The Interviewer's Guide to Technical Interviews

Photo by Kelly Sikkema on Unsplash

Most guides on technical interview questions focus on helping candidates. This guide is meant for interviewers.

• Structured interviews give every candidate the same planned questions and scoring. A 2022 re-analysis of decades of hiring research ranked them first among the methods it compared. Google’s data also favors them over unstructured interviews.
• Before the interview, tie each question to one or two skills, write an example of a strong answer, and agree on one rating scale.
• Reward how candidates think: the questions they ask, the assumptions they state, and how they handle what they don’t know.
• Ask about the candidate’s own work and follow up live, which is hard for AI to fake. Drop any question that gets easy with practice.

What Makes a Good Technical Interview Question?

A good technical interview question tests one or two skills the job needs. It comes with a written example of a strong answer at the level you’re hiring for. Every interviewer scores the answers on the same scale. Without those three things, interviewers fall back on gut feeling, no matter how clever the question is.

Why Do Great Interview Performances Turn Into Bad Hires?

If you’ve hired engineers for a while, you’ve probably brought on someone brilliant for four hours and disappointing for four months. The usual response is to make the questions harder or to add more rounds.

But the problem is usually in the preparation. Nobody has agreed on the interview process, so each interviewer goes by instinct and their subjective impression. That’s where bias and charm do their best work.

What Should You Define Before Writing a Single Question?

Start with the four to six skills the role can’t do without. Then describe each skill at the level you’re hiring for. Mid-level and senior engineers can both answer a question well, in very different ways. Finally, tie each question to one or two skills and give every interviewer the same scale, so your debrief compares scores instead of adjectives.

It sounds like paperwork, but Google found that ready-made questions and scoring guides saved its interviewers 40 minutes per interview on average.

What Do Coding Questions Actually Reveal?

A coding exercise should show you how someone works. A clunky solution, worked out while thinking out loud, tells you more than a perfect one typed in silence. The perfect one might only mean they’ve seen the problem before.

For example, ask them to write a function that splits a restaurant bill evenly, then try it with $100 and three diners. Candidates who ask where the extra cent goes are showing you how they’ll handle unclear requirements. So are those who return 33.33333333333333 and move on.

How Do You Spot Real System Design Experience?

The best system design questions force trade-offs, where you gain one thing by giving up another. For example: “Our payment provider sometimes takes 30 seconds to respond. Redesign checkout so customers aren’t stuck watching a spinner and nobody gets charged twice.”

Anyone can say “put a queue in front of it.” Experienced candidates cover timeouts, retries, and messages that keep failing. They make sure a retried payment can’t charge anyone twice. And they name the costs, like order statuses that briefly lag behind reality.

And don’t grade against the design you imagined as the ideal. A different design still earns full marks if the candidate can defend it.

What Should the Behavioral Round Actually Test?

In practice, senior hires rarely fail because they can’t code. They struggle with disagreements, vague requirements, poor communication, and problems nobody owns. Yet many teams ignore the behavioral round.

Mix two kinds of questions.

Past-experience questions dig into real events: “Tell me about an outage you were close to. What did you do in the first hour, and what changed afterward?”

Situational questions test judgment where experience runs thin: “Your product manager wants a feature live by Friday, and you’ve just noticed it writes customer data to the logs. What do you do?”

The first kind shows you a track record. The second shows how someone reasons without one, which matters for career changers and anyone stepping up a level.

Score answers with STAR (situation, task, action, and result), but don’t expect candidates to follow that order. Rehearsed candidates often describe the situation in detail, then rush past their own part and say “we” throughout. So ask, “Which part of that was your call?” If the answer stays vague, their role probably was too.

How Should AI Change Your Interviews?

AI raises two separate issues for interviewers. First, some remote candidates will have an AI tool open in another window, and generic questions make it easy to lean on.

Tighter monitoring tends to turn into an arms race, but better questions don’t. A chatbot can write a believable story about an incident. Keeping that story straight through quick, specific follow-ups is much harder.

Second, your engineers probably use AI assistants every day, so test that skill too. Give candidates a polished, AI-written solution with a planted bug and ask for a code review. The best ones read it with suspicion and look for the bugs. They treat AI output like a confident stranger’s Reddit post: useful, but never final.

Which Interview Questions Should You Retire?

Trivia questions, like asking for PostgreSQL’s default port, test facts any engineer can look up in seconds. Brainteasers, like guessing how many golf balls fit in an airplane, mostly reward puzzle practice. Google concluded in 2013 that they were a “waste of time” that didn’t predict anything.

Questions copied from the top search results have another problem: many motivated candidates have already practiced them. If knowing a question in advance would hand someone the answer, it mostly measures preparation.

How Do You Know Your Process Is Working?

Some of the best evidence shows up when a candidate doesn’t know the answer. A reply like “I haven’t tuned that myself, but I’d reproduce the problem and read the logs before touching anything” beats a lucky guess. Your scorecard should say so, because charisma is easy to mistake for skill.

Then check whether your scores predict anything. Google did: it compared scores from tens of thousands of past interviews with how those hires later performed. Laszlo Bock, who ran HR at Google then, said they found “zero relationship”.

Interviewing never gets easy, but you can measure it. Six to twelve months after someone joins, compare their interview scores with how they’re actually doing.

FAQ

What is a structured technical interview?

It’s an interview where every candidate for a role gets the same prepared questions, each tied to a defined skill. You write down strong answers ahead of time, and every interviewer scores on the same scale.

Are LeetCode-style questions useless?

Not entirely, but they reward practice, and candidates drill exactly these puzzles on sites like LeetCode. Realistic tasks, like improving the performance of slow code, show reasoning more clearly.

Should candidates be penalized for saying “I don’t know”?

No, admitting a gap is a strength, especially when it comes with a plan to close it. Engineering work is full of unknowns. How someone handles one in the interview previews how they’ll handle the next one at work. A confident guess that happens to be right tells you much less.