AI Agents Fail 95% of Tasks, New Research Reveals

New research reveals that AI agents rarely meet professional work standards. A benchmark from Scale AI and the Center for AI Safety found that even advanced AI agents succeed less than five percent of the time when completing real-world tasks from start to finish. This benchmark, called the Remote Labor Index, measures how well AI agents perform on paid digital work. It found that the top-performing agent automated only 2.5% of projects to a professional standard at its launch. This number has barely moved, remaining below five percent. The research focuses on an AI agent's ability to complete an entire task, like a human professional would. Tasks were taken from freelance platforms and covered 23 sectors, including video editing and data analysis. The key issue is reliability. AI agents can complete parts of tasks but struggle with end-to-end completion. This information is important for understanding the current capabilities of AI in professional settings.