Sign in to keep reading
“Red Flags” is free with an account. Your progress and bookmarks follow you across devices.
Sign in to continueNEW: ML Mock & Coaching with a 20+ year experience coach. Save $20 on sessions for Premium.ML Mock & Coaching now available
NEW: ML Mock & Coaching with a 20+ year experience coach. Save $20 on sessions for Premium.ML Mock & Coaching now available
The five moves that end an AI-assisted coding round on their own — why every one of them is a trust signal rather than a skill signal, and how to recover if you have already made one.
“Red Flags” is free with an account. Your progress and bookmarks follow you across devices.
Sign in to continueThe Observer Test
Could you answer these today? Tick what you already know.
Red Flag Drills
Drill 1 — Record yourself. Screen-record a 45-minute session with audio. Watch it back looking only for the five. They are much easier to spot in someone else's session, and a recording makes yours into someone else's.
Drill 2 — Calibrate your language. For one week, never say "it works" without having just run it. Notice how often you were about to.
Drill 3 — The setup check. Right now, without looking: what model are you running, in what mode, with what autocomplete behavior? If you hesitated, go find out.
Drill 4 — Rehearse the recovery lines. Say the three recovery sentences above out loud until they are available under stress. The whole point is that they are there when you need them.
The other four dimensions are graded. This one is a gate.
You can be weak on AI Fluency and get hired. You can produce mediocre Design and be told you would grow into it. Those are scaled judgments, weighed against everything else in your loop. Red flags do not work that way, and understanding why is more useful than memorizing the list.
Every red flag has the same shape: it makes the interviewer unsure whether what they watched was real.
That is a different kind of problem from being unskilled. An interviewer who saw a weak candidate can write "weak on verification, strong on design, would place at mid-level." That is a useful report. An interviewer who is not sure what they observed cannot write anything — not a negative recommendation, just an absence of one.
This is the part candidates misread. It is rarely moral disapproval. It is that a recommendation has to be built from evidence, and these behaviors destroy the evidence.
An interviewer who cannot verify what happened will decline rather than guess, every time, because guessing is how bad hires happen and their name is on it.
Pasting from a chat window that is not shared. Alt-tabbing to a browser off-screen. Any part of your process that happens where they cannot watch.
Why it lands so hard: the round is specifically an observation of your workflow. Work that happens off-screen is not evaluated as neutral, it is evaluated as hidden — and the interviewer has no way to distinguish "checking documentation" from something worse.
The fix: share the whole screen, not a window. If you need to look something up, say so and do it in view. "Let me check the docs for the signature — I'll pull it up here." Costs nothing, removes the entire category.