Anthropic has long been saying that white-collar jobs are set to be disrupted by AI, and it’s seen some compelling evidence for its claims from its own internal hiring tests.
Anthropic has revealed that its latest Claude Opus 4.5 model achieved the highest score ever recorded on a challenging performance engineering take-home exam used to evaluate job candidates. Working within the standard two-hour time limit imposed on human applicants, the AI model surpassed every human candidate who has ever taken the assessment.
Anthropic characterizes the exam as “notoriously difficult,” designed specifically to evaluate technical ability and judgment under time pressure. The company now uses this test as an internal benchmark for assessing new AI models alongside its original purpose of screening engineering candidates.
However, Anthropic was quick to contextualize the achievement’s limitations. The company emphasized that the exam doesn’t measure other essential professional competencies that candidates bring to the role, including collaboration skills, communication abilities, and the intuition that develops through years of hands-on experience.
The result nonetheless marks a significant milestone in AI capabilities and raises broader questions about artificial intelligence’s expanding role in technical professions. Anthropic acknowledged these implications, noting that the outcome prompts questions about “how AI will change engineering as a profession.”
The company is actively investigating these dynamics through dedicated research initiatives. Its Societal Impacts and Economic Futures research programs are examining how AI-driven changes will unfold across multiple industries and occupations. Anthropic indicated it plans to release additional findings from this research in the near future. Anthropic has long been talking about AI-led job loss — CEO Dario Amodei had previously said that 90% of coding would be done with AI in 6 months, and at the end of 6 months had declared that 90% of coding at Anthropic was indeed being done by AI. He has also said that AI could eliminate 50% of entry-level white-collar jobs by 2030, and with the company’s AI systems doing better than all human candidates at real tests to assess their engineering ability, it appears that his prediction could well be on track.