Center for AI Safety study reveals AI struggles with complex remote work, completing Less than 3%

A Center for AI Safety study found top AI agents automated less than 3% of remote work tasks acceptably; Manus AI succeeded 2.5%, GPT-5 just 1.7%. Failures included poor quality and incomplete results. Despite layoffs citing AI, experts warn current AI lacks long-term memory and skill learning, challenging claims of imminent full job automation.