Digital Economy

Amazon launches Nova Act: the new AI agent that autonomously navigates the web

The new technology core functionality of the future Alexa+ update

3 April 2025

3' min read

Amazon enters the arena of agent-based artificial intelligence with the launch of Nova Act, a sophisticated multipurpose AI agent designed to autonomously interact with web browsers in performing everyday tasks. The innovative technology represents a significant step forward in the Seattle-based company's strategy to gain market share in this emerging and competitive technology sector where OpenAI's Operator and Anthropic's Computer Use are already present.

The Nova Act project and the AGI workshop

Nova Act is the first public product of the AGI Lab in San Francisco, an initiative led by former OpenAI researchers David Luan and Pieter Abbeel. Both had already founded their own startups - Luan created Adept, while Abbeel co-founded Covariant - before being hired by Amazon last year to lead its efforts in the field of AI agents.

David Luan explained that he designed the Nova Act SDK, a toolkit for prototyping agents, with the aim of reliably automating short and simple tasks. The tools provided allow developers to precisely define when human intervention is required within an agent workflow. The hope is that this will enable the realisation of more reliable, though not completely autonomous, agentic applications. Developers can access these tools through the nova.amazon.com portal, which also serves as a showcase for Amazon's entire range of Nova models.

Advanced features and performance

Nova Act's suite of tools includes advanced capabilities that enable AI agents to intuitively navigate complex web pages, fill out forms with relevant data, interact with interactive elements such as calendars or date selectors, and even understand the visual context of website user interfaces, a problem that has historically hampered the effectiveness of automated agents.

Amazon claims that Nova Act outperformed competing OpenAI and Anthropic agents in various internal tests. For example, in the ScreenSpot Web Text test - which evaluates the interaction of an AI agent with text displayed on the screen - Nova Act scored 94 per cent, outperforming OpenAI's CUA (88 per cent) and Anthropic's Claude 3.7 Sonnet (90 per cent). It should be noted, however, that Amazon did not use more popular benchmarks, such as WebVoyager, for evaluating agents.

Alexa+ integration

Amazon is not the first company to develop agent technology of this kind, but what distinguishes its approach is the extraordinary distribution opportunity that the Alexa ecosystem offers, implementing these advanced functionalities in Echo devices already present in tens of millions of homes worldwide.

Amazon revealed that the Alexa+ user experience will be significantly enhanced by Nova Act capabilities. Users will be able to delegate tasks to the voice assistant that previously required direct human intervention, such as managing appointments, organising events, or searching for and purchasing products online. According to inside sources, the new Alexa+ will be able to maintain much more sophisticated contextual conversations, remembering users' preferences and habits to offer personalised assistance far beyond its current capabilities.

Application development perspectives

Amazon said that developers adopting the Nova Act SDK will have access to an ecosystem of tools designed to democratise the creation of agent experiences. The toolkit, designed to balance power and simplicity, offers modular components that allow developers to focus on business logic while avoiding the technical complexities of web interaction.

Industry experts suggest that this integration could represent a turning point both for the future of Alexa and for the mainstream adoption of agentic technology, gradually transforming the voice assistant from a simple reactive tool to a true proactive assistant capable of anticipating and meeting users' needs with an unprecedented level of autonomy.