72 Hours to Build a Company from Scratch: Notes from Day One of Grok Bot Galaxy*
In a studio right next to Dreamforce in San Francisco, three people stepped in front of the camera. They had no business idea. No company name. No product. All they had was a microphone, a camera, and an AI agent app they had just downloaded.
This was the opening scene of Grok Bot Galaxy, the event hosted by xAI (now known as SpaceXAI). From September 15 to 17, it was an attempt to build a company from scratch, live on air for 72 hours. I watched the full recording of the first day, and in this post I’m sharing the company-building side of things rather than the role-specific sessions: the journey from idea to first live product.
What Is Grok Bot Galaxy?
After SpaceX acquired xAI in February 2026 at a $250 billion valuation, the two merged under the SpaceXAI umbrella. Grok Bot Galaxy is a three-day event this new organization held to showcase its autonomous agent platform, Grok Bot, hosted live at The Howard in San Francisco and streamed globally.
The rule of the experiment was simple: three SpaceXAI engineers, Matt Palmer, Lauren Tan, and Roshan Sadanani, would step in front of the cameras with no preparation and build a company for 8.5 hours a day, three days in a row, using only Grok Bot and the AI agents it creates. No idea. No name. Not even an industry. Everything would take shape during the stream, right in front of the audience.
“We’re going to build a company in three days... we actually don’t even know what idea we’re going to build yet.” — The very first line of the stream showed they were truly starting from zero.
The Team and Roles: Three People, One Bot Army
The team was small, but the division of labor quickly became clear, and they even assigned themselves titles with a sense of humor:
- Lauren Tan – CTO (jokingly, “Chief Potato Officer”). The most technical member; she set up the infrastructure (GitHub repo, Vercel deployment, PlanetScale database) and orchestrated the engineering agents by building her own “potato factory” of bots.
- Roshan Sadanani – CPO (Chief Product Officer). He ran the product and market side: user stories, prototypes, and coordinating the agents researching restaurants and venues.
- Matt Palmer – “CEO” (“the grown-up,” in their words). He took on operations and logistics: the company playbook, Notion documentation, Slack and account setup, and go-to-market (GTM) planning.
The trio’s real job wasn’t writing code; it was managing an organization of bots. Roshan’s wry confession later in the day summed it up perfectly: “my bot sidebar is pretty lonely,” said at a moment when he already had three bots. Hours later, that number would jump tenfold.
From Idea to Company: “Ship by Thursday”
Most of the morning went into ideation. They scanned suggestions coming in on X (Twitter) and traded ideas with guests who joined the stream (Cody Sanchez on growth and marketing, Peter Yang on product). A painting company, a piano-tuning business, a “company that builds companies,” and many other ideas were put on the table, eliminated, and reconsidered.
Eventually the decision crystallized: run a restaurant pop-up experience in San Francisco, and turn what they learned along the way into an “operating system” (platform) product that makes it easy to set up pop-up events. In other words, they would dogfood their own product by running a real event, and then sell the software that came out of that experience to other pop-up organizers.
For the company name, they found a pragmatic solution: “Ship by Thursday” was chosen as a temporary brand name until a final one is settled, and the domain (shipbythursday.com) was purchased on the spot through Vercel. Alternative names considered for the product included “Grokpot” and “peanutpopup”; it looks like this will be settled through iteration.
How Many Bots by the End of the Day, and What Do They Do?
This was the most striking part of the stream: over the course of the day, three humans created more than 10 named bots, each acting like a “digital employee” with its own responsibilities:
>Steve – The first bot Roshan downloaded; the main bot handling overall orchestration.Dr. Eggbot – Roshan’s task-delegation bot; it spawns research and outreach agents on its own.Grokpot – The prototyping bot; it produced alternative landing page designs.Tater – Lauren’s engineering bot; it made the tech stack decisions (Vercel, PlanetScale, TypeScript) and wrote the code.Hashbrown – The “code reviewer” bot that reviews the pull requests Tater opens.Ops bot and Creative Director bot – Matt’s bots, handling Slack and account setup, and naming and branding decisions.Marketing bot and Design bot – Worked on guerrilla marketing ideas (the 50 busiest street corners in SF) and flyer design.Knowledge Base Manager bot – Tasked with monitoring all bot conversations and keeping the “company truth” in Notion up to date.Outreach/Research bots – Agents that scan catering companies, restaurants, and venues in San Francisco to collect contact information, and are even hooked up to phone-calling infrastructure.At the end of the stream, the team left long-running agents working through the night: a “Scout” bot to research venues and draft emails, and a budget bot to handle the event budget and RFP processes. The idea: the bots keep working while the team sleeps, and when the team wakes up in the morning, emails awaiting approval will be ready.
Where Does the Company Stand at the End of the Day?
The picture at the end of day one is clear: the infrastructure is ready, the product is not yet. Concretely, here’s what got done:
- A Slack workspace, GitHub repo, Vercel project, and Notion “company doc” were set up.
- PlanetScale database integration was in progress; authentication via Clerk was planned for the sign-up flow.
- A target date was set for the pop-up event: October 15.
- No venue yet, no chef/catering partnership, permit processes not yet researched, no budget drawn up.
In short: the software skeleton came together in a single day, but the “real world” part of the job (venue, permits, vendors) hasn’t even started. A guest who joined the stream as an event consultant (the founder of an events company) summed it up: no matter how fast the bots are, putting on an event in San Francisco is a slow, human-centered process that requires dozens of permit types and vendor coordination.
5 Lessons from the Company-Building Process
- Bots are organized like a team, not like a toolkit. Instead of a single “super bot,” the team built specialized bots (engineer, reviewer, marketer, knowledge manager), each with its own context and memory. It resembles a human organization: when roles are clear, coordination gets easier.
- Speed is no substitute for clarity. By the end of the day, the technical infrastructure was running flawlessly, yet the team was still asking, “Are we solving the right problem?” Freedom to explore doesn’t mean clarity of decision.
- Do it as a human first, then automate. The team decided to first experience the pop-up manually themselves (making phone calls, writing emails) and derive the software from that experience, not the other way around. It’s a deliberate choice against the risk of building software without real-world validation.
- Context is the most valuable asset. The idea of using the Notion doc as the “single source of truth” for all the bots came up again and again throughout the day. This principle, which holds for human teams, holds for bot teams too.
- Distribution comes before product. The words of an entrepreneur-investor guest on the stream were striking: among the AI startups they had looked at, the number one reason for failure wasn’t the product but a lack of distribution. As building products with bots gets cheaper, what really makes the difference is the ability to reach an audience.
Closing: The Experiment Continues
At the end of day one, the team wrapped up tired but with a clear question: “Are we solving the right problem?” That question is really at the heart of every company-building process. The difference is that this team now has a bot army working through the night to answer it.
So, is it possible to build a real company in 72 hours by handing most of the human labor over to agents, or is this just an accelerated skeleton? The answer will become clear over the next two days.
Want to follow the livestream? Days 2 and 3 of Grok Bot Galaxy continue on air starting today (Tuesday). You can join the event and the livestream here: x.ai/galaxy
I’m eagerly looking forward to day two: will they find a venue, will the permits come through, what did the bots accomplish overnight? We’ll see together. I’ll keep sharing my notes.
*This blog post was created with artificial intelligence.
Comment(s)