Intermediate · 9 steps
Get dangerous with a coding agent
This path makes you dangerous with a coding agent — not by teaching one framework, but by teaching the loop: how to drive the agent, constrain it, and catch its mistakes. You'll go from "it wrote something, I guess I'll run it" to steering a fast, reliable build-review-ship cycle, ending with a real weekend project deployed.
- 1Guide
How to actually work with a coding agent
Start with the mental model — what the agent is actually good and bad at — so everything after is deliberate rather than hopeful.
- 2Term
Context window
The one concept that explains most agent failures; understand it before you try to fix behavior downstream.
- 3Guide
Write a rules file your agent respects
Your highest-leverage move: a rules file makes the agent respect your conventions on every task instead of just this one.
- 4Guide
Structure a project an agent understands
A project the agent can navigate is a project it stops guessing in — this compounds the rules file.
- 5Guide
Prompt patterns for coding agents
Now that the environment is set up, learn the prompting patterns that get a usable diff on the first try.
- 6Guide
Debug with AI without going in circles
Agents will still get stuck; this is the method to break the loop instead of thrashing with 'still broken' replies.
- 7Guide
Review AI-written code safely
The safety net that makes speed sustainable — how to review fast without rubber-stamping.
- 8Guide
When NOT to reach for AI
Being dangerous includes judgment: knowing which tasks to do yourself is what separates competent from reckless.
- 9Guide
From idea to deployed in a weekend
The capstone — put the whole loop together on a real project and actually ship it.