AI can’t automate remote work yet, study finds

Widely used artificial intelligence (AI) systems, such as Manus and ChatGPT, can automate no more than 2.5% of complex remote work projects.

About one-third of American employees are worried that AI may reduce their future work opportunities, according to a Pew Research Center survey, and those working remotely are no exception.

However, a recent study by the Center for AI Safety brings good news to teleworkers. It found that AI tools achieved automation rates ranging between only 1.3% and 2.5% in fields like product design, data analysis, and video animation.

Can AI automate jobs?

We created the Remote Labor Index to test AI’s ability to automate hundreds of long, real-world, economically valuable projects from remote work platforms.

While AIs are smart, they are not yet that useful:
the current automation rate is less than 3%. pic.twitter.com/yLudf6Kc6l

undefined Dan Hendrycks (@DanHendrycks) October 29, 2025

The study tested six AI systems: Manus, Grok 4, Sonet 4.5, GPT5, ChatGPT agent, and Gemini 2.5 Pro to complete the following projects:

• Build an interactive dashboard for exploring data from the World Happiness Report
• Create 3D animations to showcase the features of a new earbuds design and case
• Create a 2D animated video advertising the offerings of a tree services company
• Develop architectural plans and a 3D model for a container home based on an existing PDF design
• Build a brewing-themed version of the “Watermelon Game,” where players merge falling objects to reach the highest level item
• Format a paper using the provided figures and equations for an IEEE conference

In total, the projects included in the study represent over 6,000 hours of real work valued at over $140,000.

After the AI systems completed the tasks, human evaluators calculated the automation rate, which is the percentage of projects that AI delivered at least as well as humans did.

The study found that the best-performing model, Manus, achieved an automation rate of 2.5%.

The automation rates for Grok and Sonet were 2.1%, followed by GPT5 and ChatGPT agents, with 1.7% and 1.3%, respectively.

According to the study, Gemini 2.5 Pro performed the worst, with an automation rate of 0.8%.

“This demonstrates that contemporary AI systems fail to complete the vast majority of projects at a quality level that would be accepted as commissioned work,” the authors wrote.

Poor quality was the most common AI failure, detected in 45.6% of AI deliverables, followed by incompletion, which affected 35.7% of deliverables.

Some failures included AI producing eight-second videos when eight-minute videos were requested, or a changing house’s appearance across different 3D views.

However, researchers noted that models are steadily improving and that progress on these complex tasks is measurable.

Unlock more exclusive Cybernews content on YouTube.