Gartner estimates that over 40% of agentic AI projects will be canceled by 2027 due to high costs, unclear business value, or inadequate risk controls, yet 60% may continue, despite the low success rates for AI agents completing multi-step tasks (30-35%). Many AI vendors misrepresent their offerings, contributing to “agent washing,” as only about 130 of thousands genuinely exhibit agentic capabilities. Researchers from Carnegie Mellon University developed a benchmark, TheAgentCompany, revealing that even the best AI agents only completed around 30% of tasks, highlighting significant limitations such as failures in communication and task execution. Similarly, Salesforce’s benchmarking on CRM tasks showed performance degradation from 58% in single-turn to 35% in multi-turn interactions, underscoring the inadequacy of current models. Gartner predicts that by 2028, AI agents could autonomously handle 15% of daily work decisions, suggesting potential growth in useful applications, despite current shortcomings.
Source link
Share
Read more