Start Here
Two lessons, one sitting. Fix the destination in your own words, then learn how the course is put together — because used like a video series it produces exactly what video series have produced before.
-
0.1
What You'll Be Able To Do
- The 7am failure, handled
- Reading code, not writing it
- A dated note, sealed to week 8
-
0.2
How This Course Works
- Six sections, every lesson
- The artifact, not the feeling
- stuck.md: trying, tried, said
Setup & Vocabulary
Hands on the tool, and learning the nouns. Installed and working on real files by Friday — then the eight words everything after this depends on, defined while they're still attached to something you watched happen.
-
1.1
What Claude Code Actually Is
- You stop being the courierWatch →
- It reads, runs and fixes
- Count your copy-pastes
-
1.2
Which Surface, And Which Plan
- App vs Code vs CoworkWatch →
- One test: does it touch files?
- Cheapest paid tier, never API
-
1.3
Install And First Run
- Install, sign in, grant accessWatch →
- A sandbox folder of throwaways
- Read the permission prompt
-
1.4
What A Project Is
- A project is a folderWatch →
- Sees inside it, nothing outside
- Two-folder demo: the wrong room
-
1.5
Where Things Live
- The home folder and ~Watch →
- Reading a path left to right
- Path bar on, never Downloads
-
1.6
Sessions And The Commands That Matter
- /help, /context, /compact, /clearWatch →
- Plan mode before multi-file edits
- Clearing wipes chat, not files
-
1.7
Permission Modes
- Ask-everything to don't-ask dialWatch →
- At the prompt, mistakes are free
- Set mode by folder, not mood
-
1.8
The Vocabulary
- Project, session, skill, pluginWatch →
- Connector, MCP, agent, subagent
- Each one, found on screen
AI Fundamentals
Why the thing you used all week behaves the way it does. Theory lands second on purpose, so every lesson here explains something you have already watched happen. Two failure modes, then the lever, then the fix.
-
2.1
What An LLM Actually Is
- Continuing text, not looking upWatch →
- Confidently wrong, by design
- Always check numbers and names
-
2.2
Context — Why It Forgets
- It re-reads the whole conversationWatch →
- Context rot: slower, costlier, worse
- One task, one session
-
2.3
Prompting That Lands First Time
- Context, task, constraints, output shapeWatch →
- The first prompt beats five corrections
- Recovery: rewrite the opening prompt
-
2.4
Memory And RAG In Plain Language
- Putting in front vs going to findWatch →
- Retrieval, without the acronym
- Sourced vs unsourced, side by side
-
2.5
The Ladder
- Ask, act, remember, repeat, run aloneWatch →
- Coding isn't a rung
- Five real tasks, placed
Structure
How Claude knows your world. A map at the top, a room per workspace, tools underneath — then you build your own and audit it against the seven ways people get it wrong.
-
3.1
The Three Layers
- An office, not a folderWatch →
- Map every time, rooms on demand
- Sketch two or three, build nothing
-
3.2
CLAUDE.md — The Map
- What you'd tell a new hireWatch →
- One screen, or every session bloats
- Ask Claude to describe it back
-
3.3
The Routing Table
- Four columns: task, folder, read, skillsWatch →
- Otherwise it reads everything, or guesses
- Same question, before and after
-
3.4
CONTEXT.md — The Rooms
- Loads only in that roomWatch →
- 80% the work, 20% behaviour
- Prove it with /context numbers
-
3.5
Build Yours, And The Seven Mistakes
- Seven failures, every one silentWatch →
- Stale structure is worse than none
- Audit yours: pass, fail, evidence
Version Control
You cannot be trusted with real things until you can undo. Git in plain language, a checkpoint you can return to, and a copy that survives the laptop. This lands before anything gets built.
-
4.1
What Git Actually Is
- Cmd+Z forgets; git doesn'tWatch →
- Ask in English, not git commands
- Break it, then bring it back
-
4.2
Commits And History
- Write messages, not "update"Watch →
- Recover one file, not everything
- Commit before multi-file changes
-
4.3
Branches And GitHub Backup
- Tracing paper over the drawingWatch →
- Wreck one branch, merge another
- Private repo, pushed every day
Build One Real Thing
Scaffolding for Level 6, not a craft exercise. Brief, build, check, deploy — no design systems, no polish. You cannot read a system you didn't build if you have never built one.
