Tesla advances AI agent project as XAI's Macrohard effort stalls

Following leadership changes and suspension of a 600-contractor data initiative, Macrohard—the ambitious AI agent venture by XAI—has been put on hold, insiders shared with Business Insider.

Meanwhile, Tesla, another Elon Musk enterprise, is boosting its AI agent initiative, "Digital Optimus," as per employees. 

Musk had humourously introduced Macrohard in August as a reference to "Microsoft," suggesting that since companies like Microsoft don't manufacture hardware, simulating them entirely with AI should be feasible. This effort aims to create an AI office worker.

Macrohard has been regarded as one of the key projects of xAI, alongside Grok Code and Grok Imagine.

Since its inception, Macrohard has undergone management reshuffles and has faced challenges scaling up, according to insiders.

In February, two Macrohard leaders departed, as informed by those aware of the exits.

During a subsequent company-wide meeting, Musk declared that xAI cofounder Toby Pohlen would take charge of the project. Sixteen days later, Pohlen announced his departure.

He faced pressure from Musk about the project's progress, despite Musk previously airing his dissatisfaction with the team's pace before Pohlen took over.

Nearly 24 xAI engineers listed working on Macrohard on X or LinkedIn. Most have either exited the company or transitioned to other teams recently, with over a dozen leaving in the past month alone.

Recently, some staff were informed that parts of Macrohard work, along with some of its computational resources, would move to Tesla's Autopilot team, insiders stated.

Tesla is also working internally on an AI agent referred to as "Digital Optimus," inspired by the robotic humanoid it started in 2021. The digital counterpart aims to be an AI that operates tasks on computers.

In February, Tesla advertised a position for an AI engineer for a computer-use agent—a similar focus as xAI's Macrohard. 

This agent is expected to autonomously handle software interactions, code creation, and immediate decision-making.

Unlike conventional systems that assess static screenshots, the team is exploring real-time control mechanisms. 

This approach allows the AI to interpret and act upon a continuous information flow, rather than evaluating a static image in steps.

This method resembles how its Full Self-Driving system processes live video feeds. Musk has consistently impressed upon xAI staff that their endeavor should imitate Tesla's Full Self-Driving operations, which utilise video data to enable real-time AI navigation of environments.