Lenny's Podcast: Product | Career | Growth

How we built Grok Bot in a month | Roman Ugarte (SpaceXAI)

1h 23min

Roman Ugarte shares how he and a small team built Grokbot, an AI knowledge work product, from scratch in one month at SpaceXAI. The key to its rapid success was starting fresh with a cloud-based, persistent agent paradigm—each bot gets its own computer and runs in the cloud—rather than layering onto an existing coding tool. The discussion covers the importance of ruthlessly simplifying the product, manually onboarding early users to gather real feedback, and the philosophy of treating AI bots as teammates rather than tools.

Summarized by Podsumo

🎧 Listen 🎙️ Ask about this episode

✨ Key Takeaways

1

Grokbot was built from scratch by a small, isolated team in just one month, moving from first line of code to internal beta in that timeframe.
2

The team made two critical early decisions: bots should run persistently in the cloud (not on a user's local machine) and each bot should have its own virtual computer to perform tasks without sharing a user's screen.
3

Roman emphasizes the importance of 'unshipping' features and killing pixels—instead of adding new buttons, the team focused on making bots reliably complete complex tasks like sourcing candidates or managing email.
4

Early success was validated by internal adoption, with teams like sales and recruiting switching from ChatGPT to Grokbot for daily work, and by hundreds of people attending a Grokbot meetup just weeks after launch.
5

The team manually onboarded 200-300 early users (including the host) to gather direct feedback, which helped them identify critical improvements like fixing browser automation for sales workflows.

💬 Notable Quotes

"The ultimate vision of Grokbot is incredibly simple. You should have a team of AI bots that help you with your job and help you with your life." — Roman Ugarte"
"Once you start breaking out of 'this is AI chat with a set of connections' to 'this is a colleague with a computer,' it just raises the ceiling of what you would think to give to AI." — Roman Ugarte"
"If you can make a thing that is not possible now possible, users are going to come to you. Through all of that, you'll gain distribution and data advantages. Moats are discovered, not planned ahead of time." — Roman Ugarte"