Stories

Told that "I don't know" was a bad signal

A senior frontend engineer, two months of interviewing for frontend, software-engineering and full-stack roles. A good deal of the work came before anyone had spoken to them: application forms with an essay box — “What excites you about this role?” — and the instruction to “Record a 2–4 minute video answering X question”.

What came after was LeetCode at two or three of the companies, and a recruiter asking them to explain concepts on the spot. Two of the processes acknowledged AI sitting inside the technical assessment. The demand underneath all of it was for an exact match: “You don’t have exactly the experience we’re looking for? Rejected.”

Then, somewhere in those rounds, nothing immediately came to mind and they said so.

“I received feedback that I had frozen because interviewing is stressful for me, and that saying ‘I don’t know’ when nothing immediately came to mind wasn’t a good signal.”

Read the full story

Read the feedback as a measurement and it stops being about the engineer. An assessment that cannot tell nervousness from incapacity is not measuring competence. It is measuring composure under observation, and then reporting the result as though the two were the same thing.

The second half of it is worse. I don’t know is the correct answer when you do not know, and on the job it is the sentence that stops a guess reaching production — the senior behaviour the round was ostensibly there to find. Marked down, it teaches the candidate to guess fluently instead, and the process ends up selecting for the one habit it would least want in the role.

Nothing else in the sequence points anywhere different. A recorded video and an essay are collected before a conversation, so the first thing scored is performance to camera. A timed algorithm round scores recall. An exact-experience filter rejects demonstrated ability for failing to match a wording. Each of these is cheap to administer at volume, and none of them is the job. The process has optimised for what it can grade, then read the grade as the person.