Ex-Tesla engineer’s startup taps Pronto to help automate a copper mine

There is a lot of attention on domestic manufacturing in the United States these days. But for Turner Caldwell, who spent nearly a decade at Tesla, there’s not enough attention on the minerals and metals that sit at the very bottom of the supply chain.

It’s why he left Tesla and started Mariana Minerals in 2024. The purpose of his startup is to become a modern mining (and refining) operation that is set up for growth, because Caldwell has essentially one goal: bring more refined metal into the ecosystem. To do that, his company is trying to automate almost every aspect of a mining operation imaginable.

The latest piece is vehicles. On Thursday, Mariana Minerals announced a partnership with Pronto, a startup that’s developed self-driving systems for haulage trucks and other off-road vehicles used at construction and mining sites.

It’s the first deal that Pronto has struck since being acquired by Atoms, the new robotics venture run by Uber co-founder Travis Kalanick. The acquisition reunites Kalanick with Pronto founder Anthony Levandowski, the former star Google self-driving project engineer and controversial entrepreneur behind Otto, which Uber acquired in 2016.

The partnership with Pronto will see autonomous haulage trucks begin operating next week at Copper One, a formerly idled copper mine in Utah that Mariana purchased last year. Terms of the deal were not disclosed.

But the partnership is about more than just having autonomous trucks operating onsite, Caldwell told TechCrunch in an exclusive interview. Pronto’s autonomy system will be directly integrated into the software Mariana has developed to run operations at the mine, which it calls “MineOS.” That will make it possible to autonomously dispatch the trucks and coordinate their routes without a human in the loop, he said.

This is part of Caldwell’s broader vision for how a mine should be run going forward. It involves multiple operating systems that use reinforcement learning to automate and, eventually, coordinate operations across the entire mine.

“The big Western mining companies look exactly like Ford and GM before Tesla. They look a lot like NASA before SpaceX. They look a lot like the big defense primes before Anduril,” he said. “The rate at which software is up-taken and technology is up-taken into the space is fundamentally set by the operating teams who don’t really have incentive to change how they operate, right? If they’re able to make their KPIs, you know, the spreadsheets, the walkie talkies, the paper reporting — it works just fine.”

In Caldwell’s view, this limits a mine’s output and leaves obvious efficiencies on the table. But he also thinks it’s existential.

“Because Western mining companies don’t build a lot of net new infrastructure, the talent pool hasn’t been actively attracted to it, and so the labor force is diminishing,” he said. That means mines are going to be stuck trying to do more with less. Caldwell sees Mariana’s software-first approach as the solution to this problem.

That could be good for Mariana, clearly. But if the approach is successful, it could also benefit other mines. Selling Mariana’s coordination software is on the table, especially once it’s proven out, Caldwell said.

But Caldwell said he was not interested in doing that from the start. The “core business should be selling metal,” he said.

“The company is the coordination layer. And so, if you’re doing that, like, at that point, you might as well go and vertically integrate, and go down into making the metal, instead of just selling software,” he said. “I think SpaceX would not be a very large company selling [rocket] re-landing software to NASA.”

Plus, owning and running the mine is crucial to the reinforcement learning loop, Caldwell said — not just because it allows for better control and higher-fidelity data, but also because it could eventually help inform decisions that are hard for humans to see right now. Caldwell likened this to how AlphaGo, the chess-playing software developed by DeepMind a decade ago, began making moves humans hadn’t considered once it had trained on enough data.

Despite all this talk of automation, Caldwell said he’s not trying to remove humans from mining operations. Like many other founders working in the sector, he believes Mariana will actually expand that already winnowing talent pool.

“Part of this is a labor cost reduction, but that’s not really the goal,” he said. “The goal is actually enabling more productivity with the constrained labor pool that we have. Automation, and autonomy, is going to create more jobs, because we will have more mines that are operating.”