[{"title":"Portugal Built Its Own AI. I Wanted to Know If It Really Spoke Portuguese to Me.","desc":"Amália: 9 billion parameters, €5.5 million, sixty researchers. A case study for understanding what 'AI sovereignty' means once you drop the speeches and look at the actual model.","category":"IA, Tech, Personal","url":"/en/amalia-portuguese-ai-sovereignty-en/","date":"Jul 21, 2026","content":"There's something people who grew up between two languages know well: the moment a tool makes you feel that one of the two counts for less than the other. The spell checker that underlines your last name in red. The form that rejects accents. And, for the past three years, the chatbot you speak to in Portuguese that answers you in flawless Portuguese… from Brazil. It's not a big deal. It's never a big deal. It's just a small reminder, repeated a thousand times, that your grandparents' language is treated as a variant of something else. So when Portugal unveiled Amália on 1 July 2026 in Lisbon,"},{"title":"Ollama in 2026: From Local Runner to AI Platform","desc":"Models, API, cloud, agents — eighteen months on from the first article","category":"IA, Development, Tech","url":"/en/ollama-2026-state-of-the-art-en/","date":"Jun 17, 2026","content":"Back in January 2025 I wrote a deployment guide for Ollama + Open WebUI focused on Docker and basic configuration. The project worked well, but it was still a runner: download a GGUF, query port 11434, done. Eighteen months and a v0.30.8 release later, the scope has changed enough to make parts of that guide obsolete. Hybrid cloud, Anthropic Messages API compatibility, native web search, agents. Here's where things stand. What Ollama Has Become The definition \"tool for running LLMs locally\" still holds, but covers less and less of the actual product. Since 2026, Ollama is also a cloud proxy: s"},{"title":"Claude Code Cheat Sheet — June 2026: Fable 5, Opus 4.8, Agent View, and Dynamic Workflows","desc":"Exhaustive up-to-date reference for Claude Code v2.1.170 — June 2026","category":"IA, Development, Claude Code","url":"/en/claude-code-cheatsheet-june-2026-update/","date":"Jun 15, 2026","content":"⚠ Update (June 12, 2026): Fable 5 and Mythos 5 were suspended globally by US government order four days after their launch. All other models remain available. Fable 5 entries in the cheat sheet are preserved for reference, marked as suspended. I republished a Claude Code cheatsheet at the end of April. Seven weeks later it has aged a lot: between v2.1.101 and v2.1.170, Anthropic shipped roughly sixty releases. Two new models landed (Opus 4.8 on May 28, then Fable 5 on June 9), the Agent View arrived, dynamic workflows now orchestrate hundreds of subagents, and a handful of new commands ( , , ,"},{"title":"Karpathy's Second Brain: A Wiki the AI Writes For You","desc":"How to turn Obsidian + Claude into a knowledge base that compiles, links and fixes itself, no RAG, no vector database","category":"IA, Claude Code, Development","url":"/en/karpathy-second-brain-obsidian-claude-en/","date":"Jun 10, 2026","content":"We all run a degraded version of the same dream: one single place that would hold everything we've ever read, watched, or learned. A backup brain. In practice, we pile up abandoned Notion pages, bookmarks we never reopen, highlighted PDFs rotting in a folder. The problem was never collecting information. It's organizing it. That's exactly where human effort collapses. In early 2026, Andrej Karpathy (OpenAI co founder, former Tesla AI director) shared a deceptively simple idea on X to solve this: what if we stopped writing the knowledge base ourselves, and handed it entirely to an LLM? He calls"},{"title":"AI sovereignty, a trillion euros and 'vassal state': what Arthur Mensch told the French National Assembly","desc":"A critical read of Mistral AI CEO's hearing before the inquiry commission on digital vulnerabilities","category":"IA, Tech, Personal","url":"/en/arthur-mensch-mistral-ai-national-assembly-hearing-en/","date":"May 15, 2026","content":"Tuesday May 12, 2026, 4:00 PM. Arthur Mensch , co founder and CEO of Mistral AI, is heard by the French National Assembly's inquiry commission \"on structural dependencies and systemic vulnerabilities in the digital sector\". Next to him sits Audrey Herblin Stoop, Mistral's head of public affairs. Across the room, MPs trying to understand where Europe stands in the global AI race. The hearing runs for over ninety minutes. Mensch lays out a built thesis: AI is infrastructure technology, and Europe has a few months to decide whether it produces or merely consumes. As in any such exercise, what he "},{"title":"Claude Code Cheat Sheet — April 2026: Opus 4.7, Auto Mode, and Computer Use in the CLI","desc":"Exhaustive up-to-date reference for Claude Code v2.1.101 — April 2026","category":"IA, Development, Claude Code","url":"/en/claude-code-cheatsheet-april-2026-update/","date":"Apr 24, 2026","content":"I published a Claude Code cheatsheet at the end of March. Less than four weeks later it is already