Day 3: The 72 Hours Are Up. The Final State of the Company Built from Scratch at Grok Bot Galaxy
The third and final day of Grok Bot Galaxy is over; the xAI/SpaceXAI team (Matt Palmer, Lauren Tan, Roshan Sadanani) has wrapped up its 72-hour livestream. Those who followed the first two days here will remember: Day 1 was about finding an idea and setting up the infrastructure, and Day 2 brought a big pivot from a restaurant pop-up idea to a game studio. As a follow-up to the first two posts, this one briefly sums up where things landed on Day 3 and what was actually achieved by the end of the three days.
The Product Went Live: Thursday Arena
On the morning of Day 3 (around 10:00 a.m.), the team actually launched the product: ThursdayArena.com, and @ThursdayArena on X. The auto-battler game, where you turn bot templates into game characters and pit them against each other head-to-head, was now playable by real users. It was deliberately rough and fast: “When we first shipped it, it was really bad, the UI was terrible, but we wanted to get it out fast,” as they put it.
To speed up the launch, they also used a classic growth tactic: new users who signed up in the first 10 minutes were given one month of Grok Bot for free ($200 worth of usage credits), and a similar credit campaign was run once more during the day.
Simplification: The Discipline of “Unshipping”
The most striking part was the team’s decision to shrink the product rather than grow it. Many of the ideas discussed on the Day 2 whiteboard were cut:
- The “rock-paper-scissors” mechanic between charisma, intelligence and agility was removed and replaced with a simple pure auto-battle mechanic.
- The idea of selling cosmetic items (hats, visual elements) was shelved.
- The stadium/arena advertising idea was kept and turned into a real ad auction system, which was tested live at the end of the day (with some hiccups, but it worked).
“As a constraint, we had to think about what we would unship, which let us simplify the product and ultimately get it to a good place.” — the team’s own summary.
The “Engineering Factory” and a Team Update
The defining feature of the last 24 hours was the bot engineering factory Lauren set up: chief bots kept a team bot in the loop, writing code, testing the game, fixing bugs, and auto-approving PRs and merging them into main. An inside joke that emerged along the way also stuck: because of Dr. Eggbot’s potato-themed naming habit, the team started saying “mash” instead of “merge,” and the jargon even made its way into Notion.
On the last day, a fourth SpaceXAI employee named Eric joined the team and was given write access to GitHub, in the team’s own words, to “5x the team’s velocity.”
So how many bots were there in the end? There’s no clear answer, because the team itself had lost count. Roshan, who on Day 1 complained that “my bot sidebar is pretty lonely,” was looking at the same screen at the end of Day 3 and saying: “We have, man, I don’t know how many Grokbots now.” Starting with three bots and ending up with too many to count is, in fact, one of the summaries of these three days: 3 humans and dozens of AI agents!
The tech stack became clear: PlanetScale for the database, Clerk for authentication, Vercel for deployment, and Stripe for payments.
End-of-Day Metrics
The team was getting a live dashboard from its own “data bot” every 15 minutes. The figures shared in the final hours of the stream:
- Around 30,000 page views.
- More than 6,000 total matches, with over 4,000 games played (including bot-vs-bot and real-user matches).
- Cumulative sign-ups approached 2,000.
- They started ranking #1 on Google for related searches.
- For the first time, a real player rather than a bot took first place on the leaderboard (29 wins, 3 losses).
Midday, the team also worked on a growth playbook prepared by its own “growth bot”: classic growth tactics such as auto-DMing people who follow them on X and partnering with game distribution platforms.
Three Days in Review: What Was Achieved?
Looking at it from start to finish, here’s the picture:
- Day 1: No idea → a pop-up restaurant experience concept called “Ship by Thursday,” with the basic infrastructure (Slack, GitHub, Vercel, domain) set up.
- Day 2: The pop-up idea was dropped → a pivot to a game studio, with a playable prototype and game mechanics designed.
- Day 3: The product was scaled down and simplified, launched (ThursdayArena.com), an ad/monetization system was tested, and real users started playing the game.
So at the end of three days, there’s a genuinely live product with registered users, thousands of matches played, a working trial of payment/ad infrastructure, and a growth plan: the first steps of a classic “pre-seed” startup, which usually take weeks or months, compressed into 72 hours with bot teams. The company’s formal incorporation, legal structure and revenue model haven’t matured yet, and the team didn’t hide that; they wrapped up the ad test with the joke that “theoretically, we earned our first sponsorship dollar.”
The Team’s Own Three Pieces of Advice
- Build in a field you know. This was exactly one of the reasons they pivoted away from the pop-up idea: they were trying to move forward in an area where they had no experience with permits, catering or venues. “The person who does something really well is the person who knows exactly what they’re doing.”
- Don’t hesitate to build a diverse bot team. For topics you don’t know, tell your bots, just like newly hired experts, to learn first and draw lessons from the best examples; feed them so they can act as your advisors.
- What really makes the difference is still the human. Even in the age of agents, the two types of people a team needs haven’t changed: someone with a clear vision and taste, and someone who relentlessly makes things happen. If you can find both in one person, that’s the best.
Closing
In the team’s own words, the real goal of this experiment wasn’t to build a game or a game studio: it was to show how Grok Bot makes it possible to work with a “bot team” in everyday work and personal life. Seen from that angle, producing a product that went from idea to real users in 72 hours, even with an imperfect, half-finished ad system and a business model that’s still unclear, is a striking result in itself.
Anyone curious can still try the product at ThursdayArena.com. As for me, I’m glad I followed this three-day series all the way through here. Now it’s time to bring my own experiments with Grok Bot to life…
Comment(s)