Module 1 of 1 · One of 4 levels
Lesson · Behavioral Interview Preparation
Behavioral interviews at SDE-2
For engineers who own features and projects with some independence. How interviewers assess ownership, ambiguity, disagreement and work across teams, with original questions, common pitfalls, likely follow-ups and worked examples.
An SDE-2 owns features and projects from start to finish with a fair amount of independence. The work typically runs for weeks to a few months, comes with some ambiguity about what exactly should be built, and depends on one or two other people or teams. Within a project, you decide how to build it and often help decide what it should include; the team's broader direction is still set by someone else. Your influence reaches past your own code: into design reviews, into the work of the teams you depend on, and into how quickly newer engineers on the team find their feet.
Behavioral questions at this scope tend to listen for whether you can run a piece of work to completion without being managed day to day. That covers surfacing risk early and honestly, working through disagreement with peers and more senior engineers without relying on authority, making sound quality calls under time pressure, and helping the people around you get better. This is an editorial summary of how behavioral questions are commonly used at this scope, not a description of any particular company's process.
Compared with an early-career answer, more of an SDE-2 story happens before and after the code: working out what done means, agreeing it with the people who asked, coordinating with whoever you depend on, and checking after launch that the work did what it was meant to. Results carry more weight too. Where you measured something and know how it was measured, say so. Where you didn't, describe the change someone could observe, and say plainly what you don't know.
Two calibration errors are common. One is pitching an SDE-2 answer as a multi-team strategy story; a single project owned well, with its trade-offs explained, is the right size here. The other is mistaking heroics for ownership. A story about working through the night to rescue a launch tells the interviewer less about you than one about spotting the risk three weeks earlier.
Choosing a level. Pick the Lesson that matches the scope of the work the role will expect and the examples you can honestly give, not your current title. Employers name and divide levels differently, so treat SDE-2 here as a description of scope rather than a rung on any company's ladder. If most of your strongest examples are well-scoped tasks that someone else defined, the SDE-1 Lesson in this Track will likely serve you better, and there is no need to stretch those stories to fit here. If your work regularly sets technical direction across several teams or quarters, look at SDE-3; if you are interviewing to lead a team of engineers, the first-line engineering manager Lesson. The four are alternatives, and someone between two levels can reasonably use both.
The clearest way to tell this Lesson from the SDE-1 one is the question each is mostly trying to answer:
| Level | The question behind most behavioral questions |
|---|---|
| SDE-1 | Can you complete your piece of work, make sound decisions within it, ask for help at the right moment and learn from feedback? |
| SDE-2 | Can you work out what success means, manage dependencies and risk, make trade-offs visible and own the outcome through launch and after it? |
Both describe scope, not stages. An SDE-2 answer is not an SDE-1 answer with bigger numbers; it covers more of the work's life, from defining it to checking what happened after it shipped.
The answers below assume the structure explained in Structuring behavioral interview answers with STAR: Situation, Task, Action and Result, with Reflection where it adds something. This Lesson is about what belongs inside that structure at SDE-2.
What fits at this level#
| Theme | What a fitting SDE-2 story usually shows | Signs it's pitched wrong |
|---|---|---|
| Ownership | A project you carried from an unclear start through launch and what happened after | The story ends at "merged", or "owned" means "wrote most of the code" |
| Risk | A problem you saw early and raised with options attached | Bad news delivered on the deadline, or absorbed through overtime |
| Disagreement | Evidence, a clear decision owner, and commitment once the call was made | Winning as the point, or a concern you never voiced |
| Quality | A reasoned trade-off, shared with the people who carry the risk | "I never cut corners", or a shortcut nobody else knew about |
| Other people | Peers and newer engineers more capable because of something specific you did | Being the person everyone depends on |
The notes under each question describe what to look for in your own experience. They are not a template for one correct story, and different stories or tellings can answer the same question well.
A project you owned from start to launch#
The question: "Tell me about a project you owned from the first task to the launch."
What the interviewer is assessing
End-to-end ownership: how you planned the work, tracked it, handled what went wrong and finished it, including the part after launch that many answers leave out.
