Comment Créer Sans Limite avec l’IA: Formation Claude Code (2026) | Guide Complet Débutant

Main focus

The video is a beginner-to-intermediate training guide for mastering Anthropic’s Claude Code (“Claude Code formation”). The speaker argues that AI tools are removing the traditional barrier to digital creation, letting non-technical users generate and manage full software projects through natural language.

Why Claude Code (framed as a career/productivity shift)

• The speaker uses external forecasts (e.g., McKinsey and the World Economic Forum) to claim AI/automation will reshape jobs.
• The argument also includes that companies will recruit people skilled in AI—especially those who can set up automation.
• Claim: mastering Claude Code can make a person several times more productive (often presented with a 5–10x productivity figure).

Training structure (how the course is organized)

The instructor says the training is split into 3 major parts:

1. Foundations What Claude Code is, how it works “under the hood,” and key concepts.

2. Practical use cases (guided hands-on demonstrations), including:

   • building websites
   • developing applications
   • automating workflows
   • creating AI agent teams

3. Business perspective How experts use Claude Code to create business opportunities.

Core conceptual model of “what Claude Code does”

The video explains software building blocks using a cooking metaphor:

• Code = precise instructions
• Program = full set of instructions that performs a task
• Application = a program with UI/screens the user interacts with
• Website = accessible pages via a URL
• Servers = host/serve content
• Database = organized storage (e.g., messages, images, orders)

Claude Code vs Claude vs Claude “Cowork” (Anthropic product ecosystem)

The video distinguishes three Anthropic products:

• Claude (chat): Q&A in the browser
• Claude “Cowork”: can act more directly on your computer/browser (file reading, browsing, task execution)
• Claude Code: the centerpiece; most powerful—creates full projects, automations, and can run multiple AI agents in parallel

Five main Claude Code capabilities (technological features)

Claude Code is described as an assistant with:

1. Natural language understanding (no programming language required)
2. Computer/file access (create/modify/delete/read documents with permission)
3. Connectors to external services (examples mentioned: Google Calendar, email, Notion, Slack, etc.)
4. Skills (specialized “modules,” e.g., accounting, research, writing in a specific style; can use existing ones or create new)
5. Autonomy / agent-like execution (can perform tasks while the user is away, enabling agent teams)

Where Claude Code can run (3 access environments)

The guide claims Claude Code can be used in three places, each with tradeoffs:

• Desktop application (easiest onboarding)
• Terminal / CLI (powerful, command-based; “learn to tame it”)
• IDE like VS Code (recommended for serious development, multi-agent workflows, and better structure/security)

Pricing and recommended plan

• States Claude Pro is required, roughly $20/month (minimum for Claude Code).
• Mentions higher tiers (e.g., “Max”) for heavier usage.

Safety & compliance guidance (major emphasis)

Before building a personal assistant with broad access (“Jarvis”), the video provides three safety rules:

1) Control permissions

Claude Code has authorization modes, including:

• systematic permission prompts (recommended for beginners)
• “approve modifications” / staged approvals
• planned execution that asks permission after planning
• YOLO/auto modes (explicitly discouraged for beginners)

It also includes a warning anecdote: a developer allegedly lost their computer due to auto-permission behavior (used as a cautionary example).

2) Protect sensitive data

It recommends avoiding sending highly sensitive categories, including:

• customer personal data (e.g., names + phone/email)
• medical data
• insurance data
• private banking details
• bank account details (RIB) and screenshots of credentials

It suggests anonymization (e.g., replacing real names with fictitious ones) and mentions GDPR-related training/module within their AI academy.

3) Be cautious with connectors, MCPs, and skills

• Warns about prompt injection risks from malicious or compromised skills/connectors.
• Recommends installing only trusted/official publishers.
• Suggests asking Claude to assess safety before installing external skills.

“Jarvis” starter kit: building persistent memory/context locally

A key tutorial component is installing a pre-configured starter kit that creates a local workspace structure.

Key points:

• Memory is not automatically shared across environments (desktop vs terminal vs IDE).
• The workflow is to manually prime each session by loading a context file (referred to as “Prime” in the workflow).
• Uses Markdown-based files such as:
  • cloud.md: defines “the soul/rules” and Jarvis behavior
  • context.md: user identity/goals/projects
  • history.md: evolving journal that gets updated over time

Practical demo flow: installing Jarvis

The speaker describes these steps:

1. Download the starter kit from free community/resources
2. Open it in Claude Code (desktop first)
3. Run install via a command such as /install
4. Claude asks questions (identity, city, goals, projects)
5. Claude updates workspace files with the new context
6. The user validates permission prompts for file edits (using safer mode)

Second major practical demo: making Jarvis “act” using connectors + skills

The tutorial upgrades Jarvis from chat-only to action-oriented by:

• Activating connectors (example shown: Canva and Notion)
• Using skills (e.g., Canva design skill, plus planning/reflection capabilities)

End-to-end task demonstrated:

• Jarvis performs benchmarking/research
• creates a business plan page in Notion
• generates a pitch deck in Canva
• coordinates tool use across platforms with selective permission approvals

Automation features: commands, skills, routines, agents

The guide explains:

• Skills can be triggered via keywords or conditions
• Skill Creator (a slash command) helps users create their own skills
• Routines/scheduling (e.g., a morning routine that monitors sources and updates you at a set time)
• Mentions “agents” and remote agents as more robust automation

Website creation pipeline (conceptual + tooling)

Later, the video shifts into building and deploying a website, using a standard dev-to-web chain:

1. Write code locally
2. Git: version control locally (checkpoints/rollback)
3. GitHub: push backups to the cloud
4. Netlify (and mentions Vercel): deploy GitHub-backed code to the internet
5. Domain name: buy a domain to replace the default Netlify URL
6. Servers/data centers: infrastructure serving the deployed files

It emphasizes:

• Never commit secrets/API keys to Git/GitHub
• Use a local secrets file and rely on gitignore

Development best practices advocated

• Use VS Code for real projects because it:
  • improves file organization (project structure)
  • supports live editing
  • supports safer handling of “secrets” and API keys
  • supports multi-agent workflows better than a single terminal window
• Adds a “workspace” structure for deliverables (e.g., YouTube, agency, academy, etc.)
• Mentions additional workspace files like an ENV/secrets file and a gitignore template

Tutorial emphasis: what to do in order

The speaker repeatedly instructs viewers to:

• practice in parallel (open their computer while watching)
• follow installation steps in sequence: starter kit → Prime/context → connector/skills → VS Code setup → website pipeline

Main speakers / sources

• Speaker/host: Yas (creator of an AI consulting firm “Chatflow AI” and the YouTube channel; narrator throughout)
• Referenced external sources: McKinsey and the World Economic Forum (labor impact claims)
• Referenced companies/tools: OpenAI/ChatGPT (historical context), Anthropic, plus Git, GitHub, Netlify, VS Code, Canva, Notion (as product/service examples)