Amazon unveiled Nova Act on Monday, a general-purpose AI agent designed to autonomously navigate web browsers and execute basic tasks. Accompanying this launch is the Nova Act SDK, a developer toolkit for building AI-driven agent prototypes.
Developed by Amazon’s newly established AGI lab in San Francisco, Nova Act will also serve as the backbone of Alexa+, an AI-enhanced upgrade to Amazon’s popular voice assistant. However, the version released today is labeled as a research preview, indicating it is still in development.
Developers can access the Nova Act toolkit via a dedicated website, nova.amazon.com, which also showcases Amazon’s broader Nova foundation models.
Nova Act positions Amazon as a competitor to OpenAI’s Operator and Anthropic’s Computer Use, both of which employ AI agent technology. Many tech companies see AI agents capable of independent web navigation as a significant advancement beyond traditional AI chatbots.
Although Amazon is not the first to introduce such technology, its integration with Alexa+ could give it the largest user base.
Amazon states that developers using the Nova Act SDK will be able to automate simple tasks such as placing food orders or making reservations. The toolkit enables AI agents to interact with web pages, complete forms, and select dates on calendars.
The company claims Nova Act outperforms OpenAI and Anthropic agents on internal tests. For instance, on the ScreenSpot Web Text benchmark, which measures an AI agent’s ability to interact with on-screen text, Nova Act scored 94%, surpassing OpenAI’s CUA (88%) and Anthropic’s Claude 3.7 Sonnet (90%). However, Amazon did not test Nova Act against widely recognized benchmarks like WebVoyager.
Nova Act is the first public product from Amazon’s AGI lab, led by former OpenAI researchers David Luan and Pieter Abbeel. Luan, who previously founded Adept, and Abbeel, co-founder of Covariant, were recruited by Amazon last year to lead its AI agent initiatives.
While automating tasks like ordering food may seem far removed from artificial general intelligence (AGI), Luan views AI agents as a step toward developing superintelligent systems. He defines AGI as “an AI system that can help you do anything a human does on a computer.”
The Nova Act SDK is designed to reliably automate short, simple tasks while allowing developers to specify when human intervention is needed in an agent’s workflow. The goal is to create dependable AI applications, albeit not entirely autonomous ones.
Amazon enters a competitive AI agent market, but the success of Nova Act could be critical for its broader AI ambitions, including Alexa+. Early tests may provide insights into the capabilities of the long-anticipated assistant upgrade, a crucial moment for Amazon’s AI strategy.
Many early AI agents from OpenAI, Google, and Anthropic have struggled with reliability, often operating slowly and making errors that a human would avoid. The coming months will reveal whether Amazon has solved these challenges or if Nova Act will face the same hurdles as its competitors.
