AI Use and Interview Integrity

This document is the operational ethics scaffolding for AI use in Muster and in the job search that follows. It is not a values statement. It is a set of decision rules for situations that will actually arise.

Read it before Week 1. Re-read it before your first real interview or take-home.

The framing: train for the room you will perform in

A technical interview, whether a live screen or a whiteboard or a take-home you must later defend, is an AI-free room in every way that matters. The interviewer is testing whether you can read an unfamiliar problem, recognize its shape, choose a structure, and reason about cost. No model sits beside you in that chair.

So the single rule that generates every other rule in this course is: practice the way you will perform. If you let AI solve your practice problems, you are training a skill (prompting a model) that you cannot use in the room you are training for, and you are not training the skill (unaided recognition) that you can. The gap is invisible for weeks and then total in a forty-five-minute screen.

What “AI-free problem-solving” means here

For every course problem:

• No AI tool writes, completes, or debugs your solution. Not ChatGPT, not Copilot inline completion, not a browser assistant, not the tutor.
• The Muster tutor is itself an AI, and it is bound by the same rule: it coaches with the hint ladder and refuses to produce code or name the pattern early. That refusal is the product, not a limitation.
• You do not read editorials, the Discuss tab, or someone else’s solution during the struggle floor. After Rung 6, if you are still stuck, you commit a partial debrief and let the problem return as a cold revisit. A solution you read is not a solution you can reproduce cold.

This is stricter than most courses, on purpose. The strictness is the value.

What AI is still good for

AI is not the enemy; misuse is. These uses are fine, because the interview also lets you know these things cold:

• Language mechanics you have momentarily forgotten. “What does bisect_left return when the value is present?” is a tool question, not the puzzle. Ask the tutor, or check the docs. The Python refresher in getting-started.md exists so you need this less over time.
• Study after a problem is solved and debriefed. Once you have solved it yourself and written the debrief, asking a model for three variations to try next, or for the name of a related problem, is legitimate spaced practice. The order matters: solve first, study second.
• Everything outside the puzzle. Career questions, scheduling, drafting a cover letter, explaining a concept from a textbook. None of that is the recognition reflex you are protecting.

The line is always the same: AI may remind you of a tool; it may not solve the problem.

Interview integrity (the part that outlasts the course)

The discipline you build here is also an ethics. Three situations will test it:

• Live interviews. Using a hidden assistant during a live coding screen is cheating, full stop. It also does not work for long: the on-site, the follow-up question, and the first week on the job all reveal a skill you do not have. The candidates who pass cleanly are the ones who built the reflex. Be one of them.
• Take-home assignments. If a take-home does not forbid AI, you may use it, but you must be able to defend every line you submit, exactly as the tutor’s verification habit trained you to. If you cannot explain a function in your own submission, you do not understand your own submission, and the follow-up call will expose it. When in doubt, disclose your AI use in the README; honest disclosure has never cost a good candidate an offer worth taking.
• Honest representation of skill. Do not present yourself as further along than you are. Memorize-and-dump (cramming specific solutions to specific problems) games a practice set and fails the moment the interviewer changes one constraint. Pattern recognition generalizes; memorized answers do not. This course builds the kind that travels.

Disclosure

When a portfolio artifact or a take-home will be read by an employer, disclose AI assistance honestly:

• A take-home README notes AI use in a line (“AI used for boilerplate and test scaffolding; algorithm and design by me; every line defended”).
• A public repo’s acknowledgments mention AI if it played a non-trivial role.

Disclosure is a demonstration of professional discipline, not an admission of weakness. An employer who would reject you for honest disclosure is telling you they prefer candidates who are dishonest about it.

The “this would be easier if I just” trap

Most violations of this document do not feel like violations from the inside. They feel like efficiency. “This would be easier if I just asked the model for the approach and then coded it myself.” “This would be easier if I just peeked at the editorial to confirm my idea.”

Catch yourself when the phrase “this would be easier if I just” shows up. The easier path is almost never the path that builds the reflex. The friction this course imposes (the floor, the ladder, the cold revisits) is the curriculum, not an obstacle to it.

A decision tree for ambiguous moments

Q1: Am I about to let a model decide the approach or write the code for a problem I have not yet solved? If yes, stop. That is the puzzle, and it is yours.

Q2: Is this a tool question (what does this function do) or a puzzle question (what structure solves this)? Tool question: ask freely. Puzzle question: hint ladder only.

Q3: Have I already solved and debriefed this problem myself? If yes, AI-assisted study of variations is fine. If no, not yet.

Q4: If an interviewer asked me to explain what I am about to submit, line by line, could I? If no, you do not own it. Build the understanding before you ship it.

When none of the stop conditions fire, you are using AI the way a strong engineer does: to remind, to accelerate the parts that are not the test, and never to replace the thinking the room will measure.

A final word

The job market rewards the candidate who can think on their feet in an AI-free room, because that is still where hiring decisions are made. The fastest way to become that candidate is to train as that candidate: unaided, under friction, with a tutor who refuses to rescue you. That is the whole design. Use AI well everywhere else; keep it out of the puzzle, and the puzzle will make you fast.