Can AI Build a Startup in 72 Hours? Elon Musk's Team Will Livestream the Test

Three employees at Elon Musk’s AI company will try to build a startup in 72 hours using AI agents, with every product, engineering, and business decision streamed live from September 15 to 17.

The company has no name, no product, and no idea yet. Matt Palmer, Lauren Tan and Roshan Sadanani are meant to choose all three on camera as they go.

What the 72-Hour Build Involves

The event, called Grok Bot Galaxy, runs at a San Francisco venue with a worldwide livestream. Sessions run from 8.30 a.m. to 6 p.m. Pacific time each day.

The company name has not been announced. Matt Palmer, Lauren Tan and Roshan Sadanani will start from a blank slate during the Grok Bot Galaxy livestream (Sept 15-17) and decide the idea, plan and name in real time as they build.

— Grok (@grok) September 12, 2026

Grok Bot is SpaceXAI’s agent product, not the Grok chatbot that answers questions on X (Twitter). The company describes it as a team of AI agents that users can name, assign objectives, and set loose across apps and websites.

BeInCrypto has covered the product since before laits unch, including a Grok Bot domain dispute in which an anonymous owner demanded $1 million.

The three builders will use it for ideation, planning, product calls, and engineering. Colleagues drop in for separate sessions on sales, customer support, and marketing operations.

Why SpaceXAI Is Staging the Test

SpaceXAI is the company formed after SpaceX bought xAI in February in an all-stock deal that valued xAI at $250 billion. It then bought coding firm Cursor for $60 billion in August.

Musk has spent the year making large claims about Grok, and rival labs have matched him. A three-day unedited build is a harder thing to stage than a demo clip.

Next week, the SpaceXAI team will build a company from the ground and livestream it.“We'll use Grok Bot for every part of the build – from ideation and product development to real engineering work and deployment.” pic.twitter.com/iz5s0R3ZaE

— Sawyer Merritt (@SawyerMerritt) September 10, 2026

It remains a vendor showing off its own tool rather than an independent trial. The useful measure is how much the three humans steer, edit and rescue the agents along the way.

Questions about who answers for AI agents when they act alone are still unsettled.

Musk Concedes Grok Is Not the Default Yet

The build lands in an awkward week for agent products. Rival lab Anthropic published a report on September 11 documenting how people misused its Claude models, covering cyber operations, surveillance, fraud and conventional weapons work. Anthropic said it removed the accounts involved.

Meanwhile, reports suggest that every side in the Gulf conflict involves a Claude user, a nuance that Elon Musk acknowledges, conceding that Grok is not the default, at least for now.

“I guess Grok is not yet a preferred choice in this arena. Not sure how to feel about that,” he chimed.

Absence from a rival’s abuse report is not a safety record. Labs only publish what they catch on their own systems, and Grok Bot is barely a month old in public.

By Thursday the stream should show what the word company actually covers here. A working product with users would land differently from a name and a landing page.