9 · AI II — the coding loop

Tuesday, Sep 22, 2026

Materials for this session are not published yet. They appear here before class.

Objectives

By the end of this session you can:

• Run the five-step delegation loop end to end.
• Scope a delegation: “write this function,” not “do my homework.”
• Read a model’s diff and judge it.
• Convert a one-off verification into a permanent check.

What we cover

• Inline and agentic coding tools.
• The five-step loop: explain the goal and the failure mode → delegate → predict the output → verify against the prediction → institutionalize the check.
• Two working rules for letting a model touch your files: start from a clean git state, so the diff shows exactly what the model did — and trust the diff, not the transcript: what it did, not what it says it did.
• Live: delegate a function from last week’s homework, predict what it should do, and catch the model’s bug by reading its diff — the git skills from session 7 are why this works.
• The naming, structure and named-constant habits, put to work: they are how you brief an assistant on how your code should look.

Verification habit. The loop itself — “runs clean” is not “is right.”