An experiment called Grok Bot Galaxy tests the ability of artificial intelligence agents to participate in founding a startup from scratch within 72 hours, rather than limiting them to individual tasks such as writing code or generating text. The experiment began in San Francisco on September 15 and continues through September 17, 2026, with the work stages broadcast live, according to the published material.
SpaceXAI is participating in the experiment, which is led by three members of the Grok Bot team: Matt Palmer, Lauren Tan, and Roshan Sadanani. Rather than starting with a ready-made product or idea, its organizers said the team began from roughly a blank page, without a specific company name, product, or final business concept.
From Writing Software to Operating Company Functions
The core test involves assigning multiple roles to artificial intelligence agents, while humans remain responsible for decisions and oversight. The cycle targeted by the experiment includes defining the idea, developing a business plan, developing and engineering the product, and then moving on to sales, marketing, customer service, and after-sales support.
The announced agenda includes sessions on engineering and development, product management, startup founding, sales engineering, sales, SDR business development, customer service, marketing operations, and marketing and after-sales. Thus, the question is not limited to whether Grok Bot can produce software; it extends to whether agents can be used as a work team performing interconnected functions within a startup.
What Did the Team Accomplish on the First Day?
According to the details from the first day, the team discussed several directions before narrowing the project’s scope, and used Grok Bot for ideation, planning, and product development. It then moved on to creating a development environment and an actual software project, favoring the construction of a rapid prototype over spending time preparing an extensive theoretical plan.
An idea related to providing tools and solutions for the restaurant sector also emerged while searching for a product that could be built and tested within the short deadline. However, the material does not present a final result proving that the idea became a completed product or a service ready for use.
From Building to the Market
The second day of the program was devoted to sales and business development, including sales engineering, sales, lead development, and customer service. This stage is important because it shifts the evaluation from the agents’ ability to build software to their ability to identify a paying customer, explain the product’s value to that customer, reach them, and then maintain the relationship.
The third day was devoted to marketing operations and after-sales support before the final Final Showcase. Therefore, judging the experiment depends on what the team actually presents: a company with a usable product and potential customers, or a prototype, an interface, and a limited marketing experience.
Why Does This Test Matter?
The experiment’s significance lies in placing artificial intelligence agents before an interconnected series of business and execution decisions, rather than a single technical task. However, it remains an experiment limited by an announced timeframe of 72 hours, and the available material does not provide independent measurements of quality, cost, or product performance after launch. Therefore, the visible result in the final showcase alone is not enough to prove that agents can replace a complete human team in founding a company.