How have tech interviews changed in the AI era?

One shift explains everything else: employers stopped trusting artefacts and started testing judgement live. Take-home assignments lost weight (anyone can produce one with AI), memorised-algorithm rounds declined (AI made recall cheap), and in their place came defend-your-code rounds (walk through work claimed as yours and justify its decisions), live modify-and-extend exercises (change this codebase while we watch how you think), and explicit AI-collaboration probes (solve this with your AI tools while we watch HOW you use them - what you accept, what you question, what you catch). The through-line: 2026 interviews test whether you can evaluate and defend software, not merely produce it - which is exactly what separates fundamentals-first candidates from vibe coders.

What rounds does a typical Malaysian tech interview have?

The common 2026 shape at startups and SMEs (the bulk of the market): (1) a screening call - part culture, part communication check, often with a hiring manager rather than HR; (2) a technical round - increasingly live building or code-walkthrough rather than whiteboard algorithms, sometimes with AI tools explicitly allowed; (3) a portfolio deep-dive - the round your projects were built for, probing decisions, trade-offs, and what broke; (4) sometimes a short practical or pairing session; (5) an offer conversation. Larger companies and banks add layers (aptitude tests, panel rounds, HR bands). Remote employers add a written-communication component and weight async clarity heavily. Very few Malaysian employers still run pure LeetCode-style gauntlets - though some larger and international ones do, so check per company.

Should I use AI during a technical interview?

Follow the interviewer’s frame, and know both games. If AI tools are explicitly allowed or provided - increasingly common - use them the way a professional does: direct them precisely, review output critically aloud ("this generated approach has an edge-case issue here..."), and let the interviewer see judgement, because that performance IS the test. Using AI clumsily - accepting everything, prompting vaguely - fails you faster than not using it. If tools are disallowed, that is the other test: whether capability survives without assistance, which is precisely why training with deliberate AI-off practice matters. Never use AI covertly when it is disallowed; getting caught (and interviewers know the tells) ends more than the interview.

How do I prepare for a tech interview with no work experience?

Your portfolio is your work experience - prepare it like one. For each project, rehearse the five questions every deep-dive asks: why this problem, why these technical choices, what was hardest, what broke and how did you find it, what would you change now? If you cannot answer those fluently about your own work, that gap - not your CV - is what fails interviews, and it is also the tell interviewers use to detect AI-generated portfolios. Then drill the live layer: explain code aloud while modifying it (practise with a study partner or mentor - the skill is speaking-while-thinking, and it trains fast), and do 2-3 mock interviews before any real one. Treat the first real interviews as calibration: most people convert somewhere between interview three and ten.

What questions should I ask the interviewer?

Ask questions that are simultaneously genuine diligence and quiet signalling. Strong ones for 2026: "How does the team use AI tools day to day?" (signals you work the modern way; their answer also tells you if the company is behind), "What does the first 90 days look like for this role?" (signals seriousness; the answer reveals whether onboarding exists), "How does code get reviewed here?" (signals you value feedback; teams that can’t answer have quality problems), and "What separates the juniors who grow fast here from those who don’t?" (the answer is a free roadmap). Avoid leading with leave policies and remote-day counts - legitimate topics, but they land better after an offer exists.

All posts

Careers · Getting hired · Interviews

The tech interview changed. Most advice hasn’t noticed.

Most interview prep online still trains you for 2019: grind algorithm puzzles, polish a take-home, memorise trivia. Meanwhile actual employers - burned by AI-generated take-homes and vibe-coded portfolios - rebuilt the interview around one question: can you defend and evaluate software, live? Here is what the rounds look like now, and how to prepare for each.

Deric Yee Updated 25 August 2026 10 min read

A confident candidate ready for a technical interview

Why the interview had to change

Understand the employer’s problem and every 2026 round makes sense. For two decades, hiring signals were artefacts: a take-home assignment, a portfolio, a certificate - things produced awayfrom the interviewer’s eyes. AI broke every one of them at once: a take-home can now be generated in an evening, a portfolio can be vibe-coded without understanding, and course certificates were already weak. Employers responded exactly as you’d predict - they moved the test into the room. If the artefact can be faked, interrogate the human about the artefact; if production is cheap, test evaluation. Hence the new canon: walk-me-through-your-code rounds, modify-this-live exercises, and the newest fixture, the AI-collaboration probe, where you solve a problem withAI tools while the interviewer watches how you direct, question, and correct them. The 2026 interview is a judgement exam wearing a coding exam’s clothes - which is why candidates trained on fundamentals with AI in the loop walk through it, and shortcut-trained candidates hit a wall precisely here.

The rounds, stage by stage

The screening callis more load-bearing than it looks: at Malaysian startups and SMEs it is often the hiring manager themselves, and the real test is communication - can you explain what you’ve built, in plain language, with genuine enthusiasm, in decent English? Prepare a 90-second version of your story and a two-minute version of your best project; rambling kills more screenings than skill gaps do. The technical round now splits into flavours: some teams pair with you on a small live task (they are watching how you think aloud and handle being stuck - narrate honestly, ask clarifying questions, treat the interviewer as a colleague); some hand you an unfamiliar codebase and ask you to find a bug or add a feature (they are testing code reading, the most-used and least-practised professional skill); a shrinking minority still run algorithm puzzles - check company reviews beforehand and drill accordingly rather than by default.

