Vibe coding just hit college classrooms.
The grading rubric scared me....
I read a post from a CS student this week.
First assignment: “vibe code” a banking app with an AI tool.
A big part of the grade was how well they “interacted” with the AI and whether the AI tool was installed.
So, if they wrote it all themselves their grade would be worse.
Huh?
This is the moment we decide what kind of builders we want to train.
AI in class is good.
Grading "vibes" over engineering is not.
What we need to teach and grade instead:
- Clear specs and constraints
- Interfaces and data contracts
- Test plans and edge cases
- Traceability from requirement to commit
- How AI followed the plan, not how much it chatted
Because coding is not vibes.
It is architecture, process, and feedback loops.
Give students a structure, then let AI help them execute it.
How I handle this in real projects:
1. I write context first
2. I break ideas into epics, user stories, and tasks
3. Then I let AI code against that plan
I use
Precursor for the planning step.
It turns a rough idea into a machine-readable plan that tools like Cursor can actually follow. Night and day difference in quality and sanity.
If I were running this class, I would grade on:
- Quality of the context doc
- Coverage of tests and edge cases
- Clarity of the handoff to AI
- Maintainability of the final codebase
AI should make better engineers, not shortcut the craft.
What do you think:
Should schools grade AI interactions, or the engineering behind them?
If you teach CS, would you try a context-first rubric this semester?
#softwareengineering #AI #developers #education #buildinpublic