Amazon has unveiled an AI agent trained to independently perform actions, such as order and pay for things, within a web browser.
Now available as a research preview, the Nova Act SDK lets developers build agents that can do everything from submitting out of office requests, to playing video games, to apartment hunting.
Amazon says that, while some use cases are well-suited for today’s technology, multi-step agents prompted with high-level goals still require constant human hovering and supervision. Nova Act SDK enables developers to break down complex workflows into reliable atomic commands.
In one example, Amazon technical staff member Vishal Vora runs through how an agent is set up to visit the Sweetgreen restaurant site every Tuesday night, add the salad he likes to a cart, choose to get it delivered to his address, add a tip and check out and pay.
Vora says that if you have to "babysit" an AI, it's not really automation. Once the Nova Act agent's workflow is setup, users can switch on "headless mode" meaning that they never have to take any actions.
"Our dream is for agents to perform wide-ranging, complex, multi-step tasks like organizing a wedding or handling complex IT tasks to increase business productivity," says Amazon.