Le plan machiavélique d'Anthropic (et comment s'en protéger)

4 weeks to build your AI Second Brain. A sovereign system that knows you, amplifies your actions, and becomes an extension of yourself. Join the Prisme One bootcamp: https://prisme.one/bootcamp-ia --------------- Every week, Anthropic releases a new product. Cloud Code, Cloud Cowork, plugins, routines, Conway. Impressive, you say? I find it scary. In this video, I dissect the strategy Anthropic has been implementing for the past six months to create the largest proprietary lock-in in tech history. Not the data they store—your way of working, your cognitive structure, your habits. All in formats impossible to export. I show you the proof (the leak of 512,000 lines of code), the concrete mechanisms, and above all, how to protect yourself with a sovereign context. What you'll learn: → Anthropic's timeline decoded: why each release is a step in a single plan (Cloud Code → Cowork → Plugins → Conway) → Conway: the permanent agent that learns your behavior, your tastes, your cognition in a proprietary format → Why the real lock-in isn't your data but your behavioral model (and why it's much worse) → What this means for employees: your boss can squeeze every last drop out of you and keep all your added value → The €100 subscription scam: why it's worth €3,000 in tokens and how it serves lock-in → Open-source models (DeepSeek V4 Pro) do the same thing for 30x less — live demo → How to build a sovereign Markdown context that survives any provider change --- Resources mentioned : Fireworks AI (open-source models with ZDR): https://fireworks.ai DeepSeek V4 Pro: https://deepseek.com OpenCode (multi-model interface): https://opencode.ai --- About: My name is Eliott Meunier, and I've been working on personal knowledge management since 2020. I created Atomic Thinking, the leading method in France for the second brain (over 15,000 people, author of a best-selling book published by Eyrolles). Today, I help freelancers build their AI Second Brain: the system that knows you, learns over time, and makes AI truly useful. --- TIME CODES: 00:00 Anthropic is riding high (but it scares me) 00:59 The release schedule decoded 01:15 Cloud Code: the terminal harness 02:16 Cloud Cowork: the consumer version 03:05 Proprietary plugins (goodbye MCP) 04:14 The 512,000-line code leak 04:49 The war against OpenClaw 05:46 Channels, Routines: centralizing everything 06:27 The Apple vs. Android analogy 07:02 Conway: the permanent agent that learns everything 07:40 What Conway records about you (4 levels) 09:04 The proprietary behavioral model 09:47 The cost of impossible migration 10:53 Why the subscription is so cheap 11:54 How to protect yourself: sovereignty 12:07 Level 1: employee sovereignty (the (trap) 13:04 Level 2: Context sovereignty 13:44 Recommendation: Markdown files 15:12 Live demo: model switching mid-conversation 17:09 Context sovereignty + multi-model = resilience 17:40 Opus $25 vs. DeepSeek $0.80 (27 times cheaper)