Nova Act from Amazon performs browser actions: AI boasts 90% precision

Amazon

Amazon - Foto: Eric Broder Van Dyke / Shutterstock.com

Amazon revealed on March 31, 2025, a groundbreaking artificial intelligence tool called Nova Act, designed to autonomously perform tasks within a web browser without human oversight. Emerging from the Nova project under the company’s AGI lab, this AI can schedule meetings, book tickets, fill out forms, or order food delivery, achieving over 90% accuracy in internal tests. Initially launched as a software development kit (SDK) for programmers, Nova Act is available in a preview version for users in the United States, allowing developers to craft customized routines. Outperforming rivals like OpenAI’s Operator in precision, this tool signals Amazon’s ambitious push into AI automation, with plans to integrate it into the enhanced Alexa+ assistant. While currently limited to the U.S., the release highlights a leap forward in digital task management, appealing to both developers and future everyday users, though Brazil and other regions await availability.

Capable of navigating websites, completing transactions, and even playing simple browser games, Nova Act offers unparalleled flexibility. During demonstrations, Amazon showcased the AI ordering a specific meal from a delivery site and searching for apartments based on biking distance from a set location. Its accuracy, exceeding 90% in fragmented tasks like clicking buttons or dodging pop-ups, sets it apart from competitors. Available in Python with support for various libraries, the SDK empowers developers to experiment with its capabilities, marking a significant step in Amazon’s broader Nova initiative, launched in late 2024 to rival generative AI leaders.

The Nova project positions Amazon as a formidable player in the AI landscape, building on a global automation market valued at roughly 200 billion dollars in 2024. Unlike traditional assistants, Nova Act’s “agentic” design enables it to analyze data and make decisions within predefined guidelines, such as skipping unwanted extras in an order. With its integration into Alexa+ on the horizon, priced at 19.99 dollars monthly (about 115 reais), Amazon aims to bring this technology to homes via Echo devices, though no timeline exists yet for international rollout, leaving global users eager for updates.

Initial features of Nova Act

Developed within Amazon’s AGI lab, Nova Act represents a milestone in multimodal AI. Programmers can configure it for specific actions, such as reserving a conference room or purchasing movie tickets, with detailed instructions. In showcased examples, the AI handled delivery orders by avoiding add-ons and played a turn-based pigeon-battling RPG in a browser, demonstrating its adaptability across diverse tasks.

Boasting over 90% accuracy in internal evaluations, Nova Act excelled in navigating complex web interfaces and completing segmented actions, outperforming tools like Anthropic’s Claude 3.5 Sonnet. The SDK, released in preview, has already drawn interest from U.S. developers, accessible via Amazon’s official platform, signaling early adoption potential among tech enthusiasts and professionals.

Plans for Alexa+ integration

Nova Act is set to enhance Alexa+, Amazon’s upgraded virtual assistant. Priced at 19.99 dollars monthly, Alexa+ blends generative AI with voice commands and will soon leverage Nova Act’s browser capabilities. Echo devices, including smart speakers, will pioneer this integration in the U.S., though Brazilian users remain without a confirmed launch date.

  • Planned Nova Act features in Alexa+:
    • Automatic scheduling of appointments via browser.
    • Voice-activated personalized delivery orders.
    • Online product searches with user-defined filters.

How Nova Act operates

Offered as an SDK, Nova Act requires basic Python skills to configure. Developers can program it for straightforward tasks like “search flights” or intricate routines like compiling rental listings based on price and proximity. The AI interacts with digital interfaces, clicks buttons, and fills fields autonomously after initial setup.

In one demo, Amazon showed Nova Act ordering a specific pizza every Friday from a delivery site, bypassing extras like soda. Another example featured it filtering apartments within a 10-minute bike ride from a location, underscoring its precision in following detailed directives. Its Python compatibility and support for popular libraries make it developer-friendly, while its independent operation distinguishes it from less flexible systems.

Defining an agentic AI

Unlike conventional AIs that merely respond to prompts, an agentic AI like Nova Act performs tasks independently within set parameters. It gathers data, evaluates options, and acts, such as selecting an optimal meeting time or progressing in a browser game. Powered by generative models, this “reasoning” ability enables seamless interaction with digital environments.

Nova Act joins a growing field of agentic tools, including OpenAI’s Operator for basic browser tasks and Google’s forthcoming Project Jarvis. Anthropic’s Claude 3.5 Sonnet offers similar automation, but Amazon emphasizes Nova Act’s superior accuracy and integration with its ecosystem, such as Alexa, as key differentiators.

Timeline of the Nova project

Nova Act stems from Amazon’s broader AI efforts. Key milestones include:

  • November 2024: Nova project debuts as a generative AI model.
  • January 2025: AGI lab ramps up multimodal AI testing.
  • March 31, 2025: Nova Act launches as an SDK in the U.S.
  • 2025 (TBD): Integration with Alexa+ planned for Echo devices.

Competition in the AI landscape

Amazon faces stiff competition in the agentic AI race. OpenAI’s Operator, introduced in 2024, handles tasks like online bookings, while Anthropic’s Claude 3.5 Sonnet automates browser navigation. Google’s Project Jarvis, still under wraps, looms as a future rival. Nova Act stands out with its 90%-plus accuracy and ties to Amazon’s vast infrastructure, including e-commerce and smart devices.

The generative AI market, valued at 1.2 trillion dollars globally in 2024, is projected to double by 2028. Amazon, commanding 37% of U.S. e-commerce per recent data, leverages Nova Act to strengthen its tech dominance, challenging Google and Microsoft in the automation and AI sectors.

Benefits for developers and users

The Nova Act SDK empowers developers to build tailored tools. Businesses can automate processes like delivery scheduling or report filing, while individuals might set up weekly shopping routines. Limited to the U.S. for now, its global rollout is anticipated given Amazon’s track record.

Integration with Alexa+ will extend automation to households. Voice commands like “order my favorite pizza” or “book a doctor’s visit” will trigger browser actions, enhancing Alexa’s utility. Nova Act’s high accuracy in real-world scenarios minimizes errors seen in competing AIs, such as incorrect website selections.

The future of digital automation

Nova Act reflects a shift toward AIs that execute practical tasks beyond answering queries. In the U.S., where 68% of adults use virtual assistants regularly per 2024 surveys, its potential is vast. In Brazil, where Alexa thrives in cities like São Paulo and Rio de Janeiro, Nova Act’s arrival could boost smart device adoption further.

Amazon aims to enhance the SDK with multilingual support and mobile app integration. Its 90% accuracy in tasks like dodging ads or filling specific fields positions it as a top-tier AI tool, pushing competitors to refine their offerings in this rapidly evolving field.

  • Nova Act capabilities:
    • Custom online shopping with precise filters.
    • Hands-free meeting scheduling.
    • Browser-based game progression via AI.