Elon Musk's AI company will attempt something rarely seen in public: building an entire startup from nothing in 72 hours, with every decision livestreamed for the world to watch. Three employees at SpaceXAI will start with no company name, no product idea, and no business plan on September 15, then use AI agents called Grok Bot to ideate, design, engineer, and launch a working company by September 17.

What Exactly Is Happening During the Grok Bot Galaxy Event?

The livestream event, called Grok Bot Galaxy, will run from 8:30 a.m. to 6 p.m. Pacific time each day at a San Francisco venue, with a worldwide broadcast available online. Matt Palmer, Lauren Tan, and Roshan Sadanani will be the three builders tasked with choosing the company name, product concept, and business strategy in real time as they work.

Grok Bot is not the Grok chatbot that answers questions on X (formerly Twitter). Instead, it is SpaceXAI's agent product, which the company describes as a team of AI agents that users can name, assign specific objectives to, and deploy across apps and websites to complete tasks autonomously. The three builders will use Grok Bot for ideation, planning, product development, and engineering work. Colleagues will drop in during separate sessions to handle sales, customer support, and marketing operations.

Why Is SpaceXAI Staging This Public Test?

SpaceXAI is the company formed after SpaceX acquired xAI in February 2026 in an all-stock deal that valued xAI at $250 billion. The company then acquired coding firm Cursor for $60 billion in August. Throughout 2026, Musk has made bold claims about Grok's capabilities, and rival AI labs have matched those claims with their own announcements.

A three-day unedited livestream is a harder test to stage than a polished demo video. The livestream will show how much human steering, editing, and rescue work is actually required when AI agents operate with minimal supervision. Questions about accountability and responsibility when AI agents act independently remain unsettled in the industry.

How to Evaluate the Startup Build Experiment

• Human Intervention: Watch how often the three employees need to correct, redirect, or rescue the AI agents when they make mistakes or go off track during the build process.
• Deliverables: Assess whether the final output is a working product with actual users, or simply a company name and landing page with no functional software.
• Real-World Viability: Determine if the startup could actually operate independently or if it requires constant human oversight to function properly.

The timing of the livestream is awkward for AI agent products generally. On September 11, rival lab Anthropic published a report documenting how people misused its Claude models for cyber operations, surveillance, fraud, and conventional weapons work. Anthropic said it removed the accounts involved.

Musk has acknowledged that Grok is not yet the default choice in certain high-stakes applications. Reports suggest that every side in the Gulf conflict involves a Claude user, a nuance that Musk recognized publicly. "I guess Grok is not yet a preferred choice in this arena. Not sure how to feel about that," he stated. 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 has been publicly available for barely a month.

By Thursday, September 17, the livestream should reveal what the term "company" actually means in this context. A working product with paying users would land very differently from a company name and a landing page with no functional software or revenue.