incomplete: Opus 4.7 shipped on April 16, Computer Use landed in the CLI, Auto Mode went GA on Max, seven new slash commands appeared (and two were removed), five hook updates arrived, and the Desktop app was redesigned on April 14. Between v2.1.83 and v2.1.101, Anthropic shipped roughly thirty releases in five weeks. So I republished an April 2026 edition — same visual architecture as the March version, rewritten to match every recent addition, and using the blog palette (dark background, gold a"},{"title":"Understanding, measuring and controlling LLM hallucinations","desc":"Notes from Aygalic Jara's talk at Devoxx France 2026","category":"IA, Development, Tech","url":"/en/llm-hallucinations-devoxx-2026-en/","date":"Apr 23, 2026","content":"Thursday 23 April 2026, 3:40 PM, Palais des Congrès. Aygalic Jara , consultant and PhD candidate at SCIAM under a CIFRE agreement with LISN (Laboratoire Interdisciplinaire des Sciences du Numérique), gave a talk whose title summed up the whole programme: \"LLMs and hallucinations: understanding, measuring, controlling\" . Room packed, topic hot. It sits at the core of his PhD work, of which a first state of the art paper was published in 2025 at the JEP TALN RECITAL conference under the title \"State of the art: evaluation, detection and mitigation of LLM hallucinations\" . LLM hallucinations are "},{"title":"SDD, Compound Engineering, BMAD: Which AI Development Philosophy Should You Choose?","desc":"Mapping structured approaches for AI-assisted development — and why combining them is probably the right answer","category":"IA, Development, Claude Code","url":"/en/sdd-compound-engineering-bmad-philosophies-en/","date":"Apr 4, 2026","content":"Vibe coding democratized AI assisted development. But by 2026, the verdict is unanimous: it doesn't scale . Serious projects demand more structure. Three families of approaches have emerged to fill that gap — each tackling the problem from a different angle. This article maps these approaches comparatively, with concrete guidance on how to choose — or combine — them. The problem: why vibe coding hits a wall The term vibe coding, popularized by Andrej Karpathy in early 2025, describes a workflow where the developer delegates code generation to an AI via natural language, without formally struct"},{"title":"Claude Code Cheat Sheet 2026: Every Command, Shortcut & Feature in One Place","desc":"An exhaustive, up-to-date reference for Claude Code v2.x — March 2026","category":"IA, Development, Claude Code","url":"/en/claude-code-cheatsheet-2026-update/","date":"Mar 31, 2026","content":"Claude Code evolves fast. Most cheatsheets available online are already missing features that shipped in early 2026. I built an exhaustive, visual reference covering everything in Claude Code v2.x as of March 2026 — and I'm sharing it for free. The Cheatsheet Available in both English and French: English — Interactive HTML French — Interactive HTML You can also download the high res PNGs: English French ( 4MB each, 2800px wide). Items marked with a green NEW badge are features added since late 2025. What's Covered This isn't a \"top 20 commands\" summary. It's the full reference: Section Coverag"},{"title":"/insights: The Command That Analyzes How You Code with Claude","desc":"Discover how Claude Code generates a comprehensive report of your development habits","category":"IA, Développement","url":"/en/claude-code-insights-command/","date":"Mar 5, 2026","content":"Announced in early February 2026 by Anthropic's Thariq Shihipar, the command is one of the latest additions to Claude Code. The concept: analyze your last 30 days of sessions and generate a detailed report of your habits. Recurring patterns, friction points, inefficient workflows — nothing escapes it. What is /insights? is a built in Claude Code command that analyzes your local session history and produces an interactive HTML report . No configuration needed, no external tracking: everything is based on data already stored on your machine. The report is generated at and opens automatically in "},{"title":"Sourdine: Meeting Transcription with 100% Local AI","desc":"An open source macOS app to transcribe your meetings without sending your data to the cloud","category":"IA, Développement, Projet, Open-Source","url":"/en/sourdine-transcription-reunions-ia-locale-en/","date":"Feb 15, 2026","content":"How many meetings do you have this week? And how many notes did you actually take? Between actively participating and taking detailed notes, you have to choose. This frustration led to Sourdine — with one major constraint: everything must stay on my machine . The Problem: Cloud AI vs Privacy Transcription solutions (Otter.ai, Fireflies, Teams/Meet transcriptions) share one thing in common: your conversations go through their servers . Problematic for: Strategic business meetings Confidential interviews Medical or legal discussions Anything that's none of Google, Microsoft, or a startup's busin"},{"title":"Human and Machine: Entropy, Intelligence, and What