Mehul Thakkar

Every week, there’s a new promise that AI will handle entire workflows from automating emails to managing complete projects. The hype sounds great, but a new benchmark from Scale AI and the Center for AI Safety shows a different reality.

In the study, AI agents were tested on real freelance tasks like writing, research, data entry, and design. Out of nearly $144,000 worth of work, the top model earned only about $1,800 completing just 2 to 3 percent of jobs successfully.

This doesn’t mean AI is weak. It means organizations are skipping the most important part of AI success: workflow design.