook👋 #VivatechAmbassador

Anthropic reveals their plan to get Fable 5 back: A new UI This is a meme, but the real Fable 5 story is more interesting than the joke It's Anthropic's first publicly-released Mythos-class model, built for multi-day agentic work, and notably it's the safeguarded twin of Mythos 5 The actual debate worth having is the new mandatory 30-day data retention." That turns a defensive note into a credibility play

Stop building AI demos that fall apart in production. This book shows you how prompting, RAG, and Agents actually work together + FREE Code repo 📖 "Building Agent-Powered Applications" by Vasyl Zvarydchuk Written by a Microsoft Principal AI Engineer, the end-to-end guide to shipping production-ready generative AI What you'll master: 1️⃣ 𝗟𝗟𝗠 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀 - Transformers, embeddings, and inference - How LLMs actually behave - NLP and core concepts - The building blocks before you build 2️⃣ 𝗣𝗿𝗼𝗺𝗽𝘁𝗶𝗻𝗴 & 𝗡𝗟𝗣 𝗧𝗮𝘀𝗸𝘀 - Prompt engineering for reliable output - Summarization and classification - Information extraction - Reasoning workflows 3️⃣ 𝗥𝗔𝗚 & 𝗙𝗶𝗻𝗲-𝗧𝘂𝗻𝗶𝗻𝗴 - Build RAG pipelines that improve accuracy - Grounding responses in real data - When to use RAG vs fine-tuning - Making the right architectural call 4️⃣ 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 - Design agents with tools and memory - Planning and orchestration - Human-in-the-loop controls - Coordinating complex tasks 5️⃣ 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 & 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 - Offline and online testing - Task-specific metrics - LLM-as-a-judge methods - Responsible AI checks 6️⃣ 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗗𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁 - Reduce failure modes - Build scalable systems - Smart architectural trade-offs - Trustworthy AI applications ✨ Bonus: → End-to-end production perspective → Real-world exemples with free github repo → Purchase includes FREE PDF eBook 👉 Get the book: https://amzn.to/4w8Kkgt 🔗 Github repo: https://lnkd.in/eZjdkDnd