Leaked email outlines Amazon’s big plans for ‘AGI,’ the AI tech all of Silicon Valley is chasing

• Amazon’s newly formed AGI group is undergoing significant restructuring.
• The team will now concentrate on six broad areas of AGI.
• The move comes on the heels of last week’s layoffs, which impacted the AGI team.

Only four months after its formation, Amazon’s new artificial general intelligence group is already restructuring.

According to an internal email obtained by Business Insider, the AGI team, led by Amazon’s head scientist and SVP Rohit Prasad, is splitting into six broad areas and offloading some Alexa-focused projects to the Devices team.

Prasad stated in an email on Monday that these changes will “help us further sharpen our focus on this mission and deliver faster for our customers.”

“I am very proud of what this organization has already delivered for Alexa and AGI,” Prasad said in a statement.

It’s worth noting that the tech world was watching OpenAI — the world’s most advanced AI company and a partner of Amazon’s main rival, Microsoft — unravel on Monday after its non-profit board fired CEO Sam Altman. Microsoft announced his hiring (though it is not yet finalized), while OpenAI employees threatened to quit en masse and join Microsoft if Altman was not reinstated.

A request for comment from Amazon was not returned.

The Amazon AGI team was not formed in response to this drama. It was founded four months ago in July with the goal of developing the “most ambitious” large language models, as previously reported by BI. Prasad was promoted at the time to report directly to Amazon CEO Andy Jassy.

While the definition of AGI isn’t universally agreed upon, it generally refers to creating machines that are as intelligent as, or more intelligent than, humans. This is distinct from generative AI, which is AI that can generate new creative works, such as fiction, emails, websites, or art, after studying patterns in previous works.

AGI is regarded as the next big goal for the AI industry, and it is being pursued by everyone from OpenAI and Microsoft to Google and numerous startups. It’s also the technology that frightens the AI doomsayers.

Prasad reiterated the AGI team’s overarching mission this week in an email: “build world-class general purpose intelligence services that benefit every Amazon business and humanity.”

Unlike OpenAI, this mission explicitly states “benefit every Amazon business” before “humanity.”

The change comes on the heels of last week’s job cuts within the AGI team, as previously reported by BI. The layoffs were part of a larger restructuring that primarily impacted Amazon’s Alexa team.

The team’s new structure includes six new focus areas for the AGI unit, revealing how Amazon intends to compete more directly with rivals Microsoft and Google in AI. It demonstrates Amazon’s intention to develop its own foundational models, as well as AGI conversational assistants and related infrastructure services.

Meanwhile, teams working on Alexa-focused features such as Alexa Ads, Alexa Personality, and Alexa Routines have been merged into the Devices team.

These are the six new teams formed within the AGI unit, as per the internal email:

• AGI Product: defines AGI services and collaborates with internal customer teams to ensure success from start to finish.• AGI Data Services: for our end customers, we build services for storing, accessing, and manually labeling data for training foundational and specialized Al models with world-class privacy and security.• AGI Foundational Models: creates the most capable foundational models (FMs) capable of completing any task in a safe and secure manner across any modality. Internal and external customers will use these models across all Amazon businesses.• AGI Sensory and Machine Learning Builder Services: creates specialized models and services like speech recognition, text-to-speech, and computer vision to power applications like Alexa, Amazon Search, Fire TV, and AWS Polly.• AGI Conversational Assistant Services: creates services to build general-purpose and specialized conversational assistants/agents by orchestrating 1000s of services and devices based on situational context to respond to a customer request or take proactive actions on their behalf.• AGI Information: constructs infrastructure for acquiring and organizing world knowledge and real-time data for training Al models, as well as experiences for accessing and acting on any information on any endpoint.