Still Sets Us Apart","desc":"When thermodynamics reveals why AI may never be like us","category":"IA, Personnel","url":"/en/entropy-human-machine/","date":"Feb 8, 2026","content":"Introduction: A Universal Law Nobody Explained to You There's a physical law you apply every day without knowing it. It explains why your desk gets messy, why your coffee cools down, why your projects drift, why your teams become disorganized, and perhaps even why you sometimes feel overwhelmed by your emotions. This law is the second law of thermodynamics : the entropy of an isolated system can only increase. In other words: without effort, everything tends toward disorder. But entropy isn't really \"disorder\" in the sense of a messy room. It's about the number of possible configurations [2]. "},{"title":"Ollama Now Supports Claude Code","desc":"Use Claude Code with local models thanks to Ollama's new Anthropic API","category":"IA, Développement","url":"/en/claude-code-ollama-local-llm-en/","date":"Jan 23, 2026","content":"Big news for Claude Code users: Ollama now supports the Anthropic Messages API , which allows you to use Claude Code with local open source models. No more exclusive dependency on Anthropic's cloud! Why This Integration Is a Game Changer Until now, Claude Code required a connection to Anthropic's servers. With this Ollama integration, you can now: Benefit Description Privacy Your code stays on your machine Costs No API fees, just your electricity Independence No single vendor lock in Offline Work without internet connection Customization Choose the model that fits your needs Requirements 1. Ol"},{"title":"Will AI Replace Developers? A Critical Analysis of Promises and Limitations","desc":"Deconstructing the prevailing narrative on AGI and the replacement of developers by artificial intelligence","category":"IA, Développement","url":"/en/will-ai-replace-developers-critical-analysis/","date":"Dec 31, 2025","content":"Panic is palpable on social media. Every new demo of a generative AI model triggers a wave of catastrophic predictions: \"It's the end for developers\", \"AGI is coming in 2 years\", \"A team of 6 developers replaced by just one with AI\". These claims deserve rigorous analysis, far from the collective hysteria. This article offers a methodical deconstruction of the dominant narrative about AI and developer replacement, drawing on recent scientific studies, economic data, analyses from researchers like Tim Dettmers (Ai2) on hardware physical limits, and a relevant video analysis by Melvynx — a Frenc"},{"title":"Case Study: Prelude of the Chambered Reborn","desc":"Day 20 - How I converted a Java game to TypeScript in one evening","category":"IA, Développement","url":"/en/claude-code-case-study-prelude-en/","date":"Dec 30, 2025","content":"To conclude this series, what better than a concrete experience report? Here's how I used Claude Code to convert Prelude of the Chambered , Notch's game (Minecraft creator), from Java to TypeScript — in a single evening . To discover the project in detail, read my dedicated article: Prelude of the Chambered Reborn: Renaissance of a Notch classic Also see my other retro game project in TypeScript: GBTS A GameBoy Emulator The Challenge The Original Project Prelude of the Chambered is a dungeon crawler created by Notch in 48h during Ludum Dare 21 in 2011. The original code is in Java, with a home"},{"title":"Claude Code vs Cursor vs GitHub Copilot","desc":"Day 19 - Comparison of AI assistants for developers","category":"IA, Développement","url":"/en/claude-code-vs-cursor-vs-copilot-en/","date":"Dec 29, 2025","content":"The AI assistant market for developers is exploding. Claude Code, Cursor, and GitHub Copilot are the three major solutions. Which one to choose? Let's compare objectively. For broader context on the state of AI in development, see my article AI and Development in 2025. And if you're considering local alternatives, check out OpenAI GPT OSS vs Ollama. Overview Criteria Claude Code Cursor GitHub Copilot Type CLI + Agent Full IDE IDE Extension AI Model Claude (Anthropic) Multiple (including Claude) GPT 4/Codex Price $100 200/month (Max) $20/month $10 19/month Platform Terminal VS Code Fork VS Code"},{"title":"Status Line and Terminal Customization","desc":"Day 18 - Configure Claude Code display","category":"IA, Développement","url":"/en/claude-code-status-line-terminal-en/","date":"Dec 28, 2025","content":"Claude Code runs in your terminal. Did you know you can customize its appearance and behavior? Let's see how to adapt the interface to your preferences. The Status Line What is the status line? The bar at the bottom of the screen that displays: Active model Tokens used Session cost Connection status Customize the status line Via the command: Status line options: Minimal : Just the model Standard : Model + tokens Detailed : Everything (model, tokens, cost, latency) Hidden : No status line Configuration in settings Vim Mode Enable vim mode Or at launch: Available vim shortcuts Mode Key Action No"},{"title":"Claude