Anthropic manager stops using coding challenges in hiring

Rieseberg, an Anthropic hiring manager, said he stopped asking candidates to complete coding challenges; his post was published on 27 October 2026.[1] He said these tests no longer provide a useful signal because Claude is better at small React challenges and assembly questions.[1] He also argued that strong performance on conventional coding tests does not necessarily identify the engineers who will have the greatest impact at work.[1] Rieseberg has not settled on a replacement.[2] For now, he reviews résumés and portfolios for small prototypes that show taste, empathy and principled thinking.[1]

Anthropic’s performance-engineering take-home is a separate assessment designed and redesigned by Tristan Hume, a lead on the company’s performance optimization team.[3] More than 1,000 candidates have completed it since early 2024, Anthropic says.[3] The task asks candidates to optimize code running on a simulated accelerator by reducing its clock cycles.[4] Anthropic’s general guidance tells applicants not to use Claude on take-home assessments unless the company explicitly allows it; Anthropic indicates an exception to that guidance for this test.[5][3]

By May 2025, Anthropic found that more than half of candidates would have achieved a better result by delegating the entire task to Claude Code.[3] Anthropic also said Claude Code produced a more optimized solution than almost all humans within the four-hour limit.[3] Anthropic said the leading human candidate at the time had made heavy use of Claude 4.[3] Those results describe this particular assessment; they do not establish how Claude performs across all engineering hiring tests.[4][3]

For its third test version, Hume adopted constrained programming puzzles in which candidates must minimize instruction count.[3] Anthropic says Claude Opus 4.5 failed to solve the first medium-hard puzzle Hume tested it on, while the version also tests building debugging tools.[3] Hume said early results showed scores correlating with the caliber of candidates’ past work, but that the version sacrifices the original take-home’s realism and varied depth.[3] Anthropic says humans can outperform its models with unlimited time and released the original take-home as an open challenge; Hume invited submissions that beat Opus 4.5.[3]