Amazon has introduced an AI agent capable of autonomously executing tasks, such as ordering and making payments, within a web browser.
Now available as a research preview, the Nova Act SDK enables developers to create agents that can handle a variety of activities, from submitting out-of-office requests to gaming and apartment hunting. While some applications are well-suited for current technology, multi-step agents that respond to high-level goals still require significant human oversight. The Nova Act SDK allows developers to decompose complex workflows into manageable, reliable commands.
In one demonstration, Amazon technical staff member Vishal Vora illustrates how an agent can be programmed to visit the Sweetgreen restaurant website every Tuesday night, add his preferred salad to the cart, select delivery to his address, include a tip, and complete the checkout process. Vora notes that if an AI needs constant monitoring, it doesn’t qualify as true automation. Once set up, the Nova Act agent can operate in “headless mode,” eliminating the need for user intervention.
“Our vision is for agents to handle a wide range of complex, multi-step tasks, like organizing a wedding or managing intricate IT projects, ultimately enhancing business productivity,” states Amazon.