Amazon has taken a significant step towards agentic AI with the introduction of Nova Act – an intelligent software agent capable of navigating websites, filling out forms, and completing simple online tasks autonomously. While it may sound like a technical novelty at first glance, Nova Act could quickly become one of the most influential tools in the growing competition between tech giants. It’s not just an experimental product from Amazon’s newly launched AGI research lab in San Francisco, but also a glimpse into what’s to come with the next generation of Alexa+, Amazon’s AI-enhanced voice assistant.
Unlike conventional chatbots that react to prompts, Nova Act actively interacts with graphical user interfaces on the web. In practical terms, this means it can order a salad from Sweetgreen, schedule a calendar event, or complete an online booking – all without the user manually guiding each step. It brings Amazon closer to a vision of AI that can operate a computer like a human – a central idea in the concept of Artificial General Intelligence.
Although Nova Act is currently available only as a research preview, Amazon is already offering a full SDK that allows developers to build their own agent prototypes. At nova.amazon.com, developers can explore the underlying foundation models and access a toolkit designed to ensure agents behave reliably and securely. One of the key features, according to Amazon, is that developers can define exactly when an agent is allowed to act independently and when human input is required.
Amazon has also shared performance benchmarks. In internal tests such as the ScreenSpot Web Text evaluation, Nova Act achieved an impressive 94 percent, outperforming OpenAI’s CUA at 88 percent and Anthropic’s Claude 3.7 Sonnet at 90 percent. However, Amazon opted not to use more widely recognised benchmarks like WebVoyager, raising questions about broader comparability.
Nova Act is the first public product from Amazon’s AGI lab, which is led by two high-profile names: David Luan, formerly of OpenAI and founder of Adept, and Pieter Abbeel, co-founder of robotics firm Covariant. Both are considered leaders in machine learning and autonomous systems. For Luan, Nova Act is more than just a tool – it’s a strategic milestone towards true machine intelligence. He defines AGI as a system that can help you do anything a person can do on a computer.
While Nova Act currently focuses on short, simple tasks, its architecture holds greater potential. AI agents from OpenAI, Google and Anthropic still struggle with reliability, speed, and autonomy across varied use cases. Amazon is aiming for consistency and transparency, giving developers greater control while offering more independence than existing models. After all, an agent that works 90 percent of the time but fails in critical moments is not fit for everyday use.
Looking ahead to Alexa+, set to launch later this year, Nova Act may become the backbone of a new Amazon ecosystem – one that doesn’t just listen, but acts. For Amazon, this could be a pivotal test of its AI strategy. The next-generation Alexa isn’t just a product update, but a statement of intent in a fast-moving industry.
Whether Nova Act truly sets a new standard remains to be seen. But one thing is certain: Amazon is entering an already crowded field – with focus, developer tools, and possibly the strongest bridge yet between generative AI and practical action. If Amazon can deliver agents that are useful, fast, and dependable, Nova Act could mark the breakthrough the market has been waiting for.
Post picture: Amazon
