Meet Nova Act, Amazon’s latest AI agent that can control web browsers and perform basic tasks on the internet, like ordering food or booking a restaurant. This new system is now available for developers to download and experiment with, marking Amazon’s entry into the AI agent competition against OpenAI and Claude.
Nova Act was developed by Amazon’s new AGI research center in San Francisco, led by former OpenAI team members David Luan and Pieter Abbeel. The system will eventually be integrated into the upcoming ‘Alexa+’ upgrade, but for now, it’s still in its experimental phase.
To access Nova Act, developers can visit the nova.amazon.com website, which collects various basic models from Amazon’s Nova family. This system is designed to help AI understand web pages, fill out forms, and select dates on calendars automatically. Amazon hopes this will enable developers to create more efficient AI applications.
Amazon claims that Nova Act outperforms its competitors, OpenAI Operator and Anthropic’s Claude 3.7 Sonnet, in internal tests, particularly in the ScreenSpot Web Text test, where Nova Act scored 94%, surpassing OpenAI (88%) and Anthropic (90%). However, the company has yet to release results from the popular WebVoyager standard, which will be crucial in determining its real-world performance.
According to David Luan, co-founder of Amazon’s AGI research center, current AI agents have limited capabilities, but they represent a crucial step towards the ultimate goal of achieving AGI, or artificial general intelligence, which can perform any task like humans. Luan emphasizes that the primary objective is to create AI that makes human life easier, even if it’s not fully automated.
