🗺️ Roadmap: Vibe Coder / AI Generalist

9 core modules + 3 optional. 58 units total: 43 in the core and 15 in optional tracks.

🎯 Overview

An AI Generalist is a full-stack problem solver who assembles solutions with AI. The course is agent-agnostic: skills transfer between Claude Code, Codex, Hermes, goose, OpenCode, nanobot, and CheetahClaws — including on a sovereign stack.

🪜 Course modules

Each module follows Kolb’s cycle: Activation → Reflection → Concept → Practice. The core is 00–08; 09–11 are optional.

Kickstart

Units: ABC: six words before you start · Your ghost idea · Map of tools · First steps · The Source or the Packaging

Concepts: token, context window, temperature, models, prompt and context, agent; the AI-tools map; the human role in vibe coding; checking a primary source.

Projects: u1-ghost-idea-plan.md, u3-kickstart-roadmap.md.

Introduction

Units: Your experience with AI · Four shifts · One clone, five types · Your first prompt

Concepts: Software 3.0; delegation, experiment, specification, and attention management; five clone types; Capture → Build → Test → Refine.

Tools: Claude Code as the working example.

Projects: my-experiments/u2-shifts-before-after.md; one clone to a first version and specification, five types as a selection map.

Setup

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

Concepts: Node, Git, and the terminal; installing Claude Code; first run; AGENTS.md and @AGENTS.md; the starter project.

Projects: u1-env-check.png, u2-claude-first-run.png, u3-claude-reads-context.md.

Stack selection

Units: What are your constraints? · Matrix of 4 stacks · Behind a censorship firewall: no VPN · Hermes + local stack · Migration plan for your workflow

Concepts: constraints → stack matrix → LiteLLM model gateway and fallback chain → local stack → migration plan.

Projects: my-experiments/u1-constraints.md, u3-cerebras-via-litellm.png, u4-hermes-mental-model.md, u5-migration-plan.md.

Prompt Engineering

Units: My worst prompt · The spec formula · Magic words · The sins of prompting · A professional prompt

Concepts: Context + Task + Instructions + Data; Role / Task / Input / Output / Constraints; meta-prompting; five prompting sins.

Projects: three prompts from bad to good; my-experiments/u5-reusable-prompt.md.

Context & Memory

Units: When the agent didn't get you · Context vs Prompt · The memory system · Tune your memory

Concepts: Context Rot and Lost in the Middle; /context; AGENTS.md, CLAUDE.md, STATE.md, and TODO.md; /memory; session summaries; parallel agents.

Automation Pipeline

Units: The info you process by hand · Trigger → Action model · Build your pipeline · What to automate next? · A pipeline that runs on its own

Concepts: trigger, action, condition; manual, scheduled, and event-driven runs; idempotency; notifications and safe launch conditions.

Tools & Extensions

Units: What's Claude Code missing? · MCP: superpowers through a protocol · Hooks: event automation · Skills: modular capabilities · Plug it in and test · Pendel: your agent audits your project

Concepts: MCP — Tools, Resources, Prompts; Hooks; Skills; choosing MCP from the module table; checks and audit.

Projects: my-experiments/u6-pendel.md.

Agent Engineering

Units: When Prompts Stop Working · Jagged Intelligence: What to Delegate · Orchestration: Gmail Agent as a Case · Prototype → Production · Design Your Clone · Agent SDK: the Agent as a Library

Concepts: the boundaries of vibe coding; Jagged Intelligence; multi-agent pipelines; production infrastructure; LiteLLM fallback routing; Agent SDK.

Projects: task decomposition, pipeline blueprint, production checklist, agent specification, and my-experiments/u6-agent-sdk.md.

AI Notebook

Units: A notebook in 15 minutes · How it works · Where it lies · Output formats · Practice

Model Training

Units: What's inside a model · Data is the main work · Fine-tuning: when and how · Evaluation: did it help · Practice: fine-tune a small model

Second Brain

Units: A second brain that stays with you · A channel: your agent within reach · Skills travel from agent to agent · Security: everything that comes from outside · Practice: build your own second brain

🧭 The next step: AI Solution Architect

The course leads to designing an agentic system around a real problem: setting AI boundaries, splitting work into AI / Tool / Code, and choosing the stack, memory, observability, and fallback route. It does not promise a job or income.