The journey so far: two weeks, one mental model
Before we sprint into the week-three finale, let’s stop and look back. Ten classes, dozens of tools — but underneath them all sits one simple map. Get that map clear in your head and every technique so far (and everything coming) clicks into place.
You have covered a lot of ground. If it feels like a blur of tools — Cursor, Claude Code, OpenRouter, MCP, skills, plugins, Jira, FeatureDev — that is completely normal. This class is a deliberate pause. We are not learning anything new today; we are tying it all together so the finale week makes sense.
01Why stop and revise?
New tools are easy to collect and easy to forget. What lasts is the mental model — the small set of ideas that explain why each tool exists and when to reach for it. A good revision does three things:
One map, not ten
Replace ten separate memories with a single picture you can redraw from memory.
Know what you don’t know
Spot the topics that are still fuzzy while there is still time to firm them up.
Ready for the finale
Week three (agents, hooks, swarms) only makes sense once weeks one and two are solid.
Read each section, then cover it and try to explain it out loud in your own words. If you can teach it to an imaginary beginner, you own it. If you stumble, that is exactly the page to reopen from Classes 1–10.
02The eight stages, revisited
Back in Class 2 we met Steve Yegge’s eight stages — a ladder describing how a developer’s relationship with AI evolves. It is the backbone of the whole course, so let’s redraw it.
| Stages 1–4 | Stage 5 (where week 2 lived) | Stages 6–8 (the finale) |
|---|---|---|
| Chatbots, autocomplete, and AI-native IDEs like Cursor. You steer closely, line by line. | The CLI agent (Claude Code): you delegate whole tasks, let diffs scroll by, and verify results. | Multiple agents, then many agents (swarms), then agents that manage other agents (orchestration). |
Every stage is about moving up one level of delegation — handing the machine more, while you move from typist to reviewer to manager. Nothing about the ladder says you must live at the top: you pick the rung that fits the task and your risk appetite.
03Control vs. autonomy — the master dial
If there is one idea to carry into the finale, it is this dial. Every technique in the course sits somewhere between tight control (you review every step) and full autonomy (you set it going and walk away).
04The two mindsets
We framed this in Class 3 and it has held up all the way through. Which mindset you adopt depends on the maturity of the project and how much a mistake would cost.
Trust, but verify
Plan → execute → review → test. Spec-driven, close supervision, everything tracked in Jira/GitHub. This is how we built the Pre-Legal SaaS.
Just go with it
YOLO builds, Ralph loops, quick throwaway prototypes. Fast, a bit chaotic, and perfect for learning the shape of a problem.
Whichever mindset you pick, you are accountable for the code you ship. The agent is a tool that accelerates you; it is not the boss. You are the boss, and it is your work product — a theme that will matter even more when you have a dozen agents running at once.
05The arc of the course, in one breath
Week 1 — See it, then understand itClasses 2–4
Generate an app from a sentence, learn how LLMs actually work, write a great agents.md, and build your first real app with an agent.
Week 2 — Go professionalClasses 5–10
Choose models deliberately, master the Claude Code CLI, add MCP / skills / plugins, run the Jira→FeatureDev→PR workflow, and ship a full SaaS end to end.
Week 3 — Go bigClass 12 onward
Sub-agents, hooks, custom plugins, and eventually swarms and orchestration — controlled chaos, with a capstone multi-agent app.
✓ Key takeaways
- The eight stages are a ladder of delegation; weeks 1–2 took you to stage 5, week 3 climbs to 6–8.
- Every technique sits on one control↔autonomy dial — slide it to match the stakes.
- Two mindsets: trust-but-verify for mission-critical work, just-go-with-it for exploration.
- You are always accountable for what ships, no matter how many agents help.
- Next up: revise the foundations — how LLMs work and how to talk to them.