🗺️ Roadmap: Vibe Coder / AI Generalist

A path from non-coder to AI generalist. 8 elective topics — take them in any order.

🎯 Overview

AI Generalist — a full-stack problem solver who solves problems with AI.

They have:

• A deep understanding of AI models and what they can do
• Knowledge of which model to use where, and how
• The ability to assemble solutions to complex problems from ready blocks

Who this path is for

• Entrepreneur — launching a startup or business and want to move faster
• Freelancer / consultant — selling expertise, scaling your output
• Hustler — looking for new ways to earn
• Employee — want to grow faster and become indispensable

In an "AI First" world, becoming an AI generalist isn't optional anymore — it's basic literacy.

🐎 Pick your horse

The course is agent-agnostic: it teaches the rider's craft, not one horse. Under the hood it's Claude Code, Codex, OpenCode, Cline, or your sovereign stack. Pick the horse that fits your budget and constraints (that's the "Stack selection" topic) — the skills carry across all of them. Swap horses and your seat stays yours.

🪜 Course topics

Each topic is broken into 3–5 cognitive units following Kolb's cycle: Activation → Reflection → Concept → Practice.

🔭 On the radar

Topics in consideration — a direction, not a final list. Want one sooner? Tell us via feedback — it moves the priority.

• NotebookLM — working with your own notes as grounded context
• Meta-notes — a note-making playbook that feeds the agent
• Telegram management — agentic channel and bot management
• Media Production — creating content (audio/video/visual) with AI

Kickstart

For people brand-new to AI coding who want to lower the cognitive load at the start.

Units: Map of AI tools · Your new role · First steps

Concepts:

• What vibe coding is, in plain words
• Minimal vocabulary (LLM, token, API, deployment, MCP, agent)
• The map of AI tools: 4 quadrants (zero-code builders, AI dev platforms, design-to-code, AI coding assistants)
• Your new role: PM + QA + DevOps

Result:

• Your own mental model of AI tools and your place in them
• First open-form question to Claude

Introduction

The mindset foundation — without it, every other topic falls flat.

Units: Your experience with AI · Four shifts · Five clones · Your first prompt

Concepts:

• Software 3.0 and why it's a different world
• Four shifts: commands → delegation, results → experiments, actions → specifications, "do it all yourself" → managing attention
• Five types of clone agents (Communication, Meeting Intelligence, Video, Learning, Automation)
• The Capture → Build → Test → Refine cycle

Tools: Claude, ChatGPT, Gemini — as a thought experiment before installing Claude Code

Projects:

• Rewrite your last prompt through the four shifts (experiment-1-shift.md)
• Pick the one clone type where you lose the most time

Setup

The workspace you live in with the agent every day.

Units: Environment check · Installing the tools · Your first project

Concepts:

• Terminal as the main interface (Warp, Wave)
• Claude Code CLI and its permission modes
• Git basics for agentic workflows
• Project structure: CLAUDE.md, TODO.md, AGENTS.md, my-experiments/, my-templates/
• Nerd Font, status bar, Chrome Extension

Tools: Warp, Claude Code CLI, GitHub, Node.js, Marp

Projects:

• Configured workspace with Claude Code, Git, Marp
• First project with CLAUDE.md + my-experiments/ structure
• Working status line with token tracking

Prompt Engineering

Writing the spec — the core skill for a vibe coder.

Units: Activation · The spec formula · Magic words · Seven sins · Practice

Concepts:

• Simple starting formula: Context + Task + Instructions + Data
• Full structure: Role / Task / Input / Output / Constraints
• The Open Claw philosophy and an experimental approach
• Meta-prompting (a prompt that improves prompts)
• Common mistakes and how to avoid them

Tools: Claude Code, any LLM for A/B prompt comparison

Projects:

• 3 prompts: from bad to good with reflection
• Your own prompt template in my-templates/
• The start of a library of working patterns

Context & Memory

How the agent holds information — and why without it, it forgets everything in 5 minutes.

