An engineering manager at Anthropic advises new employees to cut back on coding practice problems and leverage AI to bring their ideas to life.

Beating AI News, Anthropic engineering manager Felix Rieseberg announced this month his team has eliminated traditional coding interviews in hiring. Rieseberg, who previously worked on Claude.ai, Cowork, and Claude Code Desktop, noted that in the past, interviewers would ask candidates to write code on the spot (such as completing a React programming problem) to assess their coding ability. But Claude now performs these tasks far better, and he observed that the highest-performing engineers on his team are not always the best at such exams. He also found that some entry-level engineers, with the help of AI, are already delivering better results than senior staff. In the past, new hires with ideas often had to draft proposals, convince supervisors, and wait for design and development resources; now they can directly build product prototypes with AI, backed by real-world results. There’s an internal example at Anthropic: employees have long used Claude Code in their terminals, but two recently joined engineers, Anthony Morris and Alice Zhao, built a more user-friendly graphical interface for it. The pair completed the prototype in just one weekend and launched the first public version about eight days later. Rieseberg advised young engineers not to focus solely on learning programming frameworks or practicing interview questions. Instead, they should spend more time understanding company business, identifying real problems, and using AI to build solutions. He believes AI is narrowing the technical gap between new and senior employees, meaning experience and tenure no longer hold the same advantage as before. However, after ditching traditional coding interviews, he has not yet found a better evaluation method. Currently, he mainly reviews candidates’ resumes and work samples, with a particular focus on small projects that demonstrate problem-solving skills and product judgment.