Code Troubleshooting","desc":"Day 17 - Solving common problems","category":"IA, Développement","url":"/en/claude-code-troubleshooting-en/","date":"Dec 27, 2025","content":"Claude Code is robust, but problems can occur. Here's a complete guide to diagnose and solve the most common errors. Connection Problems Error: \"API Key invalid\" Solutions : 1. Verify the key: 2. Reconfigure: 3. Check permissions on console.anthropic.com Error: \"Rate limit exceeded\" Solutions : 1. Wait for the indicated delay 2. Reduce request frequency 3. Upgrade to a higher plan (Max 20x) 4. Use to reduce tokens Error: \"Connection timeout\" Solutions : 1. Check internet connection 2. Check status: status.anthropic.com 3. Retry with longer timeout: Context Problems Error: \"Context window excee"},{"title":"Claude Code Billing and Cost Optimization","desc":"Day 16 - Understanding and mastering your consumption","category":"IA, Développement","url":"/en/claude-code-billing-costs-en/","date":"Dec 26, 2025","content":"Claude Code is powerful, but it consumes tokens. Understanding billing helps optimize your costs. Let's see how to monitor and master your budget. Beyond financial cost, don't forget the ecological impact of AI every token has a carbon footprint. The Pricing Model Rates by Model (December 2024) Model Input (1M tokens) Output (1M tokens) Claude Sonnet 4 $3.00 $15.00 Claude Opus 4 $15.00 $75.00 Claude Haiku 3.5 $0.80 $4.00 What Consumes Tokens Action Tokens consumed Your prompt Input tokens Claude's response Output tokens Files read Input tokens CLAUDE.md context Input tokens Conversation histor"},{"title":"CI/CD and Headless Mode with Claude Code","desc":"Day 15 - Automate your pipelines with AI","category":"IA, Développement","url":"/en/claude-code-cicd-headless-en/","date":"Dec 25, 2025","content":"Claude Code isn't limited to interactive use. With headless mode , you can integrate it into your CI/CD pipelines. Let's see how to automate intelligently. Headless Mode: p The (or ) flag allows running Claude Code without interaction: Claude reads the prompt, executes the task, and returns the result to stdout. Basic Syntax Headless Mode Options Option Description Execute with this prompt Structured JSON output Text output (default) Limit iterations Restrict tools GitHub Actions Integration Automatic Code Review Workflow Test Generation Workflow GitLab CI Integration Production Use Cases 1. A"},{"title":"Claude Code in VS Code and JetBrains","desc":"Day 14 - The graphical experience in your favorite IDEs","category":"IA, Développement","url":"/en/claude-code-vscode-jetbrains-en/","date":"Dec 24, 2025","content":"So far, we've used Claude Code in the terminal. But there are also extensions for VS Code and JetBrains . Let's see how to leverage the graphical interface. Two Complementary Approaches Aspect Terminal IDE Speed ⭐⭐⭐ ⭐⭐ Visualization Text only Visual diffs Integration Native shell IDE workspace Flexibility Maximum Guided Claude Code for VS Code Installation 1. Open VS Code 2. Extensions (Ctrl+Shift+X) 3. Search for \"Claude Code\" 4. Install the official Anthropic extension Or via terminal: Configuration After installation, configure via VS Code settings: VS Code Features Claude Code Panel Access"},{"title":"MCP: Connecting Claude Code to Your Tools","desc":"Day 13 - Integrate GitHub, Jira, databases, and more","category":"IA, Développement","url":"/en/claude-code-mcp-integration/","date":"Dec 23, 2025","content":"I've already explored the MCP protocol in my article Model Context Protocol (MCP): Revolution in LLM Integration. Today, we'll see how to use it concretely in Claude Code. Reminder: What is MCP? Model Context Protocol is an open standard that allows Claude to connect to external tools: Databases Issue trackers (Jira, GitHub Issues) Cloud services Internal APIs It's the \"USB C of AI\" : a universal interface. The Three Transport Types Transport Usage Example HTTP Remote servers Cloud services Stdio Local processes CLI tools SSE Legacy (deprecated) Old servers Installing an MCP Server HTTP Server"},{"title":"Hooks: Automating Event Reactions","desc":"Day 12 - Trigger automatic actions in Claude Code","category":"IA, Développement","url":"/en/claude-code-hooks/","date":"Dec 22, 2025","content":"Hooks allow you to execute automatic actions in response to Claude Code events. They're the bridge between Claude and your development tools. Today, we'll see how to use and create them. What is a Hook? A hook is an event handler that executes when Claude Code does something specific: Before/after tool execution At session start/end When user submits a prompt etc. The 9 Hook Types Hook Trigger Can Block Session start No Session end No Before tool execution Yes After tool execution No Prompt submission Yes Claude notification No User stop No Subagent end No Before context compaction No Hook Con"},{"title":"Claude Code Plugins and Marketplace","desc":"Day 11 - Extend Claude Code with the community ecosystem","category":"IA, Développement","url":"/en/claude-code-plugins-marketplace/","date":"Dec 