What strong evidence sounds like
What "owned" meant: who asked for the work, and what you were accountable for. How you broke it down and kept track of progress. At least one point where the plan met reality and you changed it. How you launched, for example behind a flag, in stages or with specific things to watch, and what happened afterwards: adoption, bugs, follow-up work, a number you tracked. A credible outcome is one you can explain how you know, whether it's a measurement or an observable change.
Common pitfalls
- Treating "owned" as "wrote most of the code".
- A story that stops at merge or launch day.
- A plan that went perfectly, which leaves nothing to discuss.
- A chronological task list with no decisions in it.
Likely follow-ups
- "What would you have cut with half the time?"
- "How did you decide it was done?"
- "What broke after launch?"
- "Who else did you need, and how did you keep them informed?"
When the plan slipped#
The question: "Describe a time the plan slipped and you had to tell the people relying on it."
What the interviewer is assessing
Whether you communicate risk early and honestly, and whether you arrive with options rather than only a delay.
What strong evidence sounds like
When you first saw signs of trouble, and how soon after that you said so. Who you told and in what form. The options you laid out, each with its cost: reduce scope, move the date, add help, ship in phases. A decision made by the person who owned it, which may not have been you. How you tracked the new plan. The cause of the slip can be your own estimate; owning that is part of a strong answer, not a weakness in it.
Common pitfalls
- Telling people on the day of the deadline.
- Blaming another team as the whole of the story.
- "I worked harder and caught up", which hides the risk management the question is about.
- Bad news with no options attached.
Likely follow-ups
- "When did you first suspect it?"
- "Why not say so earlier?"
- "How did the people relying on it react?"
- "What do you estimate differently now?"
A technical disagreement you didn't get to decide#
The question: "Tell me about a technical disagreement where you were not the person who made the final call."
What the interviewer is assessing
Influence without authority. The interviewer wants to see you bring evidence to a disagreement, make your case to the person who decides, and then commit to the decision, especially if it went against you.
What strong evidence sounds like
The technical stakes, explained briefly enough that a non-specialist could follow. How you came to understand the other position. The evidence you brought: a prototype, a measurement, a precedent from your own system, a failure analysis. Where and how you raised it: in the design review, in writing, one to one. Who decided, and how you supported the decision afterwards. If you can, say what happened later and whether either of you turned out to be right; the story doesn't need you to have been.
Common pitfalls
- Winning as the point of the story.
- Reopening the decision after it was made.
- Saying nothing at the time and "I told you so" later.
- A disagreement about style presented as one about architecture.
Likely follow-ups
- "What if the decision had turned out badly?"
- "What was the best argument against your view?"
- "What did committing to the decision look like in practice?"
- "Would you raise it differently now?"
Depending on a team with other priorities#
The question: "Describe a time your work depended on another team that had different priorities."
What the interviewer is assessing
How you negotiate a dependency you don't control: making the cost of delay visible, respecting the other team's constraints and having a fallback.
What strong evidence sounds like
You understood why the other team's priorities differed, rather than treating them as obstructive. You made the dependency concrete: an agreed interface, a date, the smallest version of what you needed. You offered something, perhaps to write part of the change yourself or to accept a reduced first version. You had a fallback if it didn't arrive. If you escalated, it was after talking to the team directly, and you told them you were doing it.
Common pitfalls
- The other team cast as the villain.
- Escalation as the first move.
- No plan B.
- A dependency that was only noticed once it had already caused a delay.
Likely follow-ups
- "What did the other team need from you?"
- "At what point would you have escalated?"
- "What was your fallback?"
- "How is the working relationship now?"
Hard feedback for a peer#
The question: "Tell me about a time you had to give a peer difficult feedback on their work."
What the interviewer is assessing
Candour with care. Can you say something uncomfortable to a colleague at your own level in a way that changes something, without damaging the working relationship?
What strong evidence sounds like
Why it mattered enough to raise. You said it privately and specifically, about the work rather than the person. You listened, and perhaps learned context that changed what you asked for. You followed up, and you can describe what happened, including a partial outcome. Code review feedback counts if it went beyond a single comment, for example a pattern of changes merged without tests that kept causing regressions for others.
Common pitfalls
- Going to the person's manager first.
- Feedback so softened that it is unclear what was asked.
- A harsh delivery justified as "just being direct".
- Feedback that was really about a personal preference.
