Topics
late
AI
Amazon
Image Credits:Thos Robinson/Getty Images for The New York Times(opens in a new window)
Apps
Biotech & Health
mood
Image Credits:Thos Robinson/Getty Images for The New York Times(opens in a new window)
Cloud Computing
Commerce
Crypto
Enterprise
EVs
Fintech
fund raise
Gadgets
stake
Government & Policy
computer hardware
layoff
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
security department
societal
Space
Startups
TikTok
expatriation
speculation
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
get hold of Us
Amazon on Monday unveiled Nova Act , a worldwide - purpose AI agent that can take control of a connection web web browser and severally perform some mere actions . Alongside the fresh agentic AI theoretical account , Amazon is releasing the Nova Act SDK , a toolkit that allows developers to build agent prototype with Nova Act .
Nova Act , developed byAmazon ’s late opened San Francisco - based AGI laboratory , will also power primal features of the company’supcoming Alexa+ ascent , a generative AI - enhance reading of Amazon ’s popular vocalisation help . The reading of Nova Act available starting today is a little less polished , however . Amazon is calling it a research preview .
Developers can reach the Nova Act toolkit on a Modern website , nova.amazon.com , which also serve up as a vitrine for Amazon ’s various Nova foundation model .
Nova Act is Amazon ’s attempt to take onOpenAI ’s OperatorandAnthropic ’s Computer Usewith world-wide - purpose AI federal agent engineering science of its own . Several leading technical school companies believe AI agents that can sail the web for users will make today ’s AI chatbots significantly more useful .
Amazon may not be the first to develop this kind of agentic technology , but via Alexa+,it may have the panoptic reach .
Amazon tell developers build up with the Nova Act SDK should be able to automate introductory action mechanism on behalf of user , such as order salads from Sweetgreen or making dinner party reservations . With the Nova Act toolkit , developer can displume together tools that permit an AI agentive role to navigate connection pages , fill out forms , or pluck dates on a calendar .
Amazon claims that Nova Act outperforms agent from OpenAI and Anthropic on several of the ship’s company ’s internal run . For example , on ScreenSpot web Text , which measures how an AI agent interact with text edition on a cover , Nova Act score 94 % , outperforming OpenAI ’s CUA ( which scored 88 % ) and Anthropic ’s Claude 3.7 Sonnet ( 90 % ) .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
However , Amazon did n’t benchmark Nova Act using more vulgar agent evaluations , such as WebVoyager .
Nova Act is the first public product to issue from Amazon ’s aforementioned AGI research laboratory , an initiativeco - led by former OpenAI researchers David Luan and Pieter Abbeel . Both previously founded startups of their own — Luan startedAdept , while Abbeel co - foundedCovariant — before Amazon charter them aside last yr to spearhead its AI factor effort .
While it may seem strange for an AGI science lab to be building AI agents that can place salad , Luan assure TechCrunch that he fancy agents as a fundamental step toward creating superintelligent AI system . Luan defines AGI as “ an AI system that can help you do anything a man does on a computer . ”
Luan says his team designed the Nova Act SDK to dependably automate short , elementary tasks , and give developer tools to incisively delineate when they want a man to interfere in an agentic workflow . He hopes it will allow developers to make more true agentic applications , albeit not necessarily fully independent ones .
Amazon is put out its first generalist AI broker in a crowded space , but it ’s a of the essence applied science that the fellowship has a bunch depend on on . former exam of Nova Act could provide a glance into some of the capabilities of the long - delayed Alexa+ , a make - or - break moment for Amazon ’s AI efforts .
Amajor trouble with early AI agentsfrom OpenAI , Google , and Anthropic is their dependableness across different domain . In TechCrunch ’s tests , the system are slow , shin to run independently for very prospicient , and are prone to mistakes a human would n’t make . It wo n’t be long until we see whether Amazon has break up the code — or whether its agent suffer from the same flaws chivy challenger .
