# Anthropic's Felix Rieseberg drops coding puzzles after Claude outperforms candidates > Anthropic's Felix Rieseberg removed coding puzzles from interviews after Claude outperformed most candidates. - Status: reported (One outlet, not independently verified yet.) - Topic: AI in Practice - Published: 2026-11-03T13:09:04+00:00 - Updated: 2026-11-03T13:09:04+00:00 - Web page: https://ai-news.ikangai.com/story/anthropic-s-felix-rieseberg-drops-coding-puzzles-after-claude-outperforms-candid/ - Seen in: Forward Future (Matthew Berman) Felix Rieseberg, an engineering manager at Anthropic, has stopped giving coding interviews to candidates for his team. He says Claude now beats most of them at the puzzles. He posted on X on October 27, 2026, that the coding interview "is dead, at least for my team at Anthropic," and said he had removed it from his team's hiring loops that month. His reasoning is in a blog post he published that same week, titled "Hunger beats tenure: Advice for junior engineers." He writes that the coding interview was a "weak but useful predictor" of how successful a candidate would be on his teams. In his account, that is no longer true. A small React challenge tells a manager about as much as quizzing someone on assembly, and Claude is now better at both. A separate Anthropic page written by Tristan Hume, a lead on the company's performance optimization team, describes the same problem in its take-home test. Since early 2024 that team has used a take-home test in which candidates optimize code for a simulated accelerator. More than 1,000 candidates have completed it. Hume writes that Claude Opus 4 outperformed most human applicants when given the same time limit, and that Claude Opus 4.5 then matched even the strongest candidates. He redesigned the test, cutting the limit from four hours to two and later switching to puzzles built on a tiny, heavily constrained instruction set. The benchmarks he publishes show what the models achieved. Claude Opus 4.5 reached 1579 cycles after two hours in the company's test-time compute harness and 1487 cycles after 11.5 hours. A casual Claude Code session reached 1790 cycles, approximately matching the best human result within two hours. These are Anthropic's own measurements, run on Anthropic's own harness. They show what the company's models did under its setup, not an independent comparison. His argument is that building a working prototype has become cheap, so impact is less bounded by years of accumulated knowledge of frameworks and libraries. He cites two relatively new Anthropic engineers, Anthony Morris and Alice Zhao. They built a working prototype of a graphical interface for Claude Code over one weekend, and the first public version shipped roughly eight days later. He also says he has not found a replacement for the interview. He is now scanning resumes and portfolios for "little prototypes that exhibit taste, empathy, and principled thinking." He asked others to tell him if they have found something better. Jaana Dogan, a principal engineer at Google, replied on X that she would be interested in interviewing how people use coding agents, because that gives her more signals about their coding skills. The wider industry appears to be moving more cautiously than that reply suggests, according to The Register. Nicole Bachaud, a labor economist at ZipRecruiter, told The Register that "We are not seeing tech companies abandon skills assessments widely across the industry, at least not yet." Her firm's October 2026 Employer Survey of 1,500 employers found that 62.5 percent of employers in the information sector require formal skill tests on initial applications. TechSpot reported that about 70 percent have moved those assessments to proctored or in-person sessions to keep AI out. Anthropic's own candidate guidance still asks applicants not to use Claude in live interviews and take-homes unless told otherwise. Rieseberg changed the process only for his own team, so candidates applying to other groups at the company will likely still face a coding round. ## Why it matters If AI outperforms candidates on standard coding tests, hiring teams may need to rethink how they assess engineering skill. ## What you can do Review whether your technical interviews still measure skills that AI tools cannot easily replicate. ## Original source - [Hunger beats tenure: Advice for junior engineers](https://felixrieseberg.com/hunger-beats-tenure-advice-for-junior-engineers/) - [Felix Rieseberg on X: "The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it](https://x.com/felixrieseberg/status/2115200995546566910) - [Designing AI resistant technical evaluations \ Anthropic](https://www.anthropic.com/research/AI-resistant-technical-evaluations) ## Further reporting - [Anthropic engineer declares the coding interview dead](https://www.theregister.com/ai-and-ml/2026/10/28/anthropic-engineer-declares-the-coding-interview-dead/5306115) - [Anthropic manager stops coding interviews because Claude beats most candidates | TechSpot](https://www.techspot.com/news/114391-anthropic-manager-drops-coding-interviews-after-realizing-claude.html) --- AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.