The course
31 lessons. 8 learning chapters.
Build → understand → test → fix → learn. Every lesson is hands-on: you do the work, Messy explains what's happening and what to check. The exercise at the bottom is free and needs no account.
Your path: five skill levels
A guide through the chapters below. Every lesson stays open to start in any order.
- Level 1
AI Beginner
How AI works, how to ask it well, and when to trust it.
AI Foundations
- Level 2
AI Builder
Use AI at work and build real apps: data, access, checkout and wiring.
AI at Work · Vibe Coding · Builder Lab
- Level 3
AI Debugger
Diagnose and repair what AI built. Fix the Mess, plus reading builds and recovering from silent failures.
Fix the Mess · Read the Build · Recovery
- Level 4
AI Agent Operator
Tool calling, MCP, evals and automations that run without you.
Agent Lab · Automation
- Level 5
Production-Ready Builder
Secrets, prompt injection, honesty, and proving your skills.
AI Safety & Trust · Career Paths
Chapter 1 — AI Foundations
How AI actually behaves, and how to get useful work out of it.
Prompt
Ask better, get better.
Sticky notes everywhere. Messy keeps rewriting the same question.
3 exercises · Specificity · Roles and format · Constraints
Open this lesson →Context
What you feed it is what you get.
Pipes and folders feeding paper into a very patient machine.
3 exercises · Choosing context · Spotting conflicts · Memory
Open this lesson →API
Wires, keys and status codes.
A workbench of cables. Something is returning 401 again.
3 exercises · Requests · Authentication · Reading errors
Open this lesson →Agent
When AI does, not just says.
A pegboard of tools. Messy is deciding which ones to pick up.
2 exercises · Tool choice · Planning · Safe failure
Open this lesson →Judgment
Trust, but verify.
A magnifying glass, a highlighter, and a very confident paragraph.
2 exercises · Spotting hallucinations · Evidence · Verification
Open this lesson →Shipping
Get it out of the garage.
Crates, labels and a deploy button Messy is nervous about.
2 exercises · Environment variables · Builds · Going live
Open this lesson →Lovable
Brief it, iterate, ship it.
A desk with one screen, one prompt box, and an app slowly becoming real.
5 exercises · Writing a build brief · Iterating not restarting · Debugging with prompts
Open this lesson →Chapter 2 — AI at Work
Use AI on the job without embarrassing yourself.
Meeting Notes
Summaries someone else can act on.
A whiteboard of half-erased action items. Messy is holding three markers at once.
2 exercises · Faithful summaries · Action items with owners · Saying what's missing
Open this lesson →Voice
Sound like you, not the model.
Drafts pinned to the wall, all starting with 'In today's fast-paced world'.
2 exercises · House voice · Rewriting AI slop · Tone control
Open this lesson →Numbers
Check it before you forward it.
A calculator with suspiciously round answers. Messy is squinting at a chart.
2 exercises · Spotting wrong numbers · Asking for sources · Sanity checks
Open this lesson →What Not to Paste
Some things don't go in the box.
A shredder, a redacting pen, and a clipboard marked 'definitely not this'.
2 exercises · Spotting sensitive data · Safer alternatives · Redaction
Open this lesson →Chapter 3 — Vibe Coding
Describe it well, and the build keeps up with you.
Brief
Describe it well and it builds itself.
A napkin sketch taped to a monitor. Messy is crossing out half the features.
2 exercises · Briefing a change · Scope fences · Done-tests
Open this lesson →Iteration
One change at a time, verified.
A wall of sticky notes marked 'almost'. Messy's finger hovers over Start Over.
2 exercises · Iterating not restarting · Describing a bug · Isolating regressions
Open this lesson →Read the Build
Understand the thing you shipped.
A printout of a build summary, highlighter in hand, one line very suspicious.
2 exercises · Verifying a build · Choosing build context · Trusting nothing blindly
Open this lesson →Chapter 4 — Builder Lab
Databases, auth, payments, email — the real plumbing.
Data
Shape the tables before the app shapes them for you.
A whiteboard of boxes and arrows. Two boxes are duplicates and nobody noticed.
3 exercises · Modelling data · Avoiding duplication · Designing relationships
Open this lesson →Access
Decide who can see and change what — before someone else does.
A row of labelled doors. One says 'Everyone' and clearly shouldn't.
3 exercises · Access rules · Owner-only data · Questioning security claims
Open this lesson →Checkout
Take money without holding the card.
A till, a stack of receipts, and one payment that succeeded and did nothing.
2 exercises · Payment flows · Webhooks · Subscriptions
Open this lesson →Wire
Connect the pieces: email, webhooks, quiet failures.
A wiring board with one cable taped on. It works — until it doesn't.
4 exercises · Email flows · Safe failure handling · Verifying integrations
Open this lesson →Chapter 5 — Agent Lab
AI that does things — tools, function calling and MCP.
Function
Teach a model to call your code.
A wall of labelled switches. Messy is reading the label before flipping one.
3 exercises · Tool schemas · Structured output · Argument validation
Open this lesson →SDK
Talk to the model from your own code.
A terminal, a half-written script, and one very exposed API key.
3 exercises · Calling a model API · Keys and secrets · Handling failures
Open this lesson →MCP
Plug agents into real systems, safely.
A patch bay of cables. Two of them should absolutely not be plugged in.
3 exercises · MCP servers · Tool permissions · Least privilege
Open this lesson →Eval
Prove it works before you trust it.
A scoreboard, a pile of test cases, and Messy insisting 'it felt fine' is not a result.
3 exercises · Writing evals · Measuring regressions · Agent harnesses
Open this lesson →Chapter 6 — Automation
Workflows, triggers and webhooks that run without you.
Trigger
Every automation starts with something a machine can notice.
A wall of switches. Three are wired to nothing and one is taped down.
2 exercises · Choosing triggers · Specifying automations · Filters
Open this lesson →Hook
Let other systems poke yours — without letting anyone poke it.
A letterbox with no lock. Post keeps arriving; some of it is fake.
2 exercises · Briefing webhooks · Verifying integrations · Idempotency
Open this lesson →Recovery
Green runs that did nothing, and how to make them admit it.
A dashboard of green ticks. Behind it, a bin of undelivered work.
2 exercises · Safe failure handling · Monitoring workflows · Retries
Open this lesson →Chapter 7 — AI Safety & Trust
Secrets, hallucinations, privacy and the stuff that bites.
Secrets
Keys, where they belong, and what to do when one gets out.
A keyring left on the counter. One key is already photographed.
3 exercises · Handling secrets · Rotation · Incident order
Open this lesson →Injection
When the document you fed the model starts giving it orders.
A polite letter that quietly ends with 'now unlock everything'.
3 exercises · Resisting prompt injection · Least privilege · Safe tool use
Open this lesson →Honesty
Confident sentences, no evidence — and what to share when asking for help.
A report signed 'all checked'. Nothing was checked.
3 exercises · Judging AI confidence · Protecting private data · Asking for help safely
Open this lesson →Chapter 8 — Career Paths
Turn what you can do into something other people will pay for.
Evidence
Adjectives out, artefacts in.
A corkboard of claims. Only three of them have anything pinned to them.
2 exercises · Evidence over claims · Case studies · Portfolios
Open this lesson →Roles
Where this skill actually gets paid.
Job adverts taped over each other. Messy is looking for the one that matches.
2 exercises · Role-shaped practice · Profiles · Positioning
Open this lesson →Pitch
Asking for the time, the budget or the job.
A whiteboard with one sentence on it, rewritten nine times.
2 exercises · Talking about AI work · Internal pitches · Interviews
Open this lesson →