Likely follow-ups
- "How did they respond?"
- "What would you have done if nothing changed?"
- "Why you rather than their manager?"
- "Have you received feedback like that yourself?"
Improving how the team works#
The question: "Describe a time you improved how your team works, not just what it ships."
What the interviewer is assessing
Whether you notice a problem in how work gets done, can get other people to adopt a change, and can show that it helped.
What strong evidence sounds like
A concrete friction with visible symptoms: reviews waiting days, a flaky test suite everyone re-runs, on-call handoffs that lose context, a release checklist that lives in one person's head. You proposed a change rather than imposing it, tried it on a small scale, and adjusted. Others adopted it. You have some evidence that it helped, and you're candid about any part that didn't stick. A small change that lasted is stronger than a large one that faded.
Common pitfalls
- A process imposed on the team because you preferred it.
- A tool you liked, adopted by nobody else.
- No evidence beyond "people liked it".
- A one-off fix described as a process change.
Likely follow-ups
- "Who pushed back, and why?"
- "Is the team still doing it?"
- "How did you know it helped?"
- "What did you drop from the original idea?"
A serious bug or incident#
The question: "Tell me about a serious bug or production incident you helped resolve."
What the interviewer is assessing
How you behave under pressure: whether you stay methodical, keep people informed and follow through on the causes once the immediate problem is over.
What strong evidence sounds like
Your role, stated precisely: leading the response, investigating, writing the fix or handling communication. How you weighed stopping the damage against finding the cause, and which came first. Who you kept informed and how often. What the cause turned out to be, explained without blaming individuals. Which follow-up actions actually happened, rather than the ones that were written down. If the incident was partly caused by your own change, say so early.
Common pitfalls
- A hero story in which you alone saved the day.
- Blame for whoever made the change that triggered it.
- A story that ends when the symptoms stopped.
- A technical walkthrough with nothing about how you worked with people.
Likely follow-ups
- "What did you tell people, and when?"
- "What would you do differently in the first fifteen minutes?"
- "Which follow-up actions got done?"
- "Could it happen again?"
Turning a vague goal into something concrete#
The question: "Describe a task with a vague goal that you had to turn into something concrete."
What the interviewer is assessing
Whether you can handle ambiguity by scoping it: finding the need behind the request, defining success in terms someone can check, and confirming it with the person who asked.
What strong evidence sounds like
The original request, close to how it was phrased. How you found the need underneath it, by talking to users or the requester, looking at data or both. A written definition of success that the requester agreed to. What you explicitly put out of scope, and how you told people. Whether the result met the definition, and what happened to the parts that didn't fit. The first worked example below answers this question.
Common pitfalls
- Building what you assumed they meant.
- Scoping for so long that nothing shipped.
- Changing the goal without telling the person who asked.
- Declaring success against a definition you wrote after the fact.
Likely follow-ups
- "How did you decide what was out of scope?"
- "Who agreed to the definition?"
- "What changed along the way?"
- "How did you measure it?"
Helping a newer engineer get productive#
The question: "Tell me about helping a newer engineer get productive."
What the interviewer is assessing
Whether you help others grow, adjust your help to the person and leave them more capable rather than reliant on you.
What strong evidence sounds like
What this particular person needed, as opposed to a generic onboarding checklist. How your help changed over time, for example pairing closely at first, then reviewing, then stepping back. A specific moment where you held back so they could work something out themselves. How they progressed, with the credit given to them. Something you learned about helping people, which can come from what didn't work.
Common pitfalls
- Presenting yourself as the reason they succeeded.
- "I answered their questions" with nothing more specific.
- Doing their work for them.
- Describing the newer engineer in a way that sounds condescending.
Likely follow-ups
- "What did you do differently for them than you would for someone else?"
- "How did you decide when to step back?"
- "What did they teach you?"
- "How do they work now?"
Shipping sooner or shipping more carefully#
The question: "Describe a time you chose between shipping sooner and shipping it more carefully. How did you decide?"
What the interviewer is assessing
Your judgement on speed against quality, whether the people who carry the risk knew about it, and whether any shortcut was tracked and paid back.
