How we built Grok Bot in a month | Roman Ugarte (SpaceXAI)
- ID
- 22352
- Status
- summarized
- Published
- 08 Sep 2026, 8:31 PM
- Fetched
- 08 Sep 2026, 9:11 PM
- Provider
- Lenny's Newsletter
- Category
- product-startup
- Original URL
- https://www.lennysnewsletter.com/p/how-we-built-grok-bot-in-a-month
- Source URL
- https://www.lennysnewsletter.com/feed
Summary
- Score
- 6.0
- Created
- 08 Sep 2026, 9:12 PM
- Tags
- Audience
- developerssaas_foundersai_agent_users
What happened
Roman Ugarte of SpaceXAI describes how a small, isolated team built Grok Bot from first line of code to working internal product in four weeks, with public launch three weeks later. Key decisions included building from scratch rather than bolting it onto Cursor (where Roman previously led Growth, scaling the company from 15 to 1,000+ before its acquisition by SpaceX), and personally onboarding nearly 300 of the first users. The interview covers their 'colleague-pilled' product philosophy and what created durable moats for Cursor in a hyper-competitive market.
Why it matters
The concrete takeaway is the manual-onboarding strategy: onboarding ~300 first users by hand rather than automating, which suggests that for early AI agent products, high-touch onboarding may be more valuable than self-serve flows. The build-from-scratch-vs-extend-existing decision is also worth examining for anyone deciding whether to ship an agent as a new product or add it to an existing codebase.
Discussion angle
Is hand-onboarding 300 users a viable go-to-market for Malaysian indie builders shipping AI agents, or does it only work when you have SpaceX-level resources behind you? Compare against self-serve launches most local founders default to.