<- all tokdocs

Grok Bot's Manual Onboarding: How Many Users, and Why It Worked

By , founder of Mavric and Apso

View on TikTok ->

Roman Ugarte, who helped build SpaceXAI's Grok Bot, told Lenny Rachitsky on Lenny's Podcast that his team personally onboarded "a couple hundred people" over about two weeks, sitting through live sessions where the product failed. A TikTok from @fieldnotesfl clips that exchange and captions it "We manually onboarded 200 to 300 users," but that range came from Lenny's question, not Ugarte's answer. The idea itself traces back to Paul Graham's 2013 essay "Do Things that Don't Scale."

Who is in the clip?

The clip shows no names, logos or product screens. The audio identifies the interviewer when the guest says "I'm glad you got a good one, Lenny." Lenny's podcast RSS feed has the source episode: "How we built Grok Bot in a month | Roman Ugarte (SpaceXAI)," published 8 September 2026, with a chapter at 11:20 titled "Manually onboarding a couple hundred people."

The show notes say Ugarte "helped incubate and build Grok Bot, the popular new knowledge-work agent from SpaceXAI," and that he previously led Growth at Cursor. Lenny says the team onboarded