Amazon unveils Nova Act, an AI agent that can control a web browser
Amazon unveiled the Nova Act on Monday, an AI agent for general purposes that can take control of a web browser and can independently perform some simple actions. In addition to the new Agentic AI model, Amazon releases the Nova Act SDK, a toolkit with which developers can build agent prototypes with Nova Act.
Nova ACT, developed by Amazon’s recently opened San Francisco-Founded AGI Lab, will also be important characteristics of the upcoming Alexa+ Upgrade of the company, a generative AI-amplified version of Amazon’s popular voice assistant. However, the version of Nova Act available from today is slightly less polished. Amazon calls it a research review.
Developers have access to the Nova Act Toolkit on a new website, Nova.amazon.comIt also serves as a showcase for the different Nova Foundation models from Amazon.
Nova Act is Amazon’s attempt to accept the Operator of OpenAI and the computer use of Anthropic with its own AI agent technology. Various leading technology companies believe that AI agents can navigate on the internet for users, make today’s AI chatbots considerably more useful.
Amazon may not be the first to develop this type of agentic technology, but it can have the widest range via Alexa+.
Amazon says that developers with the Nova Act SDK should be able to automate basic promotions on behalf of users, such as ordering salads at Sweetgreen or making dinner reservations. With the Nova Act Toolkit, developers can bring together tools with which an AI agent can navigate web pages, fill in forms or choose data on a calendar.
Amazon claims that Nova Act performs better than agents from OpenAi and Anthropic on various of the company’s internal tests. For example, on Screenspot -Webstekst, which measures how an AI agent interacts with text on a screen, scored Nova Act 94%, scored better than the CUA of OpenAI (which scored 88%) and the Claude 3.7 -Sonnet of Anthropic (90%).
However, Amazon is not benchmark Nova Act with the help of more usual evaluations from agent, such as web voyager.
Nova ACT is the first public product that comes from Amazon’s earlier AGI Lab, an initiative Led by former OpenAi researchers David Luan and Pieter Abbeel. Both previously established startups of their own startups – Luan started skilled, while Abbeel co -founder of Covarian – before Amazon hired them last year to lead the efforts of AI agent.
Although it may seem strange that an AGI -LAB AI agents can order that Sweetgreen can order, Luan told WAN that he sees agents as an important step in the direction of creating super intelligent AI systems. Luan defines Agi as “an AI system that can help you do everything a person does on a computer.”
Luan says that his team has designed the Nova Act SDK to reliably automate short, simple tasks and give developers aids to determine exactly when they want a person to intervene in an agent workflow. He hopes that developers will be able to create more reliable agent applications, although not necessarily completely autonomous.
Amazon releases his first generalist AI agent in a busy space, but it is a crucial technology that the company has a lot on. Early tests from Nova Act can give a look at some of the possibilities of the long delayed Alexa+, a make-or-break moment for the AI efforts of Amazon.
A major problem with early AI agents from OpenAi, Google and Anthropic is their reliability in different domains. In the TechCrunch tests, the systems are slow, struggle to work independently for a long time and are susceptible to mistakes that a person would not make. It will not be long before we see if Amazon has cracked the code – or whether the agents suffer from the same mistakes that competitors test too much.
