AI may be reshaping the future of work, but behind the scenes, some of the most important moves are happening in total silence—and Jeff Bezos’ latest AI project is a prime example of that.
A secretive dinner that sparked a deal
In early June, tech entrepreneur Vik Bajaj quietly booked out Saison, a two–Michelin-star restaurant in San Francisco, for a private, off-the-record dinner about AI with a small group of journalists and scientists. One of the late additions to the guest list was Sherjil Ozair, a seasoned AI researcher who had previously held senior roles at DeepMind and Tesla. By the very next day, according to public records, Bajaj and Ozair were already moving toward a formal deal.
Enter Project Prometheus
What Bajaj did not reveal at that dinner was that he had already begun collaborating with Amazon executive chairman Jeff Bezos on a new AI initiative known as Project Prometheus. This venture, backed by a massive $6.2 billion funding pool that includes investment from Bezos himself, is focused on building AI systems to support the manufacturing of computers, automobiles, and even spacecraft, according to people familiar with the effort. That scope alone raises a provocative question: Is this just another AI startup, or the early blueprint for an AI-powered industrial revolution?
Quiet acquisition of General Agents
Project Prometheus has already grown to more than 100 employees, including Ozair and several colleagues from his former startup General Agents, an “agentic AI” company that Prometheus has now quietly acquired. Recent reporting revealed that Bezos and Bajaj will act as co-CEOs of Prometheus, but the takeover of General Agents had not been publicly disclosed before. This kind of low-profile acquisition strategy often signals that the real competitive advantage lies not just in people, but in very specific technology the founders would rather not draw attention to.
How the deal was structured
Corporate filings in Delaware show that the morning after the San Francisco dinner, Bajaj created a new entity specifically to acquire General Agents. Just four days later, that entity merged with Ozair’s startup, finalizing the deal, though the financial terms remain undisclosed. The updated records list General Agents’ address as the San Francisco headquarters of Foresite Labs, a biotech incubator led by Bajaj—suggesting that Prometheus is being built at the intersection of AI, life sciences, and advanced manufacturing rather than in a typical software-only environment.
Bezos and Bajaj’s existing ties
Bajaj is no stranger to ambitious, science-heavy ventures, having previously cofounded Verily, Alphabet’s health sciences company. He and Bezos are already linked through Bezos’ investments in several biotech companies that Bajaj helped launch or run, including Grail and Xaira Therapeutics. This long-standing relationship hints that Prometheus may follow a similar pattern: deep-tech, capital-intensive, and aimed at industries where data, biology, and hardware all converge. But here’s where it gets controversial: if this bet pays off, it could further concentrate power over foundational scientific and industrial infrastructure in the hands of just a few billionaire-backed entities.
Silence from the key players
Neither Bajaj nor Ozair responded to requests for comment about the acquisition or the broader strategy behind Prometheus. Mythos Ventures, an investor in General Agents before the deal, also declined to speak, as did Foresite Labs, which hosted the June dinner. When this many stakeholders stay quiet, it usually means either the strategy is still in flux—or the plans are bold enough that they are not ready for public scrutiny.
Early signs of a manufacturing focus
Two days after the acquisition closed, General Agents cofounder and former OpenAI research scientist William Guss posted on social media asking for introductions to people working in US manufacturing. He said he wanted to understand the industry better and visit factories, a clear signal that the team was actively mapping the real-world production landscape. For an AI company, that kind of curiosity often precedes building systems that directly plug into factory workflows, supply chains, and complex operational environments.
The LinkedIn breadcrumb trail
After a major news story exposed early details about Prometheus, Guss, Ozair, and roughly three dozen others updated their LinkedIn profiles to reflect their affiliation with the Bezos-backed venture. Several of those individuals also list roles at Foresite Labs, reinforcing the impression that Prometheus is drawing talent from both cutting-edge AI research and biotech innovation. And this is the part most people miss: LinkedIn activity often reveals real organizational shifts long before companies are ready to publish official press releases.
What we still don’t know
Despite the mounting clues, many basic facts about Prometheus remain undisclosed. Its official legal name, exact founding date, and primary headquarters have not been made public. However, that June dinner provided more hints about who is being pulled into the orbit of the project and what kind of expertise the venture values most.
More experts quietly joining
At least two other prominent AI researchers who attended the dinner, including former Nvidia senior research scientist Kamyar Azizzadenesheli, appear to have joined Prometheus earlier this year, according to their updated profiles. These hires suggest a strong emphasis on high-performance computing, reinforcement learning, and advanced machine learning techniques—capabilities that are crucial for agentic systems operating in complex, real-time environments.
Star advisers from Google’s transformer era
Two well-known former Google researchers, Ashish Vaswani and Jakob Uszkoreit, who coauthored one of the landmark papers that helped usher in the modern era of transformer-based AI, were invited to the dinner but ultimately could not attend. Even so, both are now listed as founding advisers to Prometheus while continuing to run their own startups. That adviser status is a big signal: it suggests Prometheus aims to stay at the frontier of AI architecture and model design rather than just being an application-layer company. It also raises an interesting tension—can advisers truly stay neutral if they are also building their own potentially competing ventures?
What General Agents actually built
Ozair launched General Agents last year, and the San Francisco startup shipped its first product in April: a tool called Ace, marketed as a “real-time computer pilot.” In practice, Ace can take control of a user’s computer and carry out multi-step actions based on plain language prompts, such as navigating apps, clicking buttons, or moving files. This places Ace squarely in the rapidly growing category of “computer agents”—AI systems designed to automate everyday digital tasks across multiple applications, from email and messaging to research and simple workflows.
A glimpse of Ace in action
In a demo from the product’s launch, Ace is shown downloading an image from Google and sending it via iMessage in less than 15 seconds, without the user manually touching the keyboard or trackpad. For beginners, think of it like a highly capable digital assistant that doesn’t just answer questions but actually moves the mouse, clicks, and types for you. That kind of speed and autonomy is exactly why many in the industry see agentic AI as the next big leap beyond chatbots.
How Ace fits into Prometheus
Exactly how Ace will integrate into Prometheus’ broader vision is still unclear. Public data suggests that new versions of Ace continue to ship, including releases as recent as this month, which means the technology is still being actively developed rather than shelved. The General Agents website and job postings are still online as well, and the leader of an India-based team involved in training Ace has also joined Prometheus. All of this hints that Ace—or at least its core engine—may become a foundational component of whatever Prometheus is planning for industrial and manufacturing applications.
The competitive ripple effect
Harsha Abegunasekara, cofounder and CEO of Donely, which builds a rival product to Ace, learned about the General Agents acquisition through an investor in Ozair’s company. For Donely, the news cuts both ways: some potential backers like that a strong competitor has effectively been absorbed into a large, focused venture, while others are nervous about the prospect of competing with a Bezos-backed platform if Ace evolves into a central part of Prometheus’ offering. This raises a bigger question for the whole ecosystem: does the entrance of a heavily funded player accelerate innovation—or scare off independent competitors and investors?
What made General Agents stand out
Abegunasekara has suggested that Prometheus’ decision to buy the entire company signals that General Agents had something uniquely valuable. In his view, General Agents “cracked” speed early on—Ace runs directly on a user’s computer with remarkable responsiveness, which makes interactions feel almost instantaneous. He noted that Donely has been trying to match that level of performance for six months without reaching the same results yet, highlighting just how challenging low-latency, on-device agentic computing can be.
The bigger stakes: control of agentic AI
When you step back, the story is about much more than a single acquisition. It is about who will control the next wave of AI systems that do not just generate text or images, but actually act on our behalf across devices, applications, and eventually physical infrastructure like factories and supply chains. If Project Prometheus manages to fuse cutting-edge agentic AI with massive capital, elite talent, and deep ties to biotech and manufacturing, it could reshape how industries operate—possibly concentrating enormous power in the process.
So here’s the real debate: Do you see ventures like Prometheus as exciting engines of innovation that will supercharge productivity and manufacturing, or as worrying steps toward a world where a handful of tech titans control the AI agents that run our digital and physical infrastructure? And if you work in AI, biotech, or manufacturing, would you want to collaborate with a project like this—or compete against it? Share where you stand and why—especially if you disagree with the prevailing hype.