21, 2025","content":"We've seen how to create slash commands, subagents, and skills individually. Plugins allow you to package them together and share them. Today, we explore the Claude Code plugin ecosystem. What is a Plugin? A plugin is a pack that can contain: Slash commands Subagents Skills Hooks MCP servers All in a standardized structure, easy to install and share. Installing Plugins Via the /plugin Command From a Marketplace From a Git Repo From a Local Folder Managing Installed Plugins List Plugins View Plugin Details Remove a Plugin Plugin Structure The plugin.json File Marketplaces Concept A marketplace "},{"title":"Skills: Model-Invoked Automation","desc":"Day 10 - Capabilities Claude activates automatically","category":"IA, Développement","url":"/en/claude-code-skills/","date":"Dec 20, 2025","content":"Yesterday, we saw subagents that you explicitly invoke with . Today, we discover Skills : capabilities that Claude activates automatically based on the context of your request. The Key Difference: Model Invoked Aspect Slash Commands Subagents Skills Invocation Explicit ( ) Explicit ( ) Automatic Trigger User User Claude Context Prompt Separate context Enrichment Skills are model invoked : Claude decides when to use them based on your request and the skill's description. How Skills Work The Process Concrete Example You have a skill with the description: \"Extract text and tables from PDF files. "},{"title":"Subagents: Delegating Intelligently","desc":"Day 9 - Create specialized agents for complex tasks","category":"IA, Développement","url":"/en/claude-code-subagents/","date":"Dec 19, 2025","content":"Slash commands automate workflows. Subagents go further: they're specialized AI agents with their own context, instructions, and tools. Today, we'll learn to use and create them. What is a Subagent? A subagent is a dedicated Claude agent that: Operates in a separate context from the main conversation Has its own system instructions Has access to specific tools Can be invoked for specialized tasks Subagent Advantages Advantage Description Isolation No pollution of main context Specialization Instructions optimized for a task Reusability Shareable across projects Control Tools limited by need Bu"},{"title":"Creating Custom Slash Commands","desc":"Day 8 - Automate your recurring workflows","category":"IA, Développement","url":"/en/claude-code-custom-slash-commands/","date":"Dec 18, 2025","content":"Built in slash commands are useful, but the real power comes from custom commands . Today, we'll create our own commands to automate recurring workflows. Where to Store Commands Project Commands (Shared with Team) These commands are versioned with the project and available to the entire team. Personal Commands (Cross project) These commands are available in all your projects. Anatomy of a Slash Command Basic Structure The YAML Frontmatter Field Description Required Description in help menu No Tools allowed for this command No The Command Body Everything after the frontmatter is sent to Claude "},{"title":"Permissions and Security in Claude Code","desc":"Day 7 - Protect your sensitive files and control actions","category":"IA, Développement, Sécurité","url":"/en/claude-code-permissions-security/","date":"Dec 17, 2025","content":"Claude Code has access to your file system and can execute bash commands. It's powerful, but requires safeguards. Today, we'll see how to secure your environment. For deeper context on AI security challenges, see my articles on LLM jailbreaking and robust development practices inspired by NASA. The Default Permission Model By default, Claude Code operates in strict read only mode: Action Default Permission Read files ✅ Allowed Search (Glob, Grep) ✅ Allowed Write/Modify files ❌ Requires approval Execute bash commands ❌ Requires approval Web access (WebFetch) ❌ Requires approval Anatomy of the P"},{"title":"Git Workflows: Commits, PRs, and Checkpoints","desc":"Day 6 - Collaborate effectively with Claude Code and Git","category":"IA, Développement","url":"/en/claude-code-git-workflows/","date":"Dec 16, 2025","content":"Claude Code integrates natively with Git. Today, we'll explore how to use this integration for professional development workflows. Automatic Commits with Attribution Commit Configuration In : Asking Claude to Commit Claude will: 1. Analyze modified files ( ) 2. Understand the context of changes 3. Generate an appropriate commit message 4. Execute the commit with configured attribution Example Generated Commit Conventional Commits Claude naturally respects Conventional Commits (see my detailed article on this topic): Prefix Usage New feature Bug fix Documentation Formatting (no code change) Ref"},{"title":"Context and Memory Management in Claude Code","desc":"Day 5 - Optimize long sessions and preserve your preferences","category":"IA, Développement","url":"/en/claude-code-context-memory-management/","date":"Dec 15, 2025","content":"A Claude Code session can last hours. But the longer the conversation, the heavier