Ingénierie agentique, structurée pour le retour sur investissement de l'entreprise

Build a correct mental model of tokens, context windows and why long prompts get worse, so you can drive any model well

Pick the right model tier for any task and know where to get strong models cheaply or free

Understand what an agent harness is and choose between Claude Code, Codex, Pi and OpenCode for a given job

Get Claude Code and Codex installed and running, choose the right plan, and protect your work and IP

Brief a coding agent so it delivers great work the first time, using axioms, framing, pushback and spec sheets

Scaffold a real project, run it locally in the browser, and put it safely under version control on a private GitHub repo

Deploy your project to the public internet on Vercel and connect a real custom domain through DNS and Cloudflare

Write a project rules file that makes your agent consistent, and grow it as a living skills library where every gotcha becomes a permanent rule

Package recurring workflows into reusable skills and slash commands you can trigger instantly, and know when each is worth building

Automate quality gates and routine steps with lifecycle hooks and scripts so they run without you thinking about them

Understand the Model Context Protocol, connect MCP servers to your agent, and judge when MCP beats a plain CLI tool

Use sub-agents and teams effectively, with the labyrinth analogy for clean delegation, while avoiding the context cliff and runaway cost

Manage long-running agent work with deliberate compaction, clean handovers, well-timed resets and the right thinking effort

Configure a professional Claude Code setup with a custom status line, the flags you use daily, and remote phone and cloud workflows

Build a clear mental map of frontend, backend and database, the marketing-versus-app split, and when a monorepo earns its keep

Add real authentication with Clerk, wire up Google OAuth, and take it from a dev instance to a production instance on your own domain

Model and query data in Convex with end-to-end type safety and live updates, and use indexes and soft deletes the way real products do

Handle secrets safely across dev and prod: env files, gitignore, recovering from a leaked secret, deploy keys, and encrypting stored user keys

Take real payments with Stripe Checkout and subscriptions, model monthly and yearly prices, and work safely in test mode before going live

Make billing reliable with verified webhooks and idempotency, handle proration on mid-cycle upgrades, run coupons, and ship embedded checkout

Migrate Clerk and Convex from dev to prod cleanly, wire DNS and Cloudflare, get indexed via Search Console, and pass Lighthouse

Choose the right automation platform for a job and generate workflow JSON with your agent

Automate and scrape the web reliably with Playwright, including the non-headless manual-login trick and the .har trick

Run agent-generated and untrusted code safely in isolated, disposable sandboxes instead of on your own machine

Build custom AI tools on top of model APIs, including image-to-structured-data workflows, instead of buying SaaS

Capture leads with magnets and forms and nurture them through email funnels, while staying legally compliant

Design systems where humans approve at the right checkpoints and every mistake becomes a rule the system learns from

Place any agentic system on a five-level autonomy scale, see why validation is the real blocker, and climb a level safely

Protect your project with type checks, linting, unit and end-to-end tests, wired into CI/CD

Apply core security measures so your app resists abuse and never leaks sensitive data

Meet core privacy obligations - GDPR, cookie consent and key US and Swiss laws - without legal overwhelm

Get discovered by search engines and recommended by AI assistants through SEO and GEO/AEO

Design products that agents can use directly, treating the API as the primary interface

Reason clearly about exponential AI progress and spot the market opportunities it opens

Plan, build, test, secure and ship a complete agentic product using everything from all five courses