Amazon has entered the AI agent race with Nova Act, a new model capable of controlling web browsers to perform tasks like booking reservations or ordering food. The company also released the Nova Act SDK, allowing developers to build custom agent prototypes using the technology.
Developed by Amazon’s San Francisco AGI lab—led by ex-OpenAI researchers David Luan and Pieter Abbeel—Nova Act will also power features in the upcoming Alexa+ upgrade. While currently labeled a "research preview," the model already outperforms competitors like OpenAI’s CUA and Anthropic’s Claude 3.7 Sonnet in Amazon’s internal tests, scoring 94% on screen interaction benchmarks.
Unlike chatbots, Nova Act is designed for action—navigating websites, filling forms, and handling simple workflows while allowing developers to set human intervention points. Luan sees such agents as stepping stones toward more advanced AI, defining AGI as systems that "can help you do anything a human does on a computer."
Amazon’s move comes as tech giants bet on AI agents to make assistants more useful. However, early agents from rivals have struggled with reliability—often being slow, error-prone, or unable to operate autonomously for long. Nova Act’s real-world performance, particularly in Alexa+, could determine whether Amazon gains an edge in this critical AI battleground.