The Vibe Coding Roadmap — 2026
Five free courses, 138 lessons, one path: from your first prompt with Claude Code, to a shipped mobile app with Codex, to a fleet of agents in Google Antigravity, to the systems layer in Agentic AI Engineering, to full prompting mastery. Click a module to jump in — or pick a role on the AI career roadmaps.
Vibe Coding using Claude Code
Ship real software at the speed of conversation.
- M01Install the CLI, learn the loop, set up project memory. End of this module you're operational.0 / 5 lessons
- M02
- M03Slash commands, skills, subagents, plan mode, hooks, and MCP — the tools that let you bend Claude Code to your workflow.0 / 7 lessons
- M04Prompts and rhythms that compound. The difference between a senior vibe coder and a tourist.0 / 5 lessons
- M05
Build Mobile Apps using Codex
Ship to TestFlight and Play Store with OpenAI's Codex as your pair.
- M01Install Codex, scaffold an Expo app, understand the constraints of mobile vs web.0 / 5 lessons
- M02
- M03
- M04The capabilities that make it feel like a real app, not a wrapped website.0 / 5 lessons
- M05
Agent Manager with Google Antigravity
Run a fleet of agents from mission control — Google's agent-first IDE.
- M01Install the IDE, meet the two surfaces, run your first Manager task, tune autonomy.0 / 5 lessons
- M02Plans, execution, verification, shipping, and the knowledge base — the working vocabulary of agent-first development.0 / 5 lessons
- M03Sequencing, parallel worktrees, second opinions, steering, and the inbox. The control flow of a fleet.0 / 5 lessons
- M04Models and rate limits, audit trails, failure modes, security, and the three metrics that keep leverage honest.0 / 5 lessons
- M05Issue → PR, docs sync, parallel triage, worktree migration, capstone. The workflows you'll actually run.0 / 5 lessons
Agentic AI Engineering
The systems track: everything around the model that makes AI products actually work.
- M01Tokens, prompting, function calling, structured outputs, the modern stack — and the latency/cost/reliability budgets that shape everything.0 / 5 lessons
- M02Curate the window, anchor answers in your own data, and know which advanced retrieval patterns earn their complexity.0 / 4 lessons
- M03From workflows to agents: architecture patterns, MCP for tools, A2A for peers, and LangGraph when orchestration gets real.0 / 5 lessons
- M04When training pays off, LoRA/QLoRA in practice, and running open-weight models locally with Ollama.0 / 3 lessons
- M05Golden datasets, failure analysis, LLM-as-judge, tracing, and the production eval loop that catches regressions before users do.0 / 4 lessons
- M06The threat surface, injection and PII defense, guardrails that fail closed, red teaming, and governance — production readiness, proven.0 / 4 lessons
Prompt Engineering Mastery
Every technique that matters in 2026 — from zero-shot to chain-of-thought to prompt injection defense.
- M01The working parts of a prompt — precise instructions, delimiters, roles, and examples — assembled into contracts that hold up under real input.0 / 6 lessons
- M02Make models show and improve their work: chain-of-thought, decomposition, voting across samples, and where 2026 reasoning models change the playbook.0 / 6 lessons
- M03Why models fabricate, and the countermeasures that ship: calibrated abstention, critique loops, ensembles, debiasing, and prompt test sets.0 / 6 lessons
- M04Prompts that survive production — strict output contracts, long-context and RAG patterns, agent system prompts, meta-prompting, and versioning.0 / 6 lessons
- M05Prompt beyond text — generating images with control, steering with negatives and weights, reading images as input, and the voice and video frontier.0 / 5 lessons
- M06The security layer every builder needs — how prompt injection, jailbreaks, and prompt leaking work, and how to defend your own app in depth.0 / 6 lessons
Start with Vibe Coding using Claude Code — the loop, the muscle memory, the prompt patterns. Move to Build Mobile Apps using Codex once you can ship a small web feature without thinking about the agent. Save Agent Manager with Google Antigravity for last — multi- agent orchestration is easier when you've operated each agent solo first. Take Agentic AI Engineering alongside or after Course 02 — it's the systems track: how the AI you've been directing actually works, and how to build production AI features of your own. And Prompt Engineering Mastery runs alongside everything — it's the technique layer, useful from your first prompt to your last red-team pass.
How do I start learning vibe coding in 2026?
Start with one agent and one real project. Our roadmap begins with Claude Code in your terminal: you learn the loop (describe an outcome, review the diff, correct, repeat), then prompt patterns, plan mode, hooks, and MCP. From there the path moves to shipping a mobile app with OpenAI Codex, and finally to orchestrating a fleet of agents with Google Antigravity.
How long does it take to learn vibe coding?
The full roadmap is roughly 92 hours of guided work — about 6 to 10 weeks part-time. Course 01 (Claude Code, ~14h) makes you productive in the first week; Course 02 (Codex mobile, ~18h) and Course 03 (Antigravity, ~20h) build toward shipping production software and running agents in parallel; Course 04 (Agentic AI Engineering, ~16h) teaches the systems layer — RAG, evals, and AI security; Course 05 (Prompt Engineering Mastery, ~24h) can be taken alongside any of them.
Do I need to know how to code before starting?
No. Course 01 assumes zero coding background — the agent writes the code, and the course teaches you to specify, review, and steer. Basic web or mobile knowledge helps in the later courses, but every lesson explains the concepts it touches as they come up.
Is this vibe coding roadmap free?
Yes. All 138 lessons across the five courses are free, with no signup and no paywall. Progress, XP, streaks, and badges are saved in your browser.
Which AI career can this roadmap lead to?
The skills here map directly onto the new AI-era roles: Vibe Coder, Forward Deployed Engineer, Agent Engineer, AI Transformation Engineer, and more. See our AI career roadmaps at vibecodeschool.com/careers for ten role-by-role paths with salaries, skills, and proof-of-work checkpoints.
Which AI coding tool should I learn first: Claude Code, Codex, or Cursor?
We recommend a terminal-first agent like Claude Code before an IDE like Cursor: it forces you to work at the level of intent instead of autocomplete, which is the core vibe coding skill. Codex and Antigravity then generalize that skill — the loop is the same, so switching tools later is cheap.