Units: Activation · Context vs Prompt · Memory systems · Practice

Concepts:

• Context problems: Context Rot, Lost in the Middle
• CLAUDE.md, TODO.md, AGENTS.md — three memory files
• Memory hierarchy: session / project / global
• Context compaction, /compact, /context
• Claude Code's built-in agents
• Parallel agents (sub-agents)

Tools: Claude Code memory system, .claude/ folder, hooks for auto-memory

Projects:

• Memory system for a real project (CLAUDE.md + TODO.md)
• Filled-in PERSONAL-CONTEXT.md
• A parallel-agents exercise

Automation Pipeline

From a manual prompt to an automated chain of actions.

Units: Activation · Pipeline theory · Build · Reflection

Concepts:

• Pipeline thinking: URL → scrape → analyze → insights
• Firecrawl and web scraping
• Prompt chaining — linking prompts into a chain
• Sources-first approach: data first, then analysis
• Generating ideas.md and research-topics.md

Tools: Firecrawl, Claude Code, Obsidian CLI (optional), web sources

Projects:

• Pipeline for a chosen URL from start to output
• Saved ideas.md and research-topics.md in my-experiments/
• First automated workflow that saves 2–5 hours a week

Tools & Extensions

From Claude Code user to architect of your own system.

Units: What's missing · MCP · Hooks · Skills · Plug it in and test

Concepts:

• MCP (Model Context Protocol): three primitives — Tools, Resources, Prompts
• Tool Search: 85% token savings on tools
• Agent Skills — modular capabilities without token cost
• Hooks — event automation: PreToolUse / PostToolUse / Stop
• Superpowers workflow: Brainstorm → Plan → Execute → Verify

Tools: MCP servers (github, stitch, chrome-devtools, vercel), Skills repo, hooks config

Projects:

• A connected MCP server with a real use case
• An installed or self-built Skill
• At least one Hook configured for a real workflow

Agent Engineering

From vibe coding to system orchestration — where prompts stop working.

Units: When prompts stop working · Jagged Intelligence · Orchestration (ENERV case) · Prototype → Production · Design your clone

Concepts:

• Vibe coding vs Agent engineering — borders and transitions
• Three eras of programming (Karpathy: Software 1.0 / 2.0 / 3.0)
• Jagged Intelligence — what AI nails, where it fails
• Multi-agent pipeline architecture (ENERV case)
• Production stack: Hetzner + Cloudflare Tunnel + Docker + n8n + Langfuse (SOVERN)
• 5 production blocks: Execution / Observability / Reliability / Cost / Security
• 5-fallback LLM routing via LiteLLM

Tools: n8n, Langfuse, Cloudflare Tunnel, Docker, LiteLLM, Hetzner CX22

Projects:

• Decomposing a real task into AI/Tool/Code nodes
• Multi-agent pipeline blueprint
• Production readiness checklist
• Implementation-ready agent specification

🌐 The Bigger Picture

Connecting all the topics into your "army of clone agents" — 5 clones that take the routine off your plate, freeing you up for strategic decisions.

Five clone candidates (see Introduction):

1. Communication Clone — writes emails and messages in your voice
2. Meeting Intelligence Clone — summarizes meetings, extracts tasks
3. Video / Content Clone — generates posts, clips, captions
4. Learning Clone — teaches you new things on a schedule
5. Automation Clone — routine pipelines (URL → summary, data → report)

Building one clone is applying multiple topics at once:

• Introduction sets the mindset — delegate, don't command
• Setup and Prompts configure the environment and write the spec
• Context gives the clone memory
• Pipeline gives it a workflow
• Tools extends its reach via MCP and Skills

⚡ Summary of Superpowers

After mastering the topics, you get four powers:

• Power to Think with AI — write specs, not commands; think in spec terms
• Power to Build with AI — assemble pipelines and tools for your tasks
• Power to Automate with AI — claw back 5–15 hours a week from routine
• Power to Network with AI — share patterns with the community and amplify each other