Skill tree · mind map · course curriculum

Level up as an AI-Augmented Developer

36 skills, 6 tiers — from how the model works to shipping AI in production. Each skill has why it matters, a mastery check, the pitfall to dodge, and hand-picked resources. Every tier ends in a project you actually build.

◇ How this course works
  • Same six tiers, two tracks. Foundations → Prompting → Code with AI → Extend your agent → Build into products → Production & frontier — every learner crosses all six, at different depth.
  • Each module maps to one tier. The skill tree's own content — why it matters, mastery check, pitfall, resources — is the required reading; this syllabus adds pacing, an exercise, and a completion bar.
  • The mastery check is the assessment. No separate quiz system — each skill's existing "you've got it when…" is the pass bar for that lesson.
  • Both tracks converge on tier-end projects. The tree already ends every tier in a hands-on project; Builder Track does it as specified, Foundations Track does a scoped-down version of the same brief.
Track A

Foundations Track

For: no prior programming background — PMs, analysts, career-switchers, students. Pace: ~10 weeks, 3–4 hrs/week, deliberately slower. Tooling: Claude.ai chat and the Claude Code chat interface; no CLI until ready.

01

How the model actually works

Tier I · Foundations (7 skills)

Change

Analogies before mechanics — skip token math and architecture diagrams; "training" explained as practice, not programming.

Exercise

Predict-the-next-word by hand, then compare to what the model does.

02

Talking to AI effectively

Tier II · Prompting (5 skills)

Change

Full depth — the highest-leverage skill for this audience. Structured output taught as plain-English templates, not JSON schemas.

Exercise

Iterate a prompt for one real work task five times, logging what changed.

03

Your first AI-assisted builds

Tier III · Code with AI (6 skills)

Change

Reframed from "coding" to "directing" — read the diff, catch obvious wrongness, learn what you don't understand yet.

Exercise

Build one small real thing end-to-end, described entirely in English.

04

Giving your AI superpowers

Tier IV · Extend your agent (6 skills)

Change

Conceptual only for MCP, agents, hooks — "build a plugin" becomes a guided walkthrough, not a build.

Exercise

Install one existing MCP server; describe in plain English what Claude could newly do.

05

AI in the real world

Tier V · Build into products (7 skills)

Change

Skip the API/streaming/fine-tuning labs. RAG and evals stay conceptual: why the AI needs your docs, how you'd know it's working.

Exercise

Use a hosted no-code RAG builder to make a small Q&A bot over a real document set.

06

Being responsible, staying current

Tier VI · Production & frontier (5 skills)

Change

Full depth on security and responsible AI — matters regardless of coding skill. "Cost & scaling" becomes "why AI answers aren't free."

Exercise

Write a one-page AI usage policy: what never to paste into a prompt.

Capstone

Ship one small, real, personally useful AI-assisted tool — built with heavy AI assistance, presented as what it does, what you directed, and what you checked before trusting it.

Track B

Builder Track

For: developers who already code, in any language. Pace: ~6 weeks, 5–6 hrs/week, assumes fluency. Tooling: Claude Code CLI from day one, API keys, a real repo throughout.

01

Foundations

Tier I · Foundations (7 skills)

Lab

Read the tier's curated resources — Karpathy, 3Blue1Brown, the attention paper. Conceptual grounding only, no lab.

Checkpoint

Explain training vs. inference and why hallucination isn't a bug — unaided, one paragraph.

02

Prompting

Tier II · Prompting (5 skills)

Lab

Build a versioned prompt library for a real repeated task in your own work, with structured output specs.

Checkpoint

Peer-review another learner's library for the context-engineering pitfall.

03

Code with AI

Tier III · Code with AI (6 skills)

Lab

A full week shipping a real feature exclusively through an AI coding assistant — every diff read before accepting.

Checkpoint

Submit the PR; a reviewer checks specifically for signs you didn't read your own AI-generated code.

04

Extend your agent

Tier IV · Extend your agent (6 skills)

Lab

Build and ship one actual plugin or MCP server — the tree's own tier project.

Checkpoint

Someone else installs and uses it without your help.

05

Build into products

Tier V · Build into products (7 skills)

Lab

Build a small RAG-backed feature against the raw API — no framework — covering streaming, tool use, and a basic eval harness.

Checkpoint

Present the eval results: what the harness catches, what it misses.

06

Production & frontier

Tier VI · Production & frontier (5 skills)

Lab

Security pass (OWASP-style) on the Module 5 project, plus a cost estimate at 10x current usage.

Checkpoint

Red-team someone else's Module 5 project for prompt injection.

Capstone

Ship a real MCP server or Claude Code plugin, published even privately, with a security review and a cost/scaling note. Strong capstones are candidates for actual inclusion in the Plugboard directory.

Where the two tracks meet

  • Shared showcase at the end. Both capstones get presented together — Foundations learners see what's next, Builder learners practice explaining technical work to a non-technical audience.
  • Foundations graduates who want to keep going slot into Builder Track Module 3 directly, after a fast-follow pass through the code-heavy versions of Modules 1–2's resources.

Want a full external course too?

This syllabus wraps the skill tree's own resources — these are separate, complete courses worth doing alongside it, one pick per track.

For Track A
Anthropic Academy — Claude 101 & AI FluencyOfficial, free, no code required — certificate on completion. DeepLearning.AI — ChatGPT Prompt Engineering for Developers~1.5 hrs, free, taught by Andrew Ng & Isa Fulford.
For Track B
Anthropic Academy — Claude Code 101 & Building with the Claude APIOfficial developer track — API, agents, MCP servers. Karpathy — Neural Networks: Zero to HeroBuild neural nets from scratch in code, up through GPT.