the context becomes. Today, we'll see how to effectively manage memory and context for productive sessions. The Hierarchical Memory System Claude Code uses a 4 level memory architecture, from highest to lowest priority: Level 1: Enterprise Policy For organizations, admins can define global rules that apply to all users. These rules have maximum priority. Level 2: Project Memory (.claude.md or CLAUDE.md) The file at the project root: Level 3: Project Rules (.claude/rules/) For modular and conditional rules: Examp"},{"title":"Prompt Engineering for Claude Code","desc":"Day 4 - Techniques to get exactly what you want","category":"IA, Développement","url":"/en/claude-code-prompt-engineering/","date":"Dec 14, 2025","content":"You now master the Explore → Plan → Code → Test workflow. But the quality of results depends enormously on how you formulate your requests. Today, we'll cover prompt engineering techniques specific to Claude Code. The Fundamental Principle: Specificity The difference between a good and bad prompt: Bad prompt Good prompt \"Add tests\" \"Add unit tests for the function covering: negative price, discount 100%, and nominal case\" \"Fix the bug\" \"The bug: at line 42 of . Fix by adding a null check.\" \"Improve the code\" \"Refactor to extract validation functions into a separate module\" Structure of an Effe"},{"title":"The Explore → Plan → Code → Test Workflow","desc":"Day 3 - The methodology that distinguishes senior developers","category":"IA, Développement","url":"/en/claude-code-workflow-explore-plan-code-test/","date":"Dec 13, 2025","content":"In my article about the state of AI in development in 2025, I mentioned the \"almost correct\" code syndrome: 45% of developers spend more time debugging AI code than writing it themselves . The solution? A structured workflow that transforms Claude from a haphazard code generator into a true development partner. The Problem: Coding Without Understanding Most developers use Claude Code like this: This is \"vibe coding\" : you vaguely describe what you want and hope it works. The Solution: The E.P.C.T. Workflow Phase 1: EXPLORE Objective : Understand before acting. Using the @explorer Subagent The "},{"title":"Claude Code Secret Syntax: #@/!","desc":"Day 2 - The shortcuts 95% of developers don't know","category":"IA, Développement","url":"/en/claude-code-secret-syntax/","date":"Dec 12, 2025","content":"Yesterday, we installed Claude Code and discovered the basic commands. Today, we level up with the secret syntax : four symbols that transform how you interact with Claude. The Four Magic Symbols Symbol Name Usage Context Activation Add to CLAUDE.md live Mentions Reference files and agents Slash Commands Built in and custom commands Shell Escape Direct bash execution Context Activation The symbol lets you enrich your without leaving Claude Code. How It Works During a session, press and type an instruction: This instruction is immediately added to your and persists for future sessions. Use Case"},{"title":"Claude Code: Installation and First Steps","desc":"Day 1 - Discover the AI assistant that will transform your coding","category":"IA, Développement","url":"/en/claude-code-installation-first-steps/","date":"Dec 11, 2025","content":"Welcome to this 20 article series to master Claude Code! Each day, we'll explore a feature that will make you a more efficient developer. Today: installation and first steps. What is Claude Code? Claude Code is Anthropic's official command line interface for interacting with Claude. Unlike traditional web chatbots, Claude Code integrates directly into your terminal and understands your codebase. What sets it apart: Direct access to your files and terminal Understanding of your project context Ability to read, write, and execute code Native Git integration Installation Prerequisites Node.js 18+"},{"title":"Prelude of the Chambered Reborn: Rewriting a Classic in TypeScript","desc":"Bringing Notch's Ludum Dare game to the modern web with Vite and software rendering","category":"Développement, Web, Tech","url":"/en/prelude-of-the-chambered-reborn-typescript-web-port/","date":"Nov 29, 2025","content":"In 2011, Markus Persson (Notch), the creator of Minecraft, participated in Ludum Dare 21 with a small retro dungeon crawler: Prelude of the Chambered . A raycasting engine à la Wolfenstein 3D, pixelated graphics, puzzles and bosses. A technical gem compressed into 48 hours of game jam. The problem? It was Java . And in 2025, who still launches a JVM to play a small indie game? The Project: Porting Java to TypeScript Prelude of the Chambered Reborn is my complete rewrite of the original code in TypeScript , built with Vite , playable directly in the browser. No plugins, no downloads, just a URL"},{"title":"GBTS: GameBoy Development with TypeScript and AI","desc":"An AI transpiler for GameBoy game development in TypeScript","category":"IA, Développement, Tech","url":"/en/gbts-typescript-gameboy/","date":"Aug 30, 2025","content":"Developing GameBoy games in 2025 is a fascinating challenge