Is the era of solving LeetCode ending? Big‑tech interviews are shifting.

Is the era of grinding LeetCode coming to an end? Big‑tech interviews are shifting

Recently, a software engineering interview reform at Google has sparked heated discussion online.

According to the new pilot program, candidatescan use Gemini during the interview, and the assessment content is also changing.

In addition to writing an algorithm problem from scratch, Google also expects candidates to read others’ code, pinpoint issues, and evaluate whether AI‑suggested solutions are reliable.

A netizen excitedly said: “AI finally killed the LeetCode interview.”

Beyond writing code, Google also tests how you read code

For many programmers preparing to join big tech companies, LeetCode is a name they know all too well.

Solving algorithm problems, studying time complexity, and writing correct code within limited time have become almost mandatory for technical interviews.

But Google is experimenting with a different assessment approach.

According to Business Insider’s May 7 report, in Google’s pilot new stage—Code Comprehension—candidates will read, debug, and optimize an existing piece of code.

AI can also take part in this process. A Google spokesperson confirmed that the tool provided to candidates in the pilot is Gemini.

Since you can ask AI directly, how should interviewers assess a candidate’s level?

Google’s answer is “AI fluency.”

Interviewers will focus on how candidates design prompts, verify AI output, and debug code. Candidates need to be able to judge whether the AI’s suggestions are reasonable and take responsibility for the final result.

This means that even if Gemini quickly finds a bug, the candidate still needs to verify whether the fix is valid and whether the change might introduce other issues.

Of course, this is currently only a limited-scope reform.

The related arrangements target the second half of this year and mainly involve junior and mid‑level software engineer positions on some U.S. teams. Organizations such as Google Cloud and the Platforms & Devices division participated in the pilot, and whether it will be rolled out more broadly depends on the actual results.

Google recruiting vice president Brian Ong said the interview process adjustments are intended to more accurately reflect how engineers actually work in the AI era.

In addition to the code‑understanding segment, Google also plans to incorporate technical design discussions of candidates’ past projects into some interviews, allowing junior candidates to tackle more open‑ended engineering problems.

What can be confirmed now is that the scope of assessment is expanding. Beyond the ability to write code from scratch, companies are beginning to evaluate whether candidates can work with existing codebases, even those generated by AI.

Karpathy: Hand‑coding ability is starting to degrade,

but reading code is a different story

Why is Google adjusting its interview process? Looking at what’s happening within Google’s own engineering teams makes it easy to understand.

In April this year, Google CEO Sundar Pichai disclosed that 75% of new code is now generated by AI and then reviewed by engineers. Last fall, that figure was 50%.

Link: https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/cloud-next-2026-sundar-pichai/

In just about half a year, AI’s involvement in software development has noticeably increased.

For individual developers, such changes are happening even faster.

In January this year, Andrej Karpathy shared his experience using Claude Code on X.

He found that his programming habits changed almost dramatically within just a few weeks.

November 2025, about 80% of his programming work was still done manually by himself, or with code autocompletion; by December, roughly 80% of the work had been handed over to an Agent, and he only handled the remaining edits and refinements.

Karpathy remarked that this was one of the biggest changes in his work style in his nearly 20‑year programming career.

Today, much of his programming work has become describing requirements to AI in natural language.

This shift even brings an unexpected side effect: Karpathy discovered thathis ability to write code manually is gradually deteriorating.

However, he distinguishes two abilities:

“Generation (writing code) and discrimination (reading code) are different abilities in the brain.”

Writing code often requires programmers to remember many specific syntax and implementation details, whereas reviewing code relies more on understanding existing logic and judging whether the implementation is reasonable. So even if someone becomes less adept at writing code from scratch, they can still read code and spot issues.

But AI-generated code is becoming increasingly difficult to audit.

Karpathy mentioned in the post that the mistakes programming models make today are no longer mainly simple syntax issues.

Instead, they may make wrong assumptions when understanding the requirements and then continue the entire implementation based on that assumption. Sometimes the model writes thousands of lines of complex code, and after an engineer points out a more concise solution, it can quickly shrink the code to around a hundred lines.

