LeetCode Is Dying, But What Will Come Next?

I had the advantage of interviewing for tech jobs in the era of ChatGPT. And for anyone who’s tried solving LeetCode using it, you know it aces any problem set in seconds. I got offers at the country’s biggest firms when I didn’t deserve to. And that didn’t sit right with me.

If tech interviews are a reflection of job ability and AI can understand and ace these interviews, wouldn’t AI be the best contender for the job?

So I set out to build an interview that tested the skills AI had made more valuable and could not easily be completed with AI.

I repurposed an AI benchmark into an interview format that does both. I wrote about it in a post titled “We Killed LeetCode.” To my surprise, it went viral.

One of the people who saw the post was NeetCode (Navdeep Singh), the number one coding interview prep brand. He reached out and asked for a demo. A few days later, he followed up, asking to invest. The person who helped millions of engineers master LeetCode interviews was now backing a replacement for it. 

More in Job Interviews in the AI AgeWhen AI Writes the Code, What Skills Are Employers Hiring For?

The Interview We Built

Instead of abstract puzzles, the interview is structured as a 30 to 90 minute asynchronous job simulation that mirrors day-one work. Every assessment is custom-built for the specific company and role.

Candidates are embedded in a real engineering environment: a broken codebase, a messy pull request or a production scenario with ambiguous, shifting constraints. They might receive a repository with a subtle bug that isn’t documented anywhere, a failing test suite and a vague description of expected behavior. The candidate’s job is to find the issue, understand why it’s happening, fix it and explain why that fix is the right one.

AI tools are allowed. Expected, even. The goal is to see whether they can direct it effectively, recognize when it’s wrong and make sound decisions about what to do next.

What We’re Scoring and Why

Instead of a binary “right or wrong,” we break evaluation down into structured signals.

Debugging Depth

How does the candidate navigate an unfamiliar codebase? How do they form and test hypotheses? Do they find the root cause or just patch until the tests pass?

Code Quality

Is the fix legible? Does it handle edge cases? Does the candidate leave the repo in better shape than they found it?

Judgment

How do they reason through tradeoffs? Can they explain the conflicting requirement and why they made a particular decision?

AI Proficiency

This is increasingly the highest-signal dimension. We look at how candidates prompt, whether they verify outputs and whether they push back when the model is wrong. Engineers who use AI well are measurably more productive than those who don’t, and this format surfaces that gap directly.

Communication

Can they explain what they found in terms a teammate can act on? We ask follow-up questions mid-session because engineering work requires constant translation between what you’re doing and what others need to understand.

Each dimension is scored independently, producing a multi-dimensional profile rather than a binary pass/fail. We weigh dimensions against their specific role, a senior engineer inheriting a legacy codebase gets scored heavily on debugging depth and design choices; a greenfield hire gets more weight on judgment and communication.

LeetCode can be gamed because the solution set is finite and the problems are well-known. Even without AI, a motivated candidate can practice the same 150 questions until they’ve memorized the patterns. Work simulations, just like real work, don’t have an answer key. You can’t memorize your way through an unfamiliar codebase.

We’ve started benchmarking scores against actual on-the-job performance for customers who are far enough into their hiring cycles. Unsurprisingly, the correlation between the interview and job performance is better than what orgs are used to.

Ace Your Next InterviewHow to Prepare for a Job Interview (With AI Prompts and Expert Tips)

What Is the Future of the Technical Interview?

I’ve had the privilege of working with the nation’s biggest companies to revamp their technical interview process in the last month. A few things are very apparent as everyone gears up for this shift. 

The engineering leaders I’ve talked to all share a sentiment: AI is fundamentally changing the way engineers build. The interview should reflect that. Managers started noticing that 40 to 50 percent of candidates were scoring perfectly on current technical screens.

In my first conversation with Jason Shiverick, Director of AI Platforms at Rivian, he shared that “LeetCode is useless, and the online assessments out there are not relevant signals. What we care about is how people actually work, and especially how they leverage AI.”

That perspective is spreading beyond engineering teams. Amelia Generalis, Chief People Officer at ID.me, points to a broader change in hiring philosophy: The future involves evaluating candidates not by restricting tools, but by understanding how effectively they use them.

And the fact that NeetCode, the person who taught a generation of engineers to pass the old test, is now an investor in what we’re building probably tells you something about which direction this industry is heading.