that blends nostalgia with extreme technical constraints. But what if we could use TypeScript — with all its modern power and features — to create retro games? This is exactly the problem I wanted to solve with GBTS (GameBoy TypeScript). The Problem: Between Nostalgia and Technical Complexity Traditional GameBoy development involves: C programming with archaic tools Manual memory management in an ultra constrained environment Complex optimizations to respect hardware limitations Huge learning curve for modern developers I had started "},{"title":"Auth0 International Email Generator: CLI for Multilingual Templates","desc":"TypeScript tool to automatically generate Auth0 email internationalization i18n templates","category":"Développement, Web, Sécurité, Auth0, TypeScript, i18n","url":"/en/auth0-international-email-generator-cli-modern-templates/","date":"Aug 28, 2025","content":"Auth0 Email Internationalization: The Challenge of Multilingual Templates Auth0 multilingual email management and authentication template internationalization represent a major challenge for developers. Between email translation , HTML template validation , language fallback management , and integration across different environments, i18n projects quickly become complex. To solve these Auth0 internationalization challenges, I developed the Auth0 International Email Generator¹. This TypeScript CLI tool automates multilingual email template generation with integrated validation, fallback systems"},{"title":"Sam Altman's AI Bubble Statements: Market Impact and Sectoral Analysis","desc":"OpenAI CEO's admission of speculative bubble causes major tech stock declines","category":"AI, Tech","url":"/en/sam-altman-ai-bubble-markets-analysis/","date":"Aug 21, 2025","content":"The technology world was shaken this week by the explosive statements from Sam Altman, CEO of OpenAI, who openly admitted that the artificial intelligence industry is experiencing a major speculative bubble period. Sam Altman's Confessions In a recent interview, Sam Altman didn't mince words regarding the current state of the AI sector. He explained that this bubble results from \"excessive investor enthusiasm for technology\" and that the consequences could prove particularly severe. \"When bubbles form, smart people get excessively excited about a grain of truth\" This declaration marks a notabl"},{"title":"OpenAI gpt-oss on Ollama: When Proprietary Models Go Open Source","desc":"OpenAI just announced its first open weight models available on Ollama. A collaboration marking a turning point in the local LLMs ecosystem.","category":"AI, Development","url":"/en/openai-gpt-oss-ollama-proprietary-models-open-source-transition/","date":"Aug 6, 2025","content":"OpenAI gpt oss on Ollama: When Proprietary Models Go Open Source OpenAI has just announced its first open weight models available on Ollama. A collaboration marking a turning point in the local LLMs ecosystem, but one that fits into a dynamic already initiated by Meta and Mistral. An Opening Movement in the Existing Ecosystem On August 5, 2025, OpenAI finally joins an already well established opening movement by releasing gpt oss , its first open weight models since GPT 2 in 2019. This collaboration with Ollama certainly marks a change in approach for a company historically oriented towards pa"},{"title":"AI and Development: Between Promises and Realities - A 2025 State of Affairs","desc":"Critical analysis of AI tools for developers: between promised revolution and field realities","category":"AI, Development","url":"/en/ai-development-between-promises-realities-2025-state/","date":"Aug 3, 2025","content":"The \"Falkland Law\", formulated by psychologist Robert Falkland, states a simple principle: if everything works well, why force change? This rule, based on observation of corporate behaviors, reminds us to weigh costs before changing just for the sake of change. When something works well, it's often better not to touch it. And if we can wait before deciding without it causing problems, we might as well take time to better understand the situation. This idea takes on its full meaning with AI in development. Faced with our tendency to always want something new, we can ask: do we really need to ch"},{"title":"Model Context Protocol (MCP): Revolution in LLM Integration with Third-Party Systems","desc":"Discover how MCP facilitates language model interaction with Atlassian tools and other connected systems","category":"AI, Development","url":"/en/anthropic-mcp-model-context-protocol-llm-integration/","date":"May 12, 2025","content":"What is the MCP Protocol? A Bit of Context Large language models (LLMs) are limited to information available at the time of their training. However, there are methods to enrich these models with more recent or detailed knowledge. One of these methods is the RAG (Retrieval Augmented Generation) system. This system works in two steps: it divides information into small units called vectors and uses sophisticated retrieval techniques to quickly and