These errors look more like the oversights made by inexperienced developers working in a hurry.

Therefore, even though Karpathy can hardly imagine returning to a development approach that relies primarily on hand‑written code, he still believes we need to closely monitor what AI produces.

This is the new thinking behind Google interviews.

If AI can complete a large amount of coding work in minutes, engineers’ tasks will increasingly shift toward understanding code, validating implementations, and intervening when AI makes mistakes.

Code generation is becoming easier, making code review skills even more important.

Google isn’t the first; some even deliberately make AI write incorrectly

In fact, Google is no longer the earliest company to try this interview approach.

Last June, design platform Canva announced on its official engineering blog that backend, machine‑learning, and frontend engineering candidates will be required to use AI coding tools during technical interviews.

Link: https://www.canva.dev/blog/engineering/yes-you-can-use-ai-in-our-interviews/

For front‑end and back‑end positions, Canva even replaced the basic interview stage with AI‑assisted coding; previously this stage focused on algorithms and data structures.

The reason is straightforward:The old questions are no longer that effective today.

Canva once used AI to test its traditional interview questions and found that the model often only needs a few seconds to generate correct answers with explanations, even without further prompting.

So, Canva decided to make the questions a bit more complex.

For example, where candidates might previously have been asked to implement “Conway’s Game of Life,” they might now be asked to design an aircraft takeoff and landing control system for a busy airport.

These open-ended problems require candidates to independently gather requirements, devise solutions, and coordinate relationships between different functionalities. AI can handle the coding tasks, but it is difficult to manage all design and engineering details with just a simple instruction.

Canva also shared observations from the pilot.

Top-performing candidates tend to proactively clarify requirements, break complex tasks into more defined parts, and then selectively leverage AI. They review the generated results and adjust their approach when issues are identified.

Some candidates with programming skills but lacking AI experience may actually struggle to guide the AI and assess whether its suggestions are appropriate.

Meta is also beginning to adopt a similar approach.

In July 2025, the media reported that Meta was developing coding interviews that allow the use of AI. Shortly after, Meta officially introduced this pilot and encouraged participating candidates to use approved AI tools.

Meta specifically emphasizes that using AI in such interviews is no longer considered cheating, because it is a tool engineers actually use in their work.

In October of this year, the recruiting technology platform Codility released a new interview feature that allows interviewers to embed hidden instructions into the AI coding tools used by candidates.

Link: https://www.codility.com/product-tours/interview/

For example, instructing the AI to deliberately write buggy code or use suboptimal programming approaches.

Candidates will not see these instructions. Interviewers can observe how candidates handle them—whether they carefully review the generated results or simply copy‑paste and accept everything at face value.

The platform also supports controlling AI usage permissions across different assessment tasks, allowing companies to evaluate both candidates’ collaboration with AI and their ability to solve problems independently.

In the past, to prevent candidates from obtaining answers via AI, interviewers might have had to look for traces of AI usage. Now, AI can directly enter the interview room and even become a tool that interviewers use to assess candidates.

Previously, a coding problem might start with “Please implement this algorithm.”

In the future, interviewers might give you a piece of AI‑generated code and ask, “This code runs already; what issues do you see with it?”

But the AI tools at hand may not help you answer this question correctly.

As some netizens joke: “Requiring candidates to use Gemini to complete the task is itself a great interview question.”

Reference link:

https://www.businessinsider.com/google-job-interview-software-engineers-ai-assistant-coding-2026-5

https://x.com/karpathy/status/2015883857489522876

https://www.canva.dev/blog/engineering/yes-you-can-use-ai-in-our-interviews/

https://www.linkedin.com/posts/meta_metacareers-interviewing-softwareengineering-activity-7386046074240737280-wnwA

https://www.codility.com/lab/product-updates/coding-agents-task-by-task-oct-2026-update/

© THE END 

Please contact this public account for permission to repost

Submit articles or request coverage: liyazhou@jiqizhixin.com