Amazon unveiled Nova Act on Monday, a general-purpose AI agent capable of controlling web browsers and independently performing simple tasks. Simultaneously launched was the Nova Act SDK, enabling developers to build agent prototypes using Nova Act.

Developed by Amazon's newly established AGI lab in San Francisco, Nova Act will provide key functionality for the company's upcoming Alexa+ (a generative AI-enhanced version of Amazon's voice assistant). However, the currently released version of Nova Act is positioned as a "research preview," and developers can access the Nova Act toolkit via nova.amazon.com.

AWS, Amazon, Cloud Services, Amazon, Cloud Computing, Servers

This product represents a clear move by Amazon to compete with OpenAI's Operator and Anthropic's Claude. Many tech giants believe that AI agents capable of navigating the web for users will significantly enhance the practicality of current AI chatbots. While Amazon isn't a pioneer in this field, its reach through Alexa+ could potentially be the widest.

According to Amazon, developers using the Nova Act SDK can automate basic tasks for users, such as online ordering food or making reservations. The toolkit supports integrating multiple functions, allowing the AI agent to browse web pages, fill out forms, or select dates on a calendar.

Amazon claims Nova Act outperformed competitors in internal testing. In the ScreenSpot Web Text evaluation, Nova Act scored 94%, exceeding OpenAI's CUA (88%) and Anthropic's Claude 3.7 Sonnet (90%). However, Amazon did not benchmark Nova Act against more common agent evaluations like WebVoyager.

Nova Act is the first publicly released product from the AGI lab, co-led by former OpenAI researchers David Luan and Pieter Abbeel. Both founded their own AI startups—Luan founded Adept, and Abbeel co-founded Covariant—before being hired by Amazon last year to lead its AI agent efforts.

Luan told TechCrunch that he believes agents are a crucial step in creating superintelligent AI systems, defining AGI as "an AI system capable of helping accomplish everything humans do on computers." He stated that the team designed the Nova Act SDK to reliably automate short tasks and allow developers to precisely define when human intervention is needed in the workflow.

A major challenge facing early AI agents is cross-domain reliability. In testing, existing systems are generally slow, struggle to operate independently for extended periods, and are prone to making mistakes that humans wouldn't. The market will soon see whether Amazon has overcome these limitations or if its agent suffers from the same problems that plague its competitors.