Skip to content

Interview PrepWorking with people · Core

Working With People ​

9. How do you review pull requests? ​

The short answer

In a fixed order, from most to least important. First, understand the goal: read the description and the ticket. Then correctness and risk: does it do what it claims, what happens on failure, retries, concurrency and bad input, and is anything security-sensitive. Then design: is it in the right place, is it simple, will it be easy to change. Then readability and naming. Leave style to linters. Comment on the code, not the person, explain the why, mark what is blocking versus a suggestion, and approve quickly when it is good enough. Keep pull requests small, and review within a day.

What to look at, in order:

PriorityLook for
1. PurposeDoes the change solve the stated problem? Is anything missing?
2. CorrectnessEdge cases, error handling, transactions, idempotency, race conditions
3. SecurityInput validation, authorization checks, secrets, injection
4. Production behaviourQueries and indexes, timeouts, logging, metrics, migrations, feature flags
5. DesignRight layer, simple, no premature abstraction
6. ReadabilityNames, structure, comments for the non-obvious

How to phrase comments:

Instead ofWrite
"This is wrong.""If the payment call times out here, we mark the order failed, but the card may have been charged. Could we look the payment up first?"
"Why didn't you use X?""Have you considered X? It would remove the manual retry loop. Not blocking."
A wall of 40 style commentsFix the linter config once

Prefix comments so the author knows what matters: blocking, suggestion, question, or nit.

10. How do you mentor junior developers? ​

The short answer

Meet them where they are, give them real work with a safety net, and help them become independent, not dependent on you. Pair on the first tasks, then gradually hand over more ownership. Answer questions with questions that teach them how to find answers, such as "what does the trace show?". Give specific, timely feedback, privately for corrections and publicly for praise. Review their code generously, explaining the why. Make it safe to ask and to make mistakes. Measure success by how quickly they need you less.

A model answer:

"When a junior joined our team last year, I set them up with a small, real bug fix in their first week, paired on it, and walked them through deploying it, so they shipped to production in days. Each week we had a 30-minute check-in where they brought one thing they were stuck on. Instead of giving answers, I asked how they would find out, and showed them our tracing and logs. After two months they owned a small feature end to end; I reviewed the design before they wrote code, which saved them a rewrite. By six months they were on the on-call rotation and reviewing other people's code. The thing I learned is to resist taking the keyboard: it is faster for me and worse for them."

DoAvoid
Real tasks with clear scopeOnly toy tasks
Explain the reasoning in reviewsRewriting their code yourself
Regular, short check-insBeing available only when things are on fire
Gradually increase ownershipThrowing them into a huge ambiguous project

11. How do you handle disagreements during a design review? ​

The short answer

Separate the idea from the person, and move the discussion from opinions to shared criteria and evidence. Start by restating the other view fairly, so they know you understood it. Agree on what matters for this decision, such as latency, cost, delivery time or operability. Compare options against those criteria, with data, a prototype or a small benchmark where possible. If it is still split, let the owner decide, record the decision and the reasons, and commit fully to it, "disagree and commit", with a point at which you would revisit it.

A model answer:

Situation. "In a design review for our notification system, a senior colleague wanted a single PostgreSQL table as the queue; I favoured a dedicated broker."

Action. "I first summarised their argument: no new infrastructure, transactions with the rest of our data. That was fair. I suggested we agree on criteria: expected volume, operational cost, and how soon we needed it. We estimated peak volume at 300 messages per second, which PostgreSQL with SKIP LOCKED handles comfortably. I proposed a quick benchmark, which confirmed it."

Result. "We went with their approach, which was the right call for our scale, and I wrote down the threshold, around 2,000 messages per second, at which we would revisit. Looking back, the useful part was moving from 'I prefer' to numbers both of us accepted."

Phrases that help: "What would change your mind?", "Which risk worries you most?", "Can we try the smallest experiment that would settle this?"

12. How do you explain a technical trade-off to non-technical stakeholders? ​

The short answer

Translate the trade-off into what they care about: customer impact, money, risk and time. Give two or three options, each with a plain outcome, a cost and a risk, recommend one and say why, and use a concrete comparison or analogy instead of jargon. Quantify where you can, such as "about one failed checkout in a thousand" or "two extra weeks". Make the decision theirs where it is a business decision, and confirm what they decided in writing.