What strong evidence sounds like
The actual risk, and whether the consequences could be undone. Who you told and how, and who agreed to the risk. What you cut and what you deliberately protected. How the shortcut was recorded, and whether the clean-up happened, including if it happened late. Either choice can be the right one; what's being assessed is the reasoning. The second worked example below answers this question.
Common pitfalls
- "I don't compromise on quality", which avoids the judgement the question asks for.
- Shipping fast with a risk nobody else knew about.
- Deciding alone something that wasn't yours to decide.
- Clean-up work "tracked" in a backlog nobody reads.
Likely follow-ups
- "What would have made you choose the other way?"
- "Who signed off on the risk?"
- "Did the clean-up happen?"
- "What was the worst case, and how likely was it?"
Worked example: "make order history faster"#
Illustrative example: a composite written for teaching, not a real candidate's experience.
The question: "Describe a task with a vague goal that you had to turn into something concrete."
Situation. An engineer owns the backend for the account area of an online retailer's website. The product manager forwards a batch of support tickets complaining that the order history page is slow, and asks the engineer to "make it faster" this quarter. There is no target and no definition of slow. The same API also serves the company's mobile app, which another team builds and releases on its own schedule.
Task. Turn "make it faster" into something specific enough to build, agree it with the product manager, and own it through launch.
Action.
- They read the tickets and look at the page's server-side timings before touching any code. The slowness is concentrated in accounts with long order histories; for most customers the page is fine. The cause is visible in the code: the endpoint loads every order a customer has ever placed and the page paginates them in the browser.
- They write a one-page proposal: the problem, who is affected, and a target of the first page of results in under one second at the 95th percentile for the accounts with the longest histories, measured from server timings. They put one thing explicitly out of scope: the order export feature, which is slow for a separate reason.
- They review the proposal with the product manager, who agrees the target, and with the support lead, who asks for a way to tell affected customers once it's fixed.
- Server-side pagination changes the shape of the API response, which would break the mobile app. The engineer proposes a new versioned endpoint, keeps the old one running until the app has moved over, and agrees a date for that with the mobile team.
- They ship in two steps behind a flag: server-side pagination first, then a database index for the query after checking its plan. They roll out to a small share of traffic, then to everyone.
- For two weeks after launch they compare the timings against the target and send the support lead a short summary for the support team to use.
Result. The web page meets the agreed target. The mobile app moves to the new endpoint about a month later on its own release schedule, so app users see the improvement later than web users. Support stops receiving complaints about the history page itself; complaints about the export continue, and the engineer files them with their timings so the product manager can prioritise that work separately.
Reflection. Writing the target down before starting made the finish line obvious and made the export cut easy to defend. The engineer would bring the mobile team into the proposal review rather than after it, because a change to the response shape was predictable from the start.
Sample wording. One possible spoken version, offered as a sample to adapt rather than to memorise:
"Our product manager forwarded a batch of support tickets about the order history page being slow and asked me to make it faster this quarter. There was no target, so before writing any code I read the tickets and looked at our server timings. The problem was really about customers with long histories: we loaded every order they'd ever placed and paginated in the browser. I wrote a one-page proposal with a concrete target, the first page in under a second at the 95th percentile for those accounts, and I put the export feature explicitly out of scope because it was slow for a different reason. The product manager agreed the target, and the support lead asked for a way to tell customers when it was fixed. The fix changed the API response shape, which would have broken the mobile app, so I added a versioned endpoint and agreed a migration date with the mobile team. I shipped pagination and then an index behind a flag, rolled it out gradually, and tracked the timings for two weeks. Web hit the target; the app caught up about a month later on its own release. Export is still slow, but it's now a separate, prioritised piece of work with numbers attached. If I did it again, I'd bring the mobile team into the proposal review, because that response change was predictable."
Why it works. The candidate turns a vague request into a definition someone else agreed to, then holds the work to it. Ownership runs from reading support tickets to checking the numbers after launch, which is the span an SDE-2 story should cover. The coordination with the mobile team is handled before it becomes a problem. The result is partial in two honest ways, the delayed app and the deferred export, and each is explained rather than buried. The scope is right for the level: one feature area, one dependency and a real decision about what not to do. The figures are plausible for the scenario and the candidate can say how each was measured.
