cupure logo
trumpdaytariffamazonelontradetariffsprime2025pay

Mechanize's CEO wants to use AI to automate every job — including his own

Tamay Besiroglu is the cofounder and CEO of the San Francisco startup Mechanize, which aims to automate all jobs.Courtesy Tamay BesirogluCofounder and CEO of Mechanize, Tamay Besiroglu, said he wants to automate all jobs using AI.Besiroglu said that the company was trying to automate software engineering first.Mechanize, backed by leaders at Google and Stripe, expects to see some AI integration take decades.Tamay Besiroglu's audacious goal is to use artificial intelligence to automate every job. So he's starting with what should be relatively easy: software engineering.Yet, even though the role involves spending a lot of time at a keyboard — and not, say, treating patients in an ER — whipping AI understudies into shape for desk work isn't as straightforward as it might seem."Software engineering is just a lot more than writing code," the cofounder and CEO of Mechanize told Business Insider.Among other things, engineers need to understand what their employer is doing and need to talk with product managers and customers, he said."Reality is surprisingly detailed, often," Besiroglu said. So, even if the company can have digital stand-ins do part of what people do, a full AI replacement for human software engineers could take a decade or more, he said.To make progress, Mechanize is attempting to teach AI through a process known as reinforcement learning, or RL, in which an AI agent is rewarded for completing a task correctly. The company builds these software "environments" to sell to AI labs, which use them to train their models.Besiroglu, 29, started Mechanize this year with Matthew Barnett and Ege Erdil. Backers include Jeff Dean, Google's chief scientist, and Stripe CEO Patrick Collison. The San Francisco startup drew some criticism when it announced its mission to hand over all jobs to AI.For Besiroglu, who previously created the AI nonprofit research organization Epoch AI, the online pushback wasn't surprising."We've been happy to generate discussion about this topic, because I think it's important," he said.Mechanize, which has fewer than 10 employees, recently moved into a new office in the South Beach section of San Francisco, across from Oracle Park, where the San Francisco Giants play. Besiroglu hopes the space will sustain the company until it reaches about 20 workers, perhaps in a year's time, he said.The following has been edited for length and clarity.Business Insider: How do you describe what your goal is to the average desk worker?Tamay Besiroglu: We would describe it the same to basically anyone. That's an important differentiator for us, relative to other companies. I personally have a very low tolerance for tailoring your message to different people. I want to not have 10 messages for 10 different people.We want to enable the development of AI that can do a lot more valuable work than it can do today, and in the fullness of time, can do every piece of valuable work that humans can do. We want to enable this full automation and thereby generate enormous amounts of value for society. That is the thing that we're focused on.We want to do that by building the piece that I think is most neglected in AI development today, which is scaling up the RL —reinforcement learning — environments.That's so you get scalable labor. Is that the idea?That's right. AI is scalable in a way that humans are not. You can copy and spin up a very large number of them, and you're not bottlenecked by something like human population growth.Our economy today is very much bottlenecked by labor. Capital is something that you could invest, and you can generate more however fast you want, basically. But with humans, population growth is like 2% or 3% per year, and that is not something that you could invest in — increasing the size of your population — very easily.AI presents the opportunity to do precisely that. You can reinvest your output into building more data centers and compute, and you can just run more AI workers. That grows your economy, which you can then reinvest again. That creates this feedback loop that results in this acceleration in overall growth.Why did you start with trying to automate software engineering?Software engineering makes sense for a couple of reasons. So one is, there is an existing technology and culture for writing very good tests — tests that check whether the piece of software is working as intended. Whether it's not breaking; whether it's not buggy; whether it's not surfacing security problems. And that is really helpful for doing RL.Because in RL, you want some reward signal. You want to be able to tell the model you did the task correctly versus incorrectly. Then you want to leverage that to reinforce the kind of patterns of behavior that resulted in it correctly performing the task.Secondly, there is this battle to win coding between Google, OpenAI, and Anthropic, and they are really keen on winning that battle. They are very willing to spend a ton of money on beating their competitors. So, that is really great for us, because it makes it much easier for us to get lucrative deals.There's been a lot of discussion about AI automating jobs. If this does occur, how will humanity get through this transition?This will be a gradual process. Some people in the industry expect this will happen very, very soon — this decade. I don't think that'll happen that soon. It'll be more of a challenge that takes decades for us to develop the technology, produce all the innovations, and build out our capital stock, and our compute to be able to do that. But I agree, there will be this transition, and I think the fact that it's more gradual and something that's drawn out over decades is definitely helpful.If I believed it was something that would take a couple of years, then I'd be much more nervous. Also, there are cases today where a large share of people's income in specific countries is not derived from labor income. There are some oil states that do this.There is Norway, as an example, where the sovereign wealth fund is worth a lot of money. I think we will look at examples like that for what our future might look like. Frankly, Norway is not that bad. I'm Dutch myself, and I have a bunch of Norwegian friends. They seem to do just fine. In many ways, they seem to have higher standards of living than people in the US.It's something that is indeed a large challenge, but I think the upside is just so enormous that the costs associated with this transition look fairly minor in comparison.One effect of that consumption could be increased externalities. If consumption is going up, for example, there are environmental costs. Have you thought about those types of things?These externalities are concerning. The environmental impacts are real, but I think there are problems where technology enables us to really make a large dent, and the development of AI is this enormous accelerant for technological progress. The development of AI, I am fairly confident, will improve our ability to minimize the harms to the environment if we want to deploy AI to do that.I'm more worried about an acceleration of economic growth and technological change, like collapsing history. If you have this acceleration, then maybe 100 years of history could happen in one decade, because things just move a lot faster.That might mean that the rates of things happening that you don't want to happen, like wars or various financial crises or pandemics, perhaps might accelerate because we're compressing time, effectively. So, that is maybe a worry, which is a little bit abstract, but I think that is the level of abstraction that is appropriate because it's hard to predict the details of precisely what will happen in 10, 20 years.You said you want to give everybody one answer about the implications for their jobs. But desk jobs you can do remotely feel more at risk than those of a nurse or a mechanic.Yeah, for that reason, we don't have a robotics division just yet. Robotics, I think, is going to take a little longer, but eventually we want to get there, too.I think robotics is just behind the types of tasks that you can do remotely. So, for that reason, I think it makes sense for us to focus on the things that you can do remotely. There's just a larger market for that currently. But I'm also pretty encouraged by some of the progress in humanoid robotics, and I think that's really valuable. I would love to have a humanoid robot that is able to clean up my office and my home, and I think that'll be really great.Do you have the sense that you'll eventually be developing yourself out of a job? What would you do in that case?It is a little ironic that we are hiring engineers to obsolete themselves, effectively, by building these environments that train models to be good at the types of things that they're good at.In the meantime, they are working on hard, important problems, and they're getting rewarded well for it. If they join us, then they get equity. And, so even once they've obsoleted themselves, they'd still be able to become rich off that, by just owning equity. That's the pitch I give our software engineers.Read the original article on Business Insider

Comments

Business News