The portfolio deep-dive is the round the whole no-experience path is built for - and the round where preparation pays most predictably, because you know the material in advance: it is your own work. Expect the five canonical probes (why this problem, why these choices, what was hardest, what broke, what would you change) plus the 2026 addition: “how much of this did AI write, and how do you know it’s correct?”The winning answer is neither “none” (implausible, and worse, old-fashioned) nor a shrug - it is a precise account of your collaboration: what you directed, what you rejected, the bug the AI introduced that you caught and how. Employers are not screening AI use out; they are screening for people who can be accountable for AI-assisted work, because that is the actual job now. Build projects worth interrogating in the first place - the criteria are in the portfolio guide - and this round becomes your strongest.

The question banks, by round

This guide covers the shape of the interview; the specific questions each round recycles have their own dedicated banks, each with the answers interviewers actually want and the live follow-ups behind them: the full junior round (openers, behavioural probes, the AI-workflow question), JavaScript, React, SQL and databases, and - for the premium lane - AI developer roles. Work this guide for the strategy, then the banks matching your target roles for the reps.

The preparation plan that matches the test

Work backwards from what is measured. Drill explaining aloud, not just solving: the core interview skill is speaking-while-thinking, and it trains fast - rubber-duck your own projects daily for a week, then do two or three mock interviews with a mentor or study partner before any real one (inside our programme this is formalised as defence interviews, where you justify your project decisions to an instructor precisely because the real market tests it). Practise both AI modes: sessions where you build with your AI tools while narrating your direction and review of them, and AI-off sessions that keep raw capability honest - the interview will test one or both, and the preparation is the same discipline that makes you good at the job. Read code you didn’t write: twenty minutes daily in unfamiliar open-source code trains the modify-this-codebase round better than any puzzle grinding. And calibrate with real at-bats: treat your first three interviews as reconnaissance - log every question, patch the gaps each one exposes, and expect conversion somewhere between interview three and ten, which is the normal curve, not a verdict on you.

Last, the meta-advice that outranks all tactics: the interview is downstream of the preparation months. A candidate with real fundamentals, an honest AI workflow, and two defensible projects needs only modest interview polish; a candidate without them cannot interview-prep their way through a judgement exam. If you are earlier in the journey, the sequence that produces interview-ready candidates is mapped in the no-experience playbook, and its first rung - six real projects with a live instructor session - is free at the free trial. The graduates who walk through 2026 interviews - several now in roles at the RM 6,000–9,000 band documented in the hiring report - were made in the months, not the week, before the interview.

FAQ

How have tech interviews changed in the AI era?
One shift explains everything else: employers stopped trusting artefacts and started testing judgement live. Take-home assignments lost weight (anyone can produce one with AI), memorised-algorithm rounds declined (AI made recall cheap), and in their place came defend-your-code rounds (walk through work claimed as yours and justify its decisions), live modify-and-extend exercises (change this codebase while we watch how you think), and explicit AI-collaboration probes (solve this with your AI tools while we watch HOW you use them - what you accept, what you question, what you catch). The through-line: 2026 interviews test whether you can evaluate and defend software, not merely produce it - which is exactly what separates fundamentals-first candidates from vibe coders.
What rounds does a typical Malaysian tech interview have?
The common 2026 shape at startups and SMEs (the bulk of the market): (1) a screening call - part culture, part communication check, often with a hiring manager rather than HR; (2) a technical round - increasingly live building or code-walkthrough rather than whiteboard algorithms, sometimes with AI tools explicitly allowed; (3) a portfolio deep-dive - the round your projects were built for, probing decisions, trade-offs, and what broke; (4) sometimes a short practical or pairing session; (5) an offer conversation. Larger companies and banks add layers (aptitude tests, panel rounds, HR bands). Remote employers add a written-communication component and weight async clarity heavily. Very few Malaysian employers still run pure LeetCode-style gauntlets - though some larger and international ones do, so check per company.
Should I use AI during a technical interview?
Follow the interviewer’s frame, and know both games. If AI tools are explicitly allowed or provided - increasingly common - use them the way a professional does: direct them precisely, review output critically aloud ("this generated approach has an edge-case issue here..."), and let the interviewer see judgement, because that performance IS the test. Using AI clumsily - accepting everything, prompting vaguely - fails you faster than not using it. If tools are disallowed, that is the other test: whether capability survives without assistance, which is precisely why training with deliberate AI-off practice matters. Never use AI covertly when it is disallowed; getting caught (and interviewers know the tells) ends more than the interview.
How do I prepare for a tech interview with no work experience?
Your portfolio is your work experience - prepare it like one. For each project, rehearse the five questions every deep-dive asks: why this problem, why these technical choices, what was hardest, what broke and how did you find it, what would you change now? If you cannot answer those fluently about your own work, that gap - not your CV - is what fails interviews, and it is also the tell interviewers use to detect AI-generated portfolios. Then drill the live layer: explain code aloud while modifying it (practise with a study partner or mentor - the skill is speaking-while-thinking, and it trains fast), and do 2-3 mock interviews before any real one. Treat the first real interviews as calibration: most people convert somewhere between interview three and ten.
What questions should I ask the interviewer?
Ask questions that are simultaneously genuine diligence and quiet signalling. Strong ones for 2026: "How does the team use AI tools day to day?" (signals you work the modern way; their answer also tells you if the company is behind), "What does the first 90 days look like for this role?" (signals seriousness; the answer reveals whether onboarding exists), "How does code get reviewed here?" (signals you value feedback; teams that can’t answer have quality problems), and "What separates the juniors who grow fast here from those who don’t?" (the answer is a free roadmap). Avoid leading with leave policies and remote-day counts - legitimate topics, but they land better after an offer exists.

Back to all posts The skills behind the rounds →

Interviews test judgement. Build some.
Made in the months, not the week, before.

Defence interviews, AI-on and AI-off practice, and projects worth interrogating are built into the programme because the market tests them. The free first rung: six real projects and a live instructor session, one signup, no card.

Start the free trial The no-experience playbook