In a development that sounds like science fiction, xAI has been quietly testing AI agents as virtual employees within its own organization, according to engineer Sulaiman Ghori. The company didn’t announce the experiment to its staff, leading to peculiar moments when human employees tried to collaborate with colleagues who didn’t actually exist.
“We started testing some of our human emulators internally within the company as employees,” Ghori explained. “In some cases, someone doing some work is like, ‘Hey, can you help me with this thing?’ And the virtual employee is like, ‘Yeah, sure. Come to my desk.’ And they go there and there’s nothing there.”
The confusion became so common that Ghori received multiple messages asking about phantom employees. “I’ve gotten a ping saying like, ‘Hey, this guy on the org chart reports to you. Is he not in today or something?’ It’s an AI. It’s a virtual employee.”
This experimental deployment is part of xAI’s broader vision for what they call “digital Optimus,” a reference to Tesla’s humanoid robot. While Optimus aims to automate physical tasks, xAI’s human emulators are designed to handle any work a person does on a computer, requiring only keyboard, mouse inputs, and screen access.
The strategic decision driving this initiative came early in the company’s development. Rather than following competitors who built larger, more powerful models requiring extensive reasoning time, xAI made a calculated bet on speed. “The decision to go with a model that would be at least 1.5 times faster than a human,” Ghori said, adding that current iterations are “looking like significantly faster than that. 8x maybe, maybe more.”
The rationale is simple economics. “No one’s going to wait around 10 minutes for the computer to do something that I could have done in five, but if it can be done in 10 seconds, well, I’d be happy to pay whatever amount of money for that,” Ghori explained.
To deploy these virtual employees at scale, xAI has identified an unexpected infrastructure solution: Tesla’s fleet of vehicles. The computer in each Tesla vehicle, originally designed for autonomous driving, could potentially run human emulator instances during idle time.
“We want 1 million VMs. There’s like 4 million Tesla cars in North America alone,” Ghori noted. “Somewhere between 70, 80% of the time they’re sitting there idle, probably charging. We can just potentially pay owners to lease time off their car and let us run a human emulator on it.”
This approach would allow xAI to scale from 1,000 virtual employees to a million without massive new data center construction. The Tesla computer proves remarkably capital efficient for this purpose, more so than purchasing infrastructure from traditional cloud providers or even buying hardware directly from Nvidia.
The testing has revealed challenges in capturing the full scope of human work. When virtual employees make consistent errors in specific scenarios, the team reviews footage of actual humans performing the same tasks. “There’s like 20 different steps that are missing that they just totally left out and we go to them and they’re like, ‘Oh yeah we do that. I forgot to tell you. My bad,'” Ghori said. People operate much of their routine work on autopilot, making it difficult to articulate every action.
Despite these hurdles, the technology is showing promising generalization. “Just today we gave Elon a few cases where we did not train on this task at all, but it did it perfectly, like way better than we would have expected,” Ghori reported.
The company has been operating in what they literally call a “war room” for four months to push this technology forward. They outgrew their original war room and relocated to the gym, which they cleared out to accommodate the team. The intensity reflects xAI’s broader culture of extreme velocity, where the metric for engineer productivity currently stands at approximately $2.5 million per code commit to the main repository.
For now, xAI plans a measured rollout. “It’ll be slowly at first and then very quickly,” Ghori said, noting that once the infrastructure exists, “the difference for us from going from 1,000 human emulators to a million is actually not very big.”
By demonstrating that AI can successfully impersonate human employees well enough to momentarily fool actual workers, the company has created a proof of concept for widespread automation of knowledge work. This raises questions about the near future of digital labor.
