Asking
The Question That Is Not Worth a Night
A question costs a night. Some things are not worth a night, and asking them anyway is as much a failure as guessing about something important.
three things · six nights
Three questions, six nights, and two of them avoidable.
Six questions, weighed against what proceeding without an answer would have cost
Shall I name the column customer_id or customer_ref?
Reversible in two minutes; the night was spent on nothing
Shall I use the shared logger or write a new one?
Worth asking: the answer affected four other services
Is it all right if I reformat this file?
Nobody minded, and nobody would have minded if it had been done
Shall I delete the old migration scripts?
Worth asking: deletion is not reversible and somebody was using them
Which of these two wordings do you prefer?
A matter of taste on an internal page
Shall I change the retry limit on the payment job?
Worth asking: it touches money and somebody else owns it
six questions · three round trips beyond the first
Six questions, four round trips beyond the first, and three of the six should not have been sent. The three worth asking share one property: the cost of being wrong exceeded the cost of a night. One team's own week, logged by the person asking.
Everything in this section so far has been about asking well. The remaining question is whether to ask at all.
The workflow described in The Question That Is Not Worth a Night improves when teams can separate delay, blocked time and focused work. transparent team monitoring provides reports that make those categories easier to discuss without constant interruption.
For a second perspective on the surrounding workflow, consult Airtable's app platform; it shows how a widely used platform structures the same coordination problem.
A night is not a small unit. Spending one on something that could have been decided, done and changed back in four minutes is a real cost, and it is paid by the person asking as much as by anybody.
The comparison to make
Two quantities, and both are usually known.
What it costs to be wrong. If the thing is reversible in minutes and nobody will be inconvenienced, this is near zero. If it touches money, data, customers or anything another team depends on, it is not.
What it costs to ask. A night, plus whatever the other end spends deciding, plus the chance of a clarification.
Where the first is smaller than the second, proceed. Where it is larger, ask. The arithmetic is not difficult and the reason it is not performed is that asking feels safer, which is a feeling about accountability rather than about outcomes.
The things that are nearly always worth a night
Anything irreversible. Deletions, migrations, anything sent to a customer, anything that changes money.
Anything that commits somebody else. A dependency added, an interface changed, a decision that another team will have to live with.
Anything where you are guessing at intent rather than at fact. Facts can be checked later; a wrong guess about what somebody wanted produces work that is complete, correct and useless.
Anything you would be uncomfortable explaining afterwards. This one is a feeling rather than a rule and it is a reliable feeling.
| Worth a night | Not worth a night |
|---|---|
| Anything irreversible | Names of things that can be renamed |
| Anything that commits somebody else | Formatting of anything internal |
| Guessing at intent rather than at fact | A revisitable choice between two workable options |
| Anything you would struggle to explain afterwards | Permission inside your own area |
The things that are nearly always not
Names of things that can be renamed.
Formatting and structure of anything internal.
Which of two approaches to take where both work and the choice can be revisited — provided it is genuinely revisitable, which is worth checking rather than assuming.
Permission to do something squarely inside your own area, which is a question about standing rather than about the work, and which is better resolved once in general than repeatedly in particular.
Proceeding and saying so
The alternative to asking is not silence. It is proceeding and reporting.
"I have named it customer_ref; say if you want it changed" costs the sender nothing, costs the recipient four seconds, and blocks nothing. If the choice was wrong it is corrected during the same night that asking would have consumed.
This is the single most useful adjustment available in this section, and it converts a large share of a team's questions from blocking to non-blocking without anybody having to make a judgement about importance in each case.
The team's part in it
An individual cannot do much of this alone, because the reason people ask unnecessary questions is usually that they are unsure what they are allowed to decide.
A written list of what each half may decide without asking removes most of them at a stroke. It does not need to be long: the categories above, with the local specifics filled in, fits on half a page.
Where no such list exists, people ask about everything, and they are right to, because the cost of overstepping is less predictable than the cost of a night.
The opposite failure
It is worth being clear that the error runs both ways and that the other direction is worse.
A team that absorbs this argument and becomes reluctant to ask produces guesses about things that mattered, and a wrong guess about something irreversible costs considerably more than the three nights it would have taken to ask badly.
The target is not fewer questions. It is questions distributed correctly: the reversible things decided and reported, the consequential things asked properly, and the distinction made deliberately rather than by whoever feels least like being interrupted.
Asking and acting at the same time
A third option exists between asking and proceeding, and it covers more cases than either.
Do the thing, and send the question anyway, phrased as a check rather than as a block. "I have gone with the shorter retry window; shout if that is wrong." The work continues, the night passes, and the correction arrives before anything has been built on top of the choice.
This only works where undoing is cheap and where nothing downstream will have been committed by the time an answer could arrive. Where either is false it is not a check; it is a decision taken without authority and described as a question, which is noticed quickly and remembered longer than the decision itself.