A new study by the online work marketplace Upwork shows that artificial intelligence agents frequently fail to complete professional tasks on their own. However, their performance improves dramatically when they collaborate with human experts, with project completion rates increasing by up to 70 percent. Michael Nuñez reports for VentureBeat that this is the first major evaluation of human and AI collaboration on real-world projects.
The research suggests the future of work will rely on partnership rather than replacement. “AI agents aren’t that agentic, meaning they aren’t that good,” Andrew Rabinovich, Upwork’s head of AI, told VentureBeat. He added that when paired with expert professionals, their performance improves significantly.
Upwork tested leading AI systems like OpenAI’s GPT-5 and Google’s Gemini 2.5 Pro on more than 300 real client jobs. The study found that agents performed best on technical tasks with clear solutions, such as coding and data science. For example, one model completed 64 percent of data science projects alone, but this figure jumped to 93 percent with human feedback.
In contrast, the AI systems struggled with creative and qualitative work that required judgment, such as writing, marketing, and design. These areas saw the most significant improvement from human input. The findings indicate that while AI can handle routine tasks, human expertise remains essential for creativity, context, and quality control.
