Amazon unveils Nova Act, an AI agent with SDK, to control web browsers

Amazon has introduced Nova Act, a general-purpose AI agent designed to navigate web browsers and execute straightforward tasks autonomously. This unveiling coincides with the release of the Nova Act SDK, a developer toolkit intended to facilitate agent prototype creation.

Developed within Amazon’s AGI lab in San Francisco, Nova Act is set to enhance Alexa+, the AI-driven upgrade to Amazon’s voice assistant. While the currently available version is a research preview, developers can explore the Nova Act toolkit via nova.amazon.com. The platform showcases Amazon’s Nova foundation models.

With Nova Act, Amazon challenges OpenAI’s Operator and Anthropic’s Computer Use, entering the domain of general-purpose AI agents. The belief among tech leaders is that AI agents capable of web navigation will significantly augment the utility of AI chatbots.

While not the first in this field, Amazon’s integration through Alexa+ could potentially ensure widespread adoption. Amazon highlights that developers using the Nova Act SDK can automate tasks like ordering from Sweetgreen or making dinner reservations. The toolkit equips AI agents to interact with web pages, complete forms, and select dates on calendars.

Amazon reports that Nova Act surpasses agents from OpenAI and Anthropic in internal evaluations. For instance, on ScreenSpot Web Text, Nova Act achieved a 94% score. This exceeds OpenAI’s CUA (88%) and Anthropic’s Claude 3.7 Sonnet (90%). However, evaluations using standard benchmarks like WebVoyager were not disclosed.

Spearheaded by former OpenAI researchers David Luan and Pieter Abbeel, Amazon’s AGI lab introduces its first public offering with Nova Act. Luan views AI agents as crucial for advancing superintelligent AI. 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 automate simple tasks reliably. It offers developers tools to define human intervention points in agent workflows. The aim is reliable, though not fully autonomous, applications. Despite a competitive landscape, Amazon’s agent is a pivotal technology. Initial Nova Act tests could preview the capabilities of the awaited Alexa+, marking a critical juncture for Amazon’s AI initiatives.

Early AI agents from companies such as OpenAI, Google, and Anthropic have exhibited reliability issues across various domains. Tests reveal slow performance, limited independent operation, and errors. The industry watches to see if Amazon’s agents can overcome these common flaws.

Also Read : AWS Launches Agentic AI Division to Drive Next-Gen Automation