A model answer:

"Our product manager wanted a feature in two weeks that touched payments. I explained it as two options. Option one: ship in two weeks, but if the payment provider is slow, a small number of customers, we estimated a few per day, might be charged twice and need a refund. Option two: three weeks, with protection that makes double charges impossible. I recommended option two, because each double charge costs us a support ticket and some trust, and refunds take days. They chose three weeks, and we wrote down the decision in the ticket."

Technical framingStakeholder framing
"We lack idempotency on retries.""A few customers a day could be charged twice."
"p99 latency would be 2 seconds.""One in a hundred customers would wait two seconds or more at checkout."
"This adds tech debt.""The next feature in this area will take about a week longer."
"We need a read replica.""About £400 a month so search stays fast during sales."

13. How do you prioritize reliability improvements against feature delivery? ​

The short answer

Make the trade-off visible and measurable, not a matter of opinion. Agree SLOs with the business, such as 99.9% successful checkouts, and an error budget: the failures that target allows. While the budget is healthy, features win; when it is being spent fast, reliability work takes priority. Rank reliability work by user impact and likelihood, using incident data, and attach each item to a business risk. Reserve a steady share of capacity, such as 20%, for reliability and debt, so it is not left until something breaks.

A model answer:

"On our team, every incident review produces actions, and I keep them in one ranked list with an estimate of user impact: how many failed checkouts, how often. In planning, I bring the top three with that number. Last quarter our error budget for checkout was nearly gone by week six because of payment timeouts, so I proposed pausing one feature for a sprint to add timeouts, a circuit breaker and a fallback provider. I showed the cost of the incidents so far, roughly 4,000 failed orders, against a one-week delay. The business agreed, and the next six weeks used less than 10% of the budget."

SignalDecision
Error budget healthyShip features; normal reliability allocation
Budget burning fastFreeze risky changes; reliability work first
Same incident twiceThe fix jumps the queue
A risk with low chance but huge cost, such as no backupsDo it now regardless

More on SLOs and error budgets in SLIs, SLOs and SLAs.

14. How do you handle a tight deadline or unclear requirements? ​

The short answer

For unclear requirements: ask who the users are and what problem they have, write down your understanding and the open questions, confirm it with the person who asked, and build the smallest version that answers the biggest question first. For a tight deadline: split the work into must-have and nice-to-have, agree on scope, not quality, cuts with the stakeholder, flag risks early rather than at the end, and keep correctness, security and data safety non-negotiable. Communicate progress before anyone has to ask.

A model answer:

"We were asked to add 'gift cards' in three weeks with no written spec. On day one I wrote a one-page note: what I thought a gift card was, the five questions I could not answer, such as partial redemption and expiry, and what I proposed to leave out. A thirty-minute call settled four of them. We shipped purchase and full redemption in week two, then partial redemption in week three. The things I would not cut were idempotent redemption and a ledger for balances, because those protect real money."

15. What questions should you ask the interviewer? ​

The short answer

Ask questions that show you think like an engineer on the team and that help you decide. Good ones are about how the team works, what is hard right now, how success is measured, and how decisions are made: "What does a typical week look like for this team?", "What is the biggest technical challenge you face this year?", "How do you handle incidents and on-call?", "How are technical decisions made and recorded?", "What would make someone successful in the first six months?". Avoid questions answered on the website, and leave salary to the recruiter stage.

TopicQuestion
The work"What would I work on in my first three months?"
Challenges"What is the hardest technical problem the team is facing now?"
Engineering practice"How do changes get to production, and how often?"
Operations"How does on-call work, and how often is someone paged at night?"
Decisions"How are design decisions made and written down?"
Growth"How do engineers here grow, and what does good look like at the next level?"
Culture"Can you tell me about a recent disagreement on the team and how it was resolved?"
Fit"Is there anything about my background you would like me to expand on?"

Combine chapters into one PDF

Pick the chapters you want. They print in the order you pick them.

Interview Prep
Talking About Your Work
Your stories
Working with people
Other topics
Nothing picked yet