efficiently access necessary data. Thus, models can provide more current responses by incorporating up to date information from external sources like f"},{"title":"SQL Assistant AI: Natural Language to SQL Automatic Conversion","desc":"An open-source tool to generate SQL queries in natural language and interact with your database","category":"AI, Development","url":"/en/sql-assistant-ai-natural-language-sql-conversion/","date":"Feb 23, 2025","content":"SQL Assistant AI: Natural Language to SQL Automatic Conversion SQL Assistant represents a personal project I developed to address a concrete need : simplifying database interaction through a conversational interface. This open source tool enables automatic natural language to SQL query conversion with direct database execution capability . The project targets developers, data analysts, and occasional database users seeking a practical solution to interact with their storage systems without perfectly mastering SQL syntax. This personal initiative continues my ongoing explorations of AI practica"},{"title":"AI Ecological Impact: Training vs Inference Environmental Costs","desc":"Deciphering artificial intelligence's carbon footprint and exploring sustainable paths","category":"AI, Tech","url":"/en/ai-ecological-impact-training-vs-inference-environmental-costs/","date":"Feb 3, 2025","content":"AI Ecological Impact: Analysis of Training vs Inference Energy Costs Artificial intelligence (AI) today plays a major role in numerous domains: online recommendations 🎥, virtual assistants 📱, scientific research 🔬, and many others. But at a time when ecological transition becomes a global priority, it's crucial to question the environmental cost of these technologies. Behind each interaction with a chatbot or AI query lies a power hungry electrical infrastructure, not always aligned with climate objectives. This article examines two major stages in an AI model's lifecycle: training and usag"},{"title":"LLM Jailbreaking: Security Analysis of Bypass Mechanisms","desc":"Techniques, discoveries, and ethics around bypassing censorship in language models","category":"AI, Security","url":"/en/llm-jailbreaking-security-analysis-bypass-mechanisms/","date":"Jan 29, 2025","content":"Analysis of Bypass Vulnerabilities in DeepSeek R1 Large language models (LLMs) integrate sophisticated filtering mechanisms to prevent generation of sensitive or potentially dangerous content. DeepSeek R1 , the model developed by Chinese company DeepSeek, presents a particularly interesting case study regarding the effectiveness and limits of these censorship systems. This technical analysis examines vulnerabilities identified in DeepSeek R1's security mechanisms through a series of empirical tests. The objective consists of understanding architectural flaws and blind spots of current filterin"},{"title":"Robust Web Development: Applying NASA Methodologies for Critical Applications","desc":"Between theoretical limitations and minimalist practices, create robust web applications","category":"Development, Web","url":"/en/robust-web-development-nasa-methodologies-critical-applications/","date":"Jan 28, 2025","content":"Robust Web Development: Applying NASA Methodologies for Critical Applications Introduction: The Balance Between Complexity and Robustness Developing web applications free from defects constitutes a central objective for the modern software industry. However, achieving this optimal quality remains complex in a constantly evolving web ecosystem, constrained by temporal and budgetary imperatives. End users, whether businesses or individuals, nevertheless demand products with reliability and performance guarantees. This problem requires a methodological approach based on two complementary theoreti"},{"title":"Web 3.0 and Blockchain: Technical Analysis of a Decentralized Revolution","desc":"Web 1.0, Web 2.0, and now Web 3.0?","category":"Web, Tech","url":"/en/web-3-0-blockchain-technical-analysis-decentralized-revolution/","date":"Jan 24, 2025","content":"Understanding Web 3.0: A Digital Revolution Internet's architectural evolution constitutes a major technological transformation process. From the emergence of static web to contemporary proposals for decentralized and semantically enriched ecosystems, this evolution has been characterized by substantial paradigmatic mutations. The concept of Web 3.0 now occupies a central position in contemporary technological debates, presented as the next evolutionary stage of the Internet. This analysis examines the foundations of this transition and its historical contextualization. Understanding Web 3.0 r"},{"title":"Bun in 2025: Critical Evaluation of an Alternative JavaScript Runtime","desc":"How Bun can position itself as a strategic tool to simplify and accelerate modern JavaScript workflows in Digital Factories","category":"Development, Web","url":"/en/bun-2025-critical-evaluation-javascript-runtime-alternative/","date":"Jan 22, 2025","content":"Bun in 2025: Critical Evaluation of an Alternative JavaScript Runtime