Amazon AGI Lab

Amazon AGI Lab developed AI agents that could act through computers and web browsers to complete business workflows. Its Nova Act service lets developers build browser automation with natural-language instructions and Python, covering software testing, form filling, data extraction, and bookings. Developers can prototype in a playground, debug in an IDE, and deploy and manage agents on AWS. The San Francisco team’s July 2026 closure came amid broader Amazon AGI layoffs; Nova Act remains available through AWS.

Amazon established the Amazon AGI SF Lab in 2024, with David Luan as its founding head and Amazon robotics scholar Pieter Abbeel involved in its launch. The initial team included employees hired from Adept. Luan departed in February 2026; Bryan Silverthorn subsequently led the lab. Its technical approach combined reinforcement learning with computer use: Nova Act trains its model, SDK, orchestrator, and browser controllers together in simulated business applications. These environments let agents practice multistep tasks through trial and error before deployment.

Nova Act became generally available as an AWS service in 2025, bringing the lab’s research into enterprise automation. Hertz used it to test rental reservation workflows, while 1Password used it for sign-in flows across hundreds of websites. In the 2025 launch announcement, Sola reported using Nova Act for business-critical processes running hundreds of thousands of times per month, illustrating its role in repetitive operational work.