INSIGHT · GLOBAL
AI agents complete only 26% of assigned work tasks in real-world financial deployments
OpenAI data cited in Tearsheet analysis shows AI agents succeed on only one in four real-world work assignments in production financial services environments. Task failure rates remain high despite frontier model improvements.
WHY IT MATTERS
Deflates hype around autonomous AI agents for back-office; forces BFSI to pivot from agent-as-replacement to agent-as-augmentation, requiring new governance and human-in-loop frameworks.
Source: Tearsheet · 2026-07-28