In a nutshell
- LeetCode isn't dead. The LeetCode-style puzzle, asked over video with no follow-ups, is dead as a hiring signal. The problem and its solution are retrievable, and tools exist specifically to retrieve them invisibly.
- It was never as strong a signal as its popularity suggested. Structured interviews and work samples have better research behind them.
- Three formats replaced it in 2026. Companies like Google, Cisco and McKinsey are adding in-person rounds. Canva, Shopify and a Meta pilot are running AI-enabled interviews. And more teams use a work sample followed by a live defense.
- For most startups, in-person isn't practical for every hire. The cheaper fix is to convert algorithm questions into practical, laddered problems, add one AI-allowed stage, and defend any offline work live.
- Algorithmic thinking still matters. It just has to be tested through follow-ups that a pre-generated answer can't survive.
Is LeetCode actually dead?
The skill isn't. The format is.
Knowing when a hash map beats a nested loop, or why a query is slow, still separates good engineers from weak ones. What died is a specific interview: a well-known puzzle, read out on a video call, scored mainly on whether the candidate reaches the optimal solution.
Canva's engineering team tested this and wrote that AI "can trivially solve traditional coding interview questions". If a tool solves the question, a solved question tells you about the tool.
Why did the remote LeetCode round break?
Three reasons, and they compound.
The answers are retrievable. Classic puzzles have known, optimal solutions. That was always true. What changed is how fast a solution can be fetched during the interview.
Tools exist to fetch them invisibly. Interview Coder, the tool that later became Cluely, was built by a student who recorded himself using it in an Amazon technical interview, as CNBC reported. Its pitch was an assistant that stays hidden from screen share. The category is now mainstream enough that a detection industry has grown up around it. I cover how these tools work, and why detection struggles, in Interview Coder and Cluely detection.
Technical rounds are the most exposed. Fabric, which sells AI interviewing with cheating detection, flagged 48% of technical-role interviews on its platform against 12% for sales. That's vendor data from a mostly India-based sample, and "flagged" isn't "proven". But it fits the logic: the more retrievable the answer, the more hidden help pays off.
So the remote puzzle round now measures some mix of skill, interview prep and tool access, and you can't separate them.
Was LeetCode ever a great signal?
Less than its reputation suggests.
The largest recent re-analysis of selection research, Sackett and colleagues in 2022, ranks structured interviews highest at a validity of about .42. Work samples come in around .33. Unstructured interviews sit at about .19.
A puzzle round can be structured, if every candidate gets the same problem and a written rubric. But it usually isn't a work sample. Reversing a linked list under time pressure isn't what most engineers do at work, and performance on classic puzzles tracks hours of practice as much as engineering ability.
So the honest framing: the remote LeetCode round lost the one thing it had going for it, a level playing field. Without that, there's little reason to keep it.
What replaced the LeetCode round in 2026?
Three formats, often combined.
1. In-person rounds
The most visible change is a partial return to in-person interviews.
Google's CEO Sundar Pichai said on the Lex Fridman podcast in June 2025 that Google will "introduce at least one round of in-person interviews… just to make sure the fundamentals are there." In the same answer he said that if candidates can use AI tools to generate better code, "I think that's an asset."
The Wall Street Journal reported in August 2025 that Cisco and McKinsey are adding face-to-face meetings, with McKinsey encouraging at least one in-person meeting before an offer. One Dallas recruiting firm estimated the share of its clients requesting in-person interviews rose from 5% in 2024 to 30% in 2025. That's one firm's estimate, not a market statistic.
And a Gartner survey, as reported by Computerworld, found 72.4% of recruiting leaders conduct in-person interviews to combat fraud.
Candidates don't seem to mind. Gartner found 62% of candidates say they're more likely to apply if the employer requires in-person interviews.
The cost is real, though. At a 2025 Google town hall, as CNBC reported, the company's recruiting VP said virtual interviews are about two weeks faster. For a startup hiring across borders, flying every finalist in isn't realistic.
2. AI-enabled interviews
The opposite response: let candidates use AI, and change what you score.
- Canva replaced its computer-science fundamentals screen with "AI-Assisted Coding" and expects candidates to use tools like Copilot, Cursor and Claude.
- Shopify's Head of Engineering, Farhan Thawar, said on the Pragmatic Engineer podcast that candidates can "use whatever they want", and that those who don't use a copilot "usually get creamed by someone who does". He also wants trivial bugs fixed by hand rather than prompted.
- Meta is piloting a coding interview where candidates have an AI assistant. An internal post reported by Wired said it "also makes LLM-based cheating less effective."