-
5.1
Brief Before Build
- "I want a thing" into a specWatch →
- Scope and constraints, up front
- A brief Claude can execute
-
5.2
Build It
- From brief to runningWatch →
- No polish, just working
- A build on your own machine
-
5.3
Checking Claude's Work
- Plausible-but-wrong is the riskWatch →
- Verify before it ships
- Find one defect yourself
-
5.4
Deploy It, Then Break It
- Put it on a live URLWatch →
- Break it on purpose
- Ship the fix yourself
The differentiator
Operate It
Everything above is building. This is keeping alive something other people depend on — reading a system you didn't write, deciding what to fix and what to escalate, and knowing what you never touch. It is the job most people actually get hired for, and the one no other course covers.
-
6.1
Reading A System You Didn't Build
- A repo you've never seenWatch →
- An accurate map in an hour
- Write the explanation down
-
6.2
Errors Are Data
- Read a log line properlyWatch →
- Stack trace to actual cause
- Diagnose from the log alone
-
6.3
Read-Only First
- Investigate without the power to breakWatch →
- Read-only access as the default
- Answer a real question safely
-
6.4
The Smallest Safe Change
- The minimum fix, nothing extraWatch →
- Prove it worked afterwards
- One line, merged
-
6.5
Monitoring
- Find out before they tell youWatch →
- What's worth watching
- One check on a schedule
-
6.6
Triage — Fix It Or Escalate It
- Decide in two minutesWatch →
- Fix it, or hand it up
- A written triage rule
-
6.7
Never-Touch Lists
- What you never touchWatch →
- Why each item is on it
- Your own list, written down
-
6.8
Keeping Secrets Secret
- Keys and tokens, handledWatch →
- What never enters a repo
- Audit the repo you have
-
6.9
The Daily Loop
- Fifteen minutes, every morningWatch →
- Same checks, same order
- A checklist actually in use
Skills & Leverage
Where competence becomes leverage. Vet and amend other people's skills, write your own, connect real services through MCP, and get work happening while you are not at the machine.
-
7.1
What A Skill Actually Is
- Instructions, not magicWatch →
- Read the file end to end
- Annotate what each part does
-
7.2
Finding And Vetting Skills
- Stars are not safetyWatch →
- What to read before running
- Three vetted, three rejected
-
7.3
Install, Test, Amend
- Never run one off the shelfWatch →
- Test before it touches real work
- Change it to fit your setup
-
7.4
Build Your First Skill
- Pick a task you repeatWatch →
- Write the steps down once
- A skill that runs
-
7.5
What Makes A Skill Good
- Two skills, side by sideWatch →
- Say why one outperforms another
- Rewrite 7.4, better
-
7.6
Connectors And MCP
- CLI vs API vs MCPWatch →
- No hand-waving on the difference
- One working connection
-
7.7
Build A Custom MCP
- Wrap a service cleanlyWatch →
- Expose only what's needed
- A local MCP running
-
7.8
Subagents And The Critic Loop
- Fan work out in parallelWatch →
- A second model reviews it
- Nothing ships unreviewed
-
7.9
Hooks, Routines And Scheduled Agents
- Hooks fire on eventsWatch →
- Routines run on a clock
- Something happens without you
Memory That Lasts
Sessions end. Knowledge shouldn't. Where a given fact belongs, a vault that outlives any one conversation, and a way to search your own material properly.
-
8.1
Three Levels Of Recall
- Three levels, not oneWatch →
- Place a fact at the right level
- A memory map of your work
-
8.2
Obsidian As The Durable Brain
- Knowledge outliving the sessionWatch →
- Plain files Claude can read
- A working vault
-
8.3
Making It Searchable
- Search your own materialWatch →
- Find it without remembering where
- A queryable index