This is the classic path: the full course-based curriculum,
preserved and still valid. The main course is now
The Estate, a mission-based path that covers the same
concepts in about half the time. Missions there reference sections here (the
Xubuntu install guide in Phase 0 especially), so this page stays alive. Progress
on the two pages is tracked separately.
A hands-on bootcamp with one student (you, Charles) and one classroom: a Dell
Latitude 3490 that starts as a blank slab and ends as a working full-stack
development machine. There is no video series to binge and no sandbox in a browser
tab. Every phase installs real tools on the real machine and uses them on real
work. When the course is done, the laptop itself is the certificate: its
configuration, its repos, and its running services demonstrate the concepts.
The curriculum is a custom hybrid built from the best free sources available in
2026 (research in docs/RESEARCH.md): Full Stack Open's
concept-then-exercise method, Boot.dev's Linux-first structure, The Odin Project's
no-answer-key projects, Anthropic Academy's AI workflow track, and Ed Donner's
public AI-engineering labs. Roughly 290 hours of work. At about 5 hours a week,
that is a loose 12 to 18 months. The timeline view below tracks hours, not dates,
on purpose.
The one rule: hand-first
This course teaches both meanings of "AI-powered developer": you will build
AI-powered apps, and you will learn to code with an AI assistant. But the 2026
consensus among people who train developers is blunt: leaning on AI before the
fundamentals stick produces someone who can prompt but cannot debug. So the
standing rule for every task before Phase 5 is:
Write it yourself first. Then, and only then, bring an AI in as a
reviewer, explainer, or debugger. Ask it why your code works, what could break it,
and how a senior dev would improve it. Never paste in a first draft you did not
attempt. (This is the CS50 "rubber duck" model. Claude Code becomes a primary tool
in Phase 5, once you have something real for it to work on.)
Check tasks off as you finish them; progress saves to this browser automatically
(localStorage). Each phase has a notes box that also saves. Before switching
browsers or machines, click Export progress and keep the JSON file; import
it on the other side. This file is yours to edit: add tasks, reword things, log
discoveries. The only rule is in the comments at the top: never renumber an
existing task ID.
Loose timeline
Hour estimates at roughly 5 hrs/week. Bars fill as you check tasks off. No dates, no guilt.
Phase 0 · The Machine
~15 hrs
Wipe the Latitude, install Xubuntu 24.04 LTS, and learn the terminal well enough
that the rest of the course never has to explain cd again. This phase
is written in full detail because you will be following it on a machine with no
working browser of its own yet. Read it on another device or print it.
0.A Build the installer (on your Windows PC)
0.B Install
# Latitude 3490 keys, at the Dell logo:
# F12 = one-time boot menu (choose the USB stick)
# F2 = BIOS setup (only if F12 doesn't offer USB)
sudo apt update # refresh the package index
sudo apt full-upgrade # install everything new
sudo fwupdmgr refresh && sudo fwupdmgr update # Dell firmware updates
sudo reboot
0.C The terminal, for real
alias ll='ls -alF'
alias update='sudo apt update && sudo apt full-upgrade'
The machine now demonstrates
A verified OS install, current firmware, a configured shell, keyed and authenticated
GitHub access over SSH, and a version-controlled workbench repo. That is a real
developer workstation baseline, and you built every piece of it from the terminal.
Phase 0 notes
Phase 1 · Web Fundamentals
~40 hrs
HTML, CSS, and JavaScript, written by hand in files you serve yourself. You already
build websites, so this phase is a refresher that closes gaps rather than a
from-zero crawl. The projects are Odin's deliberately: Odin gives no solution code,
and the struggle is the curriculum.
1.A The refresher: shadow the build of this document
1.B Structure and style
1.C JavaScript
The machine now demonstrates
A projects directory of hand-written sites in individual git repos, served from a
local HTTP server, including a feature you shipped into an existing codebase.
Phase 1 notes
Phase 2 · Frontend
~50 hrs
The modern toolchain: Node, npm, Vite, React. Spine: Full Stack Open, which runs
every exercise on your own machine and grades by GitHub repo. Do the exercises; the
reading alone teaches nothing.
The machine now demonstrates
A per-project Node toolchain under nvm, Vite dev servers, and a set of React
exercise repos with commit histories that show the work.
Phase 2 notes
Phase 3 · Backend
~50 hrs
Express, REST, a real database daemon running on the laptop, authentication, and
automated tests. FSO Parts 3 and 4 are the spine; we swap in PostgreSQL (via FSO
Part 13) because SQL is the default of the industry and of your future products.
The machine now demonstrates
A complete local stack: a React frontend talking to an Express API backed by a
PostgreSQL daemon, all three running on this laptop, with a test suite proving the
API behaves.
Phase 3 notes
Phase 4 · Ops
~35 hrs
The layer most courses skip and most jobs assume: SSH, Vim survival, containers,
and a CI pipeline. Spine: FSO Parts 12 and 11, plus the topic skeleton of Frontend
Masters' backend path. After this phase the Latitude is not just a dev box, it is
an ops box.
The machine now demonstrates
Key-only SSH access from another machine, a containerized full stack brought up
with one compose command, and a repo where every push runs a CI pipeline to green.
Phase 4 notes
Phase 5 · AI Workflow
~20 hrs
The hand-first apprenticeship ends; the AI-assisted workflow begins, done
deliberately instead of by vibes. You now have four phases of real code for an AI
agent to work on, and the judgment to evaluate what it does. Spine: Anthropic
Academy's developer track (free, certificates included).
The machine now demonstrates
A configured AI development environment: Claude Code installed, project-level
CLAUDE.md files, at least one working MCP server, and a git history showing
AI-assisted refactors you reviewed and owned.
Phase 5 notes
Phase 6 · AI Apps
~40 hrs
Building the AI-powered apps themselves: LLM API calls, embeddings, retrieval
(RAG), tool-using agents, and your own MCP server. Lab material: Ed Donner's public
repos and Hugging Face's free Agents course.
Cost note. The course stays free, but LLM API calls meter by the
token. Options: a provider free tier, OpenRouter's free-tier models, or fully local
via Ollama (on 8 GB, stick to 3B-class models).
A few dollars of credit also goes a surprisingly long way at lab scale; your call.
The machine now demonstrates
Working AI applications end to end: an agent loop you wrote yourself, a local RAG
pipeline over real notes, and a custom MCP server bridging your own API into an AI
assistant.
Phase 6 notes
Capstone · A Full-Stack AI App, Yours
~40 hrs
One app that uses everything: designed by you, built with Claude Code as your pair
(you now know how to be the senior in that pair), React frontend, Express +
Postgres backend, at least one real AI feature, containerized, CI-tested, and
deployed. Idea candidates from your own world: a client-brief intake assistant for
chuck design, a family study coach for the hub, or this course app reborn with its
own backend and RAG over your entire logbook. Your pick; write the spec first.
The machine now demonstrates
Everything. An OS you installed, a toolchain you assembled, services you administer,
pipelines you automated, an AI workflow you command rather than obey, and a shipped
full-stack AI application. That is the course certificate, and it boots.