How a weaker version differs. A weaker version jumps to the fix: "The page was slow, so I added pagination and an index and it got much faster." It may describe good engineering, but there is no definition of success, no agreement with the person who asked, no mention of the app it would have broken and no way to check "much faster". A different weak version overreaches, turning the same work into a vague story about rearchitecting data access across teams, which invites follow-ups the candidate can't support.
Worked example: renewal reminders on a fixed date#
Illustrative example: a composite written for teaching, not a real candidate's experience.
The question: "Describe a time you chose between shipping sooner and shipping it more carefully. How did you decide?"
Situation. A subscription business announces a price change that takes effect on a fixed date, and commits to emailing every affected subscriber two weeks beforehand. An engineer who owns the notifications work has three weeks to deliver. The fastest path is a one-off batch job that queries subscribers and sends the emails. What the engineer would normally build is a proper scheduled-campaign feature, with templates the marketing team can edit themselves, which would take longer than three weeks.
Task. Deliver on the date, decide which risks are acceptable, and make sure the people who carry those risks know about them.
Action.
- They separate what could be undone from what couldn't. A template stored in code is awkward but can be changed in a deploy. An email sent to the wrong person, or sent twice, can't be recalled.
- They propose a plan to the tech lead and the product manager: ship on time with the template hard-coded and no self-service editing for marketing, but spend the care on the sending path. That means a dry-run mode that produces the recipient list without sending anything, a per-subscriber idempotency key so a retried job can't send a second email, and a staged send to internal accounts first, then a small share of subscribers, then everyone.
- They write the trade-off into the ticket: what was cut, the accepted cost (marketing needs an engineer to change the copy for about a month), and a follow-up ticket for the template editor with a target sprint.
- Two days before the send, they run the dry run and compare the count with the finance team's count of affected subscribers. The numbers don't match: subscribers with paused subscriptions are included. The product manager confirms that paused subscribers should receive a different message later, so the engineer excludes them, re-runs the dry run and gets matching counts.
- On the day, they send in the planned stages and check the send log at each one.
Result. The reminders go out on the date. The send log shows one email per subscriber, and paused subscribers receive none. The accepted cost is real: marketing needs an engineer for two copy changes in the following month. The template editor lands one sprint after its target sprint, and the engineer says so when asked rather than rounding it off.
Reflection. "Fast or careful" turned out to be the wrong question. The useful question was where care would buy the most, and the answer was the step that couldn't be undone. Under date pressure, the engineer now identifies the irreversible steps first and protects those.
Sample wording. A sample of how a candidate might say this, in their own words rather than these:
"We had a price change on a fixed date and had promised every affected subscriber an email two weeks before. I had three weeks. The proper version, a campaign feature marketing could edit themselves, wasn't going to fit. So I split the risks by whether we could undo them. A hard-coded template is annoying but fixable. An email to the wrong person, or a duplicate, isn't. I proposed to my lead and product manager that we cut the template editor and put the effort into the send path: a dry run, an idempotency key per subscriber, and staged sends. I wrote the trade-off into the ticket, including that marketing would need an engineer for copy changes for about a month. The dry run turned out to matter. Its count didn't match finance's, because paused subscribers were included, and the product manager confirmed they should get a different message later. I excluded them and the counts matched. The emails went out on the date with one send per subscriber. The template editor landed a sprint later than I'd planned, which I'd own as a miss. The thing I'd keep is that the decision wasn't fast versus careful. It was where to be careful."
Why it works. The candidate shows judgement rather than a slogan: they work out which part of the work deserves the care and spend it there. The risk is shared with the people who own the outcome, and the cost to marketing is stated in advance, not discovered. The dry-run mismatch is evidence that the care was well placed, and the candidate doesn't inflate it into a disaster narrowly avoided. The late clean-up is admitted. The work involves product, marketing and finance, which is the kind of cross-functional coordination an SDE-2 project often needs, while the decision itself stays inside the engineer's own feature.
How a weaker version differs. One weaker version insists on doing it properly, misses or nearly misses the date, and presents that as a stand for quality. Another ships the batch script, reports that it worked fine and offers no reasoning, which leaves the interviewer unable to tell judgement from luck. A third tells the strong story but drops the late template work, and that omission tends to surface the moment the interviewer asks whether the clean-up happened.