Are Coding Interviews Easier?

A lot of software interviews were built around a simple bet: give one person one coding problem for 45 minutes, and their answer will reveal how they think. That bet gets shakier when a chat window can produce a clean solution to a standard array or graph question in seconds. (leetcode.com, openai.com) That is why some candidates now say interviews feel easier. The familiar “LeetCode medium” style problem is still hard for a human under a timer, but it is often no longer hard for an artificial intelligence coding assistant that can draft code, explain tradeoffs, and repair bugs on demand. (leetcode.com, openai.com) LeetCode itself is a giant interview-prep machine with more than 4,150 practice questions and company-focused study plans. If a company keeps asking the same recognizable patterns, candidates can now combine memorized patterns with artificial intelligence help and get much closer to a polished answer than the old format assumed. (leetcode.com, leetcode.com) This is not just a candidate rumor anymore. Anthropic published a January 21, 2026 post saying its own performance-engineering take-home had to be redesigned three times because newer Claude models kept beating or matching the signal the company wanted from the test. (anthropic.com) Anthropic said more than 1,000 candidates completed that take-home, and Claude Opus 4.5 eventually matched even the strongest applicants under the same time limit. The company’s conclusion was blunt: the team “no longer had a way to distinguish” top human output from its best model on that format. (anthropic.com) That does not mean humans suddenly became better engineers in 2026. It means a narrow test can stop measuring the person and start measuring the tool sitting next to the person, the same way a calculator changes what a timed arithmetic quiz tells you. (anthropic.com) At the same time, companies cannot just ban artificial intelligence and pretend it is 2019. GitHub said in a 2,000-person international survey that more than 97% of respondents had used artificial intelligence coding tools at work at some point, so the tools are already part of normal developer workflow. (github.blog) GitHub’s research now describes advanced developers less as pure “code producers” and more as people who delegate, verify, and orchestrate. In other words, modern engineering skill increasingly includes knowing when an artificial intelligence suggestion is useful, wrong, insecure, or incomplete. (github.blog) That shift breaks the old interview scoring rubric. If two candidates both reach the correct answer, the harder question is no longer “who wrote every line alone,” but “who can inspect generated code, test edge cases, explain failure modes, and adapt it to a messy codebase without making a hidden production mistake.” (github.blog, openai.com) Some companies are already moving in that direction by making tasks less like puzzle hunts and more like actual work. Anthropic said longer-horizon evaluations, unusual problem setups, and tasks tied to real debugging or optimization gave it more signal than standard short-form exercises that a model could ace on first pass. (anthropic.com) The likely end state is not “no more coding interviews.” It is interviews that look more like code review, bug triage, system changes, and tool-assisted implementation, where the candidate is judged on judgment instead of just speed on a memorized pattern. (github.blog, github.blog) So when people say coding interviews feel easier, they are probably noticing a real mismatch. The test was designed for an era when writing the answer was the scarce skill, and hiring is now drifting into an era where checking, steering, and stress-testing the answer may be the scarcer one. (anthropic.com, github.blog)