The logic is neat. You can't cheat on a rule that doesn't exist. And the interview gets closer to the actual job: reading what the AI wrote, catching what's wrong, and knowing when to stop prompting.
I go through when this makes sense, and when it doesn't, in should you allow AI in coding interviews.
3. Work sample plus live defense
The third replacement is less headline-friendly and, for small teams, often the most practical.
The candidate does a scoped, paid piece of realistic work, usually with AI allowed. Then, in a live session, they explain it and change it while you change the requirements. The work sample shows how they work. The defense shows the work is theirs. Someone else can do a take-home. They can't sit the defense for the candidate.
The details matter: a realistic starter repo, deliberate flaws for the candidate to catch, a time box, and scoring on judgment rather than output. I cover the design in take-home assignments and AI.
How do these formats compare?
| Format | What it measures well | Resistance to hidden help | Cost to you | Best for |
|---|---|---|---|---|
| Remote LeetCode on video | Puzzle prep, some reasoning | Low | Low | Very little now |
| In-person round | Fundamentals, identity, communication | High | High: travel, scheduling, slower | Final rounds, high-access roles |
| AI-enabled interview | Judgment with tools, verification, speed | High, because AI is allowed | Medium: needs a new rubric | Teams where engineers use AI daily |
| Work sample + defense | Realistic work and ownership | High, if the defense is live | Medium: build once, pay per candidate | Startups, contract roles, async hiring |
| Laddered practical problem | Reasoning, adaptation, ownership | Medium to high | Low | Replacing the puzzle round directly |
Which should a startup pick?
You probably can't fly every candidate in. You can do the rest.
My default for a remote startup hiring mid-level engineers:
- A recruiter or founder screen built on ownership questions about their own work.
- A laddered technical round in place of the puzzle: a practical problem in their stack, shown on screen, with constraints that change live.
- An AI-allowed work stage, either a paired build or a paid take-home.
- A live defense or code review, independent of AI.
- In person for the final round when the role has access to production, money or customer data, and when it's feasible.
Label each stage as AI-allowed or independent, and put it in writing. The free AI interview policy template is a starting point.
What should I do with my existing LeetCode questions?
Don't throw them away. Convert them.
A puzzle becomes useful again when it's the first rung of a ladder rather than the whole interview:
"Start with the simplest version that works, even if it's slow. We'll improve it together."
Then change something on screen:
"Now the input is 50 million records and doesn't fit in memory. What changes?"
Then ask about the decision:
"You switched to a heap there. Why not just sort? And how would you know it's wrong in production?"
Then connect it to their life:
"When did you last hit a problem like this at work? What did you ship?"
The base answer is the part a tool can help with. The other three rungs are where the score should live. The full method, with worked examples, is in technical interview follow-up questions.
Isn't an in-person round just LeetCode on a whiteboard?
It can be, and that would waste the trip.
Flying a candidate in to solve the same retrievable puzzle removes the hidden help, but it keeps the other weakness: you're still mostly measuring puzzle prep. If you're paying for an in-person round, use it for things that are hard to do any other way. Have them debug a real failing test with you. Sketch a system together and change the requirements halfway. Review a flawed pull request out loud. Ask about their past work and keep asking until the details run out.
The in-person round's unique value is that you know who's in the room and that nothing is feeding them answers. Spend that on the questions that matter most, not on the ones a tool would have answered anyway.
Won't all this make my loop longer?
Not if you replace rather than add.
The common mistake is to bolt a new stage onto an old loop: keep the online assessment, keep the puzzle round, and add a work sample and an in-person day on top. Candidates drop out, and strong engineers with other offers go first.
Instead, swap one for one. The laddered practical problem takes the puzzle round's slot. The AI-allowed work stage replaces the unsupervised online test, which was effectively AI-allowed already. The defense replaces the second algorithm round many loops still have. You end up with roughly the same number of stages, each measuring something you can name.
Should candidates still practise LeetCode?
Some companies still ask puzzles, especially in person, so prep isn't wasted. But if you're hiring, don't reward prep alone. Reward the candidate who can explain why their solution works, adapt it when the problem changes, and tell you about a time they needed it.
What should I do next?
Take the free cheat-risk audit to see which of your stages still depend on retrievable answers. If you'd like the replacement already built, with reference loops by role, laddered questions by stack, and AI-allowed work samples with defense scripts, that's Unscripted, the paid kit. Either way, the first move is the same: retire the remote puzzle round, or turn it into the first rung of something harder.