Day 2: Pivot! At Grok Bot Galaxy, the Restaurant Idea Gets Shelved and a Game Studio Is Born*
*This blog post was created with artificial intelligence.
Yesterday I wrote about the first day of Grok Bot Galaxy, the xAI/SpaceXAI team’s (Matt Palmer, Lauren Tan, Roshan Sadanani) attempt to build a company from scratch in 72 hours: a restaurant pop-up experience and a platform to be built on top of it. Day two, however, opened with a surprising sentence: “The agents told us to pivot.”
- Organizing a physical event in San Francisco (permits, venue, catering) was not something that could be finished in two days. As the guest experts also confirmed, it’s a human-centered bureaucratic process that takes weeks or months.
- The format wasn’t “immersive” from a livestream perspective either. Permit research and vendor phone calls didn’t make for entertaining content to watch.
- The user uploads their own Grok Bot template, and the bot is automatically turned into a game character (“minted”).
- Each character has three stats: charisma, intelligence (int), and dexterity (dex). These are distributed from a fixed pool of points (base 100 points) and determined by an LLM based on the bot’s description text.
- There’s a rarity system: base, rare, ultra, legendary. Higher rarity means a higher total stat score.
- Players build a three-bot team and compete head-to-head against a rival team; matches are recorded as watchable/shareable replays.
- The revenue model is a deliberate choice: not “pay-to-win.” Instead of selling power, it relies on cosmetic items (hats, visuals) and “stadium”/arena advertising, like sideline advertising in sports that doesn’t disrupt the game’s balance.
“We need to shorten the time to fun.” — Roshan, summing up the goal of tweaking stats live in the prototype and making it instantly playable.
- A playable prototype is up: team building, character stats, and the head-to-head match mechanic are working and can be tuned live.
- The landing page and the asset playground (SVG character visuals) are live.
- Generative music experiments were done for the game (lobby music with Strudel and Suno).
- The ads marketplace and database integration aren’t finished yet; the team left that for the next day.
What Happened Overnight? Why the Pivot?
At the end of the first day, the team handed the pop-up idea over to agents working through the night and collected their feedback. When they returned in the morning, they faced two realities:
So the team made a radical decision: switch to a product the audience could be part of. The new idea: a game studio where you can create your own Grok Bot templates, turn those bots into game characters, and pit them against each other. The company’s working name became “Cupcake.”
The New Product: Bots Enter the Arena
The game mechanics they designed together on a live whiteboard were quite clear:
How Did the Team and the Bot Army Grow?
The roles stayed the same; only the focus shifted. Lauren handled engineering and agent orchestration through her own plugin set, PStack; Roshan drove game design and the prototype/playability side; Matt took on the landing page, SEO/AEO, and the ads infrastructure.
The bot fleet grew noticeably. Joining yesterday’s names like Steve, Tater, Dr. Eggbot, and Hashbrown were newcomers: Wisk (a game designer bot), Krum (a game-testing bot that goes into the app and clicks around hunting for bugs), a new founding engineer bot, a knowledge base bot that kept Notion up to date, and new bots put to work on the ads/marketing side. At the end of the day, Lauren made a humorous confession: “I have so many bots now.”
A notable detail: Lauren asked her agents “where are we getting stuck, and how can we speed up our workflow?” Dr. Eggbot analyzed the entire bot history and reported the bottlenecks on its own, for example “the serial factory is waiting on human approval.” Bots are no longer just doing tasks; they’re optimizing their own organization too.
Where Do Things Stand at Day’s End?
The team’s own summary is spot on: “We went from no app to an app with an engine and a bot army that carried the weight of the work today.”
In short: from yesterday’s “infrastructure ready, product not yet” state, today they moved to a “pivoted, still rough but live product with a playable core loop.” The real question remains open: was this pivot the right bet, or will tomorrow bring another one?
Two Quick Lessons
Speed lowers the cost of a pivot. Changing the idea after a day’s investment is a decision that might take weeks at a classic startup. Here, it became clear by morning thanks to feedback from agents working overnight. Perhaps the real benefit of agent speed is “wasting less time on the wrong idea.”
Bots aren’t just workers anymore; they’re organizational consultants too. Dr. Eggbot analyzing the bottlenecks of its own bot fleet is an example that goes beyond the “human boss, bot team” relationship: bots can also review the way they work. Of course, the growing number of agents has also started to make staying organized seriously harder.
Wrapping Up: The Final Day Is Coming
Day two began with the restaurant idea being abandoned within a day and ended with a prototype of a playable bot-battle game. Tonight is the final run: the last chance to actually finish the product, launch the ad infrastructure, and maybe pull off one final surprise.
A quick note, by the way: this event made a lot of noise, and I think the whole world piled onto Grok Bot, because for the past 24 hours the system has frequently been down. Of course, accidents like this are normal after such an ambitious event…
Want to follow the livestream? Day 3, the final day of Grok Bot Galaxy, is live today. You can join the event and the livestream here: x.ai/galaxy
Comment(s)