For months, the only real clues about what Jeff Bezos was building at a company called Prometheus were a name, a vague description of engineering and manufacturing, and the résumé of the physicist he had picked as his partner. That changed this year. The company has raised $12 billion at a valuation near $41 billion, and its founders have finally said what they are building: an “artificial general engineer,” software meant to design, test, and help manufacture complex physical products, from jet engines and spacecraft to computer chips, cars, and new drug compounds.

The partner is Vik Bajaj, a physicist and chemist who spent years at Google X, the research group known for its moonshot projects. Bajaj worked on Wing, the drone delivery effort, and on the self-driving car project that became Waymo, before co-founding the biotech incubator Foresite Labs. That background signals something about the bet being made here. Building a machine that can design a jet engine or a chip depends on physical science as much as software.

Behind that bet is what current systems cannot do. Claude and ChatGPT were trained on a huge amount of digitized human knowledge, mostly text, code, and images pulled from the internet. That training makes them strong at language, and increasingly strong at images and video. It does not make them strong at material science, thermodynamics, or the way a supply chain behaves, because none of that knowledge was part of the training data.

Bezos and Bajaj made this case together for the first time on CNBC in June. Bezos told the interviewer that large language models excel at what he called knowledge work, and that Prometheus has to build and reach entirely different data sets, ones that are hard to access, to train a system on physical engineering itself. From there, he made a much larger claim, framing the company as a bet on the oldest driver of wealth, invention.

He pointed to the plow and the steam engine as past inventions that made entire societies richer, and described the effort as an attempt to speed up that same cycle, from idea to working object, across physical engineering. Pressed on what faster invention would mean for workers, Bezos predicted labor scarcity rather than job losses, with productivity rising enough that some households would no longer need two incomes to get by. The picture he described looks closer to a mid-twentieth-century, single-earner household than the two-income norm most families live with today.

Whether that vision holds depends on how fast invention can move, and software already shows what fast looks like. Code ships quickly, gets tested by real users, and gets rewritten within days, and AI now automates much of that loop as well. Physical engineering does not move at that speed, especially somewhere like rocket engineering. A design can be simulated and refined quickly inside a computer, but building a working jet engine or spacecraft still requires physical prototypes, physical certification, and physical testing, steps no software tool can compress no matter how good the design process becomes. That difference sets the real pace at which an artificial general engineer could change a slow-moving industry like aerospace.

Prometheus has another plan beyond the software itself. According to the Financial Times, the company is separately preparing to raise tens of billions of dollars for a holding structure that would buy the kind of industrial businesses its own technology is meant to disrupt.

That structure would solve two problems at once. Manufacturers rarely share the operational detail that explains why a part failed or how a supply chain behaves, and that detail is exactly what a system trained on physical engineering needs. Owning the businesses is one way to reach it. Owning them also keeps the entire cycle in-house, research, testing, and the physical act of production, rather than building the tool and licensing it to an existing manufacturer such as Boeing or GE.

For all that ambition, both founders are careful to call this early. Bezos describes a team of about 150 people with, in his own words, many miles to go, and he would not set a timetable for when a product built this way reaches a factory floor. The compressed design loop being built still runs into the slower physical loop of prototypes, certification, and testing that no software can shortcut. The tens of billions being raised to buy industrial businesses will face the same test, one only the next few years of manufacturing can answer.

Sources

CNBC, CNBC Exclusive: Transcript: Prometheus Co-Founders and Co-CEOs Jeff Bezos and Vik Bajaj Speak With CNBC’s David Faber on ‘Squawk on the Street’ Today — cnbc.com
Engadget, Jeff Bezos will head a new engineering-focused AI startup called Project Prometheus — engadget.com
Tech Funding News, Bezos’ $30B AI lab eyes tens of billions more to remake manufacturing — techfundingnews.com