Meet Nova Act: Amazon’s AI Agent for Web Automation
Amazon has just thrown its hat into the rapidly evolving ring of AI agents with the unveiling of Nova Act, a general-purpose AI system capable of navigating the web and performing basic online tasks autonomously. Announced Monday, Nova Act is more than just a research preview — it's a signal of Amazon’s intent to compete head-on with OpenAI and Anthropic in the race to build practical, web-savvy AI agents.
What is Nova Act?
At its core, Nova Act is an AI agent that can take control of a browser, execute web-based tasks, and interact with interfaces the way a human might — clicking, filling forms, and making selections. While the current version is an early-stage release, Amazon is already positioning it as a foundation for its upcoming Alexa+ upgrade — a generative AI-enhanced reimagining of its voice assistant with agentic capabilities baked in.
The new agent comes with the Nova Act SDK, a toolkit designed to help developers prototype agent applications. It’s available now at nova.amazon.com, where Amazon is also showcasing its broader Nova foundation model lineup.
A New Contender in the Agent Wars
Nova Act enters a space already buzzing with competitors. OpenAI’s Operator and Anthropic’s Computer Use have both aimed to bridge the gap between passive AI chatbots and active, goal-completing agents. Amazon’s edge? Distribution. Through Alexa+, it may be able to bring agentic AI to millions of devices overnight — something no rival can currently match.
Early use cases Amazon envisions are practical and relatively simple: placing a Sweetgreen order, making restaurant reservations, or picking a date on a calendar. Tasks that are tedious for users — but manageable for AI with the right guardrails.
Performance and Benchmarks
Amazon claims Nova Act is not just functional — it's performant. On internal tests like ScreenSpot Web Text, Nova Act achieved a 94% success rate, besting OpenAI’s CUA (88%) and Anthropic’s Claude 3.7 Sonnet (90%). That said, the company hasn’t yet released results on more widely recognized benchmarks such as WebVoyager, leaving some room for skepticism about real-world reliability.
Inside Amazon’s AGI Lab
Nova Act is the first public output from Amazon’s newly minted AGI lab in San Francisco, co-led by AI heavyweights David Luan (formerly of OpenAI and Adept) and Pieter Abbeel (formerly of Covariant). Their vision for AGI is pragmatic: build systems that can help with anything a human can do on a computer. In that light, Nova Act’s focus on short, goal-oriented tasks is not a distraction — it’s a foundational building block.
Luan emphasizes that the SDK allows developers to define when a task should require human intervention, a nod to the current limitations of fully autonomous agents and a bet on reliable semi-autonomy as the stepping stone toward AGI.
What’s Next?
The release of Nova Act is a strategic milestone for Amazon’s AI ambitions. Alexa+ will be under intense scrutiny when it finally rolls out, and Nova Act is its technical heart. If the agent delivers on its promises — reliability, speed, accuracy — it could signal a turning point in how AI integrates into everyday digital workflows.
But the challenges are real. Existing AI agents from top labs still struggle with consistency, context, and staying on task. Whether Amazon has solved these issues or merely delayed them remains to be seen.
For now, Nova Act is a promising leap — not just for Amazon, but for anyone watching the future of AI agents unfold.
