The pasted answer transfers the information and skips the judgment rehearsal

An essay published on dev.to argues that answering a junior engineer's question by pasting a clean explanation out of a chat window transfers the information and nothing else ^[1]. The bill arrives later: about twelve months out, by the author's account, when you need someone to own a system and find that nobody on the team can reason about it without help ^[2]. The worked example is a partition key ^[3]. The author reconstructs what the spoken version would have contained. First, query patterns, on the grounds that a partition key is not a property of the table but a bet about how the table gets read ^[4]. Second, the time the team partitioned by customer ID and produced two million tiny files, what that did to the metadata layer, and how long it took to find ^[5]. Third, a question back to the junior about what the query pattern would look like in a year, letting them be wrong, then talking about why ^[6]. Pasted text carries the first part. It cannot carry the second, because it does not know about the incident ^[7], and it cannot carry the third at all, because the third is not information: it is the junior thinking badly in front of someone who can see where the thinking goes wrong ^[8]. Two things make this decision unusual. Nobody else on the team is being asked the question, so the senior is the only person who can make the trade ^[9]. And the senior is also the only person who pays for it, twelve months on, which is what separates it from skipping a code review, a cost that lands today ^[10]. The essay cites two 2026 papers. A study by researchers at UC Berkeley and Yale, published in Harvard Business Review in early 2026, tracked 200 employees at a technology company over eight months and found that senior engineers were spending hours reviewing and fixing AI-generated code rather than mentoring juniors ^[11]. The displaced mentoring time did not become leisure; it became review ^[12]. A Communications of the ACM paper the same month described the junior-side mechanism: a problem arrives beyond current ability, a working solution arrives in seconds, the ticket closes, and the mental model the difficulty would have built never gets built ^[13]. Every step of that sequence reads as success ^[14]. The proposed rule is deliberately cheap, on the reasoning that anything costing 20% of a week will not survive contact with the week ^[15]. Research on mentoring load, as the author cites it, puts a junior at roughly 20 to 30% of a senior's time in the early months, and the author concedes five minutes does not fit that budget and is smaller on purpose ^[16]. A dozen such conversations in a week costs an hour ^[1]. The mechanics: use the chat window to prepare rather than reply, asking it to lay out the tradeoffs, which takes about thirty seconds ^[17]; ask before you answer, so a wrong answer exposes the actual gap ^[18]; supply one thing that is in no documentation, such as the incident or the reason the team does it differently from the internet ^[19]; then stop, because it is five minutes and not a class ^[20]. The filter is whether a correct answer would still leave the junior unable to make the same decision next time; window frame syntax gets forwarded without guilt ^[21]. The essay dismisses the obvious objection, that a lead is not a tutor and juniors can read, as answerable in one line ^[22]. It then names the harder one against itself: on-demand explanation is unlimited, instant, patient and free of social cost, while senior mentoring is scarce, inconsistent and badly delivered about half the time ^[23]. The supplied text breaks off mid-sentence there, unresolved.