Every now and then, a Silicon Valley startup launches with such an “absurdly” described mission that it’s difficult to discern if the startup is for real or just satire.
Such is the case with Mechanize, a startup whose founder – and the non-profit AI research organization he founded called Epoch – is being skewered on X after he announced it.
Complaints encompass both the startup’s mission, and the implication that it sullies the reputation of his well-respected research institute. (A director at the research institute even posted on X, “Yay just what I wanted for my bday: a comms crisis.”)
Mechanize was launched on Thursday via a post on X by its founder, famed AI researcher Tamay Besiroglu. The startup’s goal, Besiroglu wrote, is “the full automation of all work” and “the full automation of the economy.”
Does that mean Mechanize is working to replace every human worker with an AI agent bot? Essentially, yes. The startup wants to provide the data, evaluations, and digital environments to make worker automation of any job possible.
Besiroglu even calculated Mechanize’s total addressable market by aggregating all the wages humans are currently paid. “The market potential here is absurdly large: workers in the US are paid around $18 trillion per year in aggregate. For the entire world, the number is over three times greater, around $60 trillion per year,” he wrote.
Besiroglu did, however, clarify to News that “our immediate focus is indeed on white-collar work” rather than manual labor jobs that would require robotics.
The response to the startup was often brutal. As X user Anthony Aguirre replied, “Huge respect for the founders’ work at Epoch, but sad to see this. The automation of most human labor is indeed a giant prize for companies, which is why many of the biggest companies on Earth are already pursuing it. I think it will be a huge loss for most humans.”
But the controversial part isn’t just this startup’s mission. Besiroglu’s AI research institute, Epoch, analyzes the economic impact of AI and produces benchmarks for AI performance. It was believed to be an impartial way to check performance claims of the SATA frontier model makers and others.
This isn’t the first time Epoch has waded into controversy. In December, Epoch revealed that OpenAI supported the creation of one of its AI benchmarks, which the ChatGPT-maker then used to unveil its new o3 model. Social media users felt Epoch should have been more up-front about the relationship.
When Besiroglu announced Mechanize, X user Oliver Habryka replied, “Alas, this seems like approximate confirmation that Epoch research was directly feeding into frontier capability work, though I had hope that it wouldn’t literally come from you.”
Besiroglu says Mechanize is backed by a who’s who: Nat Friedman and Daniel Gross, Patrick Collison, Dwarkesh Patel, Jeff Dean, Sholto Douglas, and Marcus Abramovitch. Friedman, Gross, and Dean did not return News’s request for comment.
Marcus Abramovitch confirmed that he invested. Abramovitch is a managing Partner at crypto hedge fund AltX, and self-described “effective altruist.”
He told News he invested because, “The team is exceptional across many dimensions and have thought deeper on AI than anyone I know.”
Good for humans, too?
Still, Besiroglu argues to the naysayers that having agents do all the work will actually enrich humans, not impoverish them, through “explosive economic growth.” He points to a paper he published on the topic.
“Completely automating labor could generate vast abundance, much higher standards of living, and new goods and services that we can’t even imagine today,” he told News.
This might be true for whoever owns the agents. That is, if employers pay for them instead of developing them in-house (presumably, by other agents?).
On the other hand, this optimistic outlook overlooks a basic fact: if humans don’t have jobs, they won’t have the income to purchase all the things the AI agents are producing.
Still, Besiroglu says that human wages in such an AI-automated world should actually increase because such workers are “more valuable in complementary roles that AI cannot perform.”
But remember, the goal is for the agents to do all the work. When asked about that, he explained, “Even in scenarios where wages might decrease, economic well-being isn’t solely determined by wages. People typically receive income from other sources—such as rents, dividends, and government welfare.”
So perhaps we all make our living from stocks or real estate. Failing that, there’s always welfare – if the AI agents are paying taxes.
Even though Besiroglu vision and mission are clearly extreme, the technical issue he’s looking to solve is legit. If each human worker has a personal crew of agents which helps them produce more work, economic abundance could follow. And Besiroglu is unquestionably right on at least one thing: a year into the age of AI agents, they don’t work very well.
He notes that they are unreliable, don’t retain information, struggle to independently complete tasks as asked, “and can’t execute long-term plans without going off the rails.”
However, he’s hardly alone in working on fixes. Giant companies like Salesforce and Microsoft are building agentic platforms. OpenAI is, too. And agent startups abound: from tasks specialists (outbound sales, financial analysis); to those working on training data. Others are working on agent pricing economics.
In the meantime, Besiroglu wants you to know: Mechanize is hiring.
