Amazon has announced the establishment of a new laboratory, named the Amazon AGI SF Lab, situated in San Francisco. This initiative aims to develop foundational capabilities for utilising artificial intelligence (AI) in both digital and physical realms. The new lab will be an extension of the company's broader Artificial General Intelligence (AGI) team, which recently unveiled Amazon Nova, a suite of cutting-edge foundation models (FMs) designed to advance AI technology.
The AGI SF Lab's mandate is to empower researchers and engineers within Amazon to achieve significant breakthroughs with improved speed and focus. The lab’s primary research objectives include enabling AI systems to execute real-world actions, adapt through human feedback, self-correct, and understand the goals set for them. Amazon expressed particular enthusiasm for integrating large language models (LLMs) with reinforcement learning (RL) to tackle complex reasoning and planning challenges, alongside developing learned world models and designing adaptable AI agents for physical environments.
The announcement follows an extensive series of AI-related news shared at Amazon's re:Invent 2025 event, where the company disclosed plans to double its investment in AI startup Anthropic to $8 billion, introduce the new Trainium2 chip engineered for the substantial computational requirements of AI, and launch six foundational LLMs under the Nova branding.
In a strategic move, Amazon revealed that it hired a majority of the team from Adept, a startup focused on automating software processes, in June 2024. This acquisition, reported to be around 66% of Adept’s workforce, underscores Amazon's efforts to fast-track its AI development and workflow automation capabilities. Following this integration, Amazon stated that it intends to leverage Adept's technology to enhance its roadmap for automating software workflows. The expertise garnered from the Adept team will be pivotal as they contribute to the new lab.
David Luan, the co-founder of Adept, has been appointed to lead the Amazon AGI SF Lab, with Pieter Abbeel, who co-founded Covariant, joining him to collaborate on the project. Abbeel’s involvement comes after Amazon also acquired a significant portion of Covariant’s team in August 2024, where they signed a non-exclusive license to utilise Covariant’s robotic foundation models. Covariant has developed what it refers to as the “universal AI platform,” known as the Covariant Brain, which has been pre-trained on an extensive dataset of robotic operations within warehouses globally.
The AGI SF Lab is poised to expand its team further, seeking to recruit several dozen individuals not only from the field of AI but also from diverse disciplines such as physics, mathematics, and quantitative finance, welcoming candidates of various experience levels. Those interested in joining the lab are encouraged to inquire via the provided Amazon email address.
Source: Noah Wire Services