AI principle: human in the loop
This course covers how to treat code from agentic AI tools as a draft that a person must review before merging. It explains why accountability for a change stays with whoever approves it. It also shows why review weighs more on junior developers, especially in security and business logic. A case study follows article_params and two drafts that drop it, including the silent permit! fix. The course separates confident tone from correct claims and ends with a four-question review habit applied to a bad draft and a good one.
What you'll learn
- ✓Treating an agentic tool's output as a draft that needs human review
- ✓Why accountability stays with the person who approves a merge
- ✓Where security and business logic risks concentrate for junior developers
- ✓How permit! removes the strong parameters safeguard without raising any error
- ✓Separating a confident tone from a claim you can check
- ✓A four-question review habit for every AI suggestion
An AI suggestion is a draft, never a merge-ready commit
An agentic tool, the kind AI tools: introduction (Cursor & Antigravity) covered, hands back a method for Article that looks finished: tidy indentation, a sensible name, no red underline. A pull request from a teammate who wrote and tested the same method by hand looks the same on screen. Syntax…
Read the full lesson free →Curriculum
7 lessonsAn AI suggestion is a draft, never a merge-ready commit
Explains why an agentic tool's output is a draft and what makes code ready for merge.
Who owns the result: accountability doesn't transfer
Shows that responsibility for a shipped change stays with the person who reviewed and approved it.
Why this matters more for a junior developer
Explains why confident AI drafts carry more risk for junior developers, especially in security and business logic.
Case study: article_params and the "just skip permit" suggestion
Follows article_params through a simplify draft that raises an error and a permit! fix that silently accepts every field.
Confident-sounding vs. correct: reading AI code skeptically
Separates an AI's fluent, confident tone from whether its specific claims about the code are true.
A review habit you can follow
Presents four ordered review questions and applies them to a dangerous draft and to a correct scope.
Putting it together: the reviewer is the safeguard
Connects the lessons and describes the person reviewing each suggestion as the step the tool's loop cannot close.