Left to right: David Luan, VP of autonomy and head of Amazon’s AGI SF Lab, and Pieter Abbeel, Amazon Scholar, robotics. | Source: Amazon
Amazon is forming a new lab in San Francisco, called the Amazon AGI SF Lab, to develop new foundational capabilities for enabling useful AI in the digital and physical worlds. The new lab’s work will build on that of the company’s broader AGI team, which recently introduced Amazon Nova, a new generation of state-of-the-art foundation models (FMs).
Amazon said the AGI SF Lab is designed to empower AI researchers and engineers to make major breakthroughs with speed and focus. It said the lab’s initial focus is on several key research bets that will enable AI to perform real-world actions, learn from human feedback, self-course-correct, and infer the lab’s goals. Amazon said it is particularly excited about the work in combining large language models (LLMs) with reinforcement learning (RL) to solve reasoning and planning, learned world models, and generalizing AI agents to physical environments.
The announcement of the AGI SF Lab comes on the heels of a barrage of AI-related announcements Amazon made last week during its re:Invent 2025 event. The news included that Amazon doubled its investment in AI startup Anthropic to $8 billion, launched the Trainium2 chip built for the heavy computing demands of AI, and launched six foundational LLMs under its Nova umbrella.
Amazon brings help from Adept, Covariant hires
Amazon announced in June 2024 it hired the majority of the team behind Adept, a startup that aimed to use AI to automate software processes. The company raised around $400 million for its technology, before Amazon hired around 66% of its staff, according to GeekWire.
When the Adept team joined Amazon, the company said it planned to license Adept’s technology to accelerate its roadmap for building AI to automate software workflows. Now, this team will be seeding the new lab. The AGI SF Lab will leverage Adept’s work in building AI agents that can handle complex workflows using the same tools we use as humans, like computers, web browsers, and code interpreters.
David Luan, co-founder of Adept, will lead the new lab. He’ll have some help from Pieter Abbeel, co-founder of Covariant, who will be working closely with Luan on the project, the company told TechCrunch. Abbeel came to Amazon in a similar way as the Adept team. In August 2024, Amazon hired Covariant’s founders and around 25% of its staff and signed a non-exclusive license to use the company’s robotic foundation models.
Covariant develops what it calls a “universal AI platform” called the Covariant Brain. Pre-trained on millions of picks from Covariant robots in warehouses around the world, the Covariant Brain enables robots to autonomously pick many SKUs.
The lab is looking hire a few dozen people — not just AI experts who have trained state-of-the-art models but also candidates from other disciplines who will bring fresh thinking to the field, such as physics, math, or quantitative finance, regardless of experience level. Interested applicants can reach out via [email protected].
Tell Us What You Think!