For many months now, AI companies have made a huge deal out of “AI agents,” meaning autonomous software systems that can make decisions and take actions on behalf of humans with minimal intervention.

But when that ambitious vision will turn into a reality remains anybody’s guess. The current crop of agentic AI models is still getting easily tripped up, often requiring humans to jump in, effectively undercutting their purpose.

The numbers remain dismal. Researchers at Carnegie Mellon University found earlier this year that even the best-performing AI agent, which was Google’s Gemini 2.5 Pro at the time, failed to complete real-world office tasks 70 percent of the time.

And this summer, OpenAI released its ChatGPT agent, promising that it can “do work for you using its own computer, handl

See Full Page