Nova Act Unveiled by AmazonOn Monday, Amazon made a significant announcement with the unveiling of NovaAct, a general-purpose AI agent. This innovative creation has the remarkableability to control web bRowsers and carry out simple tasks autonomously.Concurrently, Amazon launched the Nova Act SDK, which serves as a valuabletool for dEVElopers to construct agent prototypes using Nova Act.
Origin and Future Role of Nova ActDeveloped at Amazon’s newly established AGI lab in San Francisco, Nova Act isset to play a crucial role in the company’s upcoming Alexa+. This is agenerative AI-enhanced version of Amazon’s voice assistant. At present, thereleased version of Nova Act is presented as a “reseARCh preview,” anddevelopers can gain access to the Nova Act toolkit through nova.amazon.com.
Competitive LandscapeThis product clearly positions Amazon in the competition against OpenAI’sOperator and AnthroPic’s Claude. Many in the tech industry believe that AIagents capable of web navigation for users will greatly enhance thepracticality of current AI chatbots. Although Amazon isn’t the first in thisdomain, its extensive reach via Alexa+ could potentially give it an edge.
Functionality of Nova Act SDKAccording to Amazon, developers leveraging the Nova Act SDK can automate basictasks for users. These include activities like ordering food online or makingreservations. The toolkit also supports the integration of multiple functions,enabling the AI agent to browse web pages, fill out forms, or select dates ona calendar.
Performance ClaimsAmazon asserts that Nova Act outperformed its competitors in internal testing.In the ScreenSpot Web Text evaluation, Nova Act achieved a score of 94%,surpassing OpenAI’s CUA (88%) and Anthropic’s Claude 3.7 Sonnet (90%).However, it should be noted that Amazon did not benchmark Nova Act againstmore common agent evaluations such as WebVoyager.
The Minds Behind Nova ActNova Act is the first publicly released product from the AGI lab, co-led byformer OpenAI researchers David Luan and Pieter Abbeel. Before joining Amazonlast year to lead its AI agent initiatives, both had founded their own AIstartups. Luan founded Adept, and Abbeel co-founded Covariant.
Vision for AGI and Agent DesignLuan told TechCrunch that he views agents as a crucial step in creatingsuperintelligent AI systems. He defines AGI as “an AI system capable ofhelping accomplish everything humans do on computers.” The team designed theNova Act SDK to ensure reliable automation of short tasks and allow developersto precisely determine when human intervention is required in the workflow.
Challenges Facing AI AgentsA major hurdle for early AI agents is cross-domain reliability. In testing,existing systems typically suffer from slowness, struggle to operateindependently for extended durations, and are prone to making errors thathumans would not. The market will soon reveal whether Amazon has managed toovercome these limitations or if its agent encounters the same issues as itscompetitors.
