On Monday, Amazon unveiled Nova Act, a groundbreaking general-purpose AI agent capable of autonomously managing web browsers to perform basic actions. This innovative tool is accompanied by the Nova Act SDK, a toolkit designed for developers to create AI agent prototypes powered by this technology.

Key Features of Nova Act

The Nova Act, developed at Amazon’s recently established AGI lab in San Francisco, is set to enhance the future Alexa+ voice assistant, which will feature generative AI capabilities. However, the debut version of Nova Act is classified as a research preview, suggesting it is still in the early stages of development.

Developer Access and Tools

Developers can access the Nova Act toolkit through a newly launched website, nova.amazon.com. This site not only provides the SDK but also showcases Amazon’s various foundational models associated with the Nova project.

With the Nova Act SDK, developers can automate essential tasks for users, such as placing orders at restaurants or organizing calendar events. The toolkit enables the creation of applications that facilitate navigation of web pages, form completion, and scheduling.

Performance Insights

According to internal evaluations, Amazon claims that Nova Act outperforms competitor technologies from OpenAI and Anthropic on several metrics. For instance, in the ScreenSpot Web Text evaluation, Nova Act achieved a score of 94%, surpassing OpenAI’s CUA, which scored 88%, and Anthropic’s Claude 3.7 Sonnet, earning 90%. Notably, Amazon did not employ more common agent evaluations, such as WebVoyager, for benchmarking.

Leadership Behind Nova Act

Nova Act marks the first significant product from Amazon’s AGI lab, which is co-led by former OpenAI researchers David Luan and Pieter Abbeel. Having previously founded startups focused on AI development, Luan and Abbeel joined Amazon to drive the company’s ambitions in AI agents.

Luan envisions AI agents as a pivotal step toward achieving Artificial General Intelligence (AGI). He defines AGI as a system capable of performing any task a human can execute on a computer.

The Future of Agent Technology

The design of the Nova Act SDK aims to allow for the reliable automation of simpler tasks while ensuring developers have mechanisms to involve human oversight when necessary. This could potentially lead to more dependable applications despite not achieving full autonomy.

Despite entering a highly competitive landscape with numerous AI agents already available, Amazon sees substantial potential in the technology represented by Nova Act, considering it vital for the company’s ongoing AI initiatives. Initial tests may shed light on the capabilities of the anticipated Alexa+ and play a critical role in defining Amazon’s AI trajectory.

Challenges and Outlook

Early iterations of AI agents from competitors have faced criticism for their inconsistencies and operational challenges. Issues such as sluggish response times and a propensity for errors highlight the difficulties these systems encounter in navigating various domains independently. The success of Nova Act will be closely monitored to determine if it can overcome existing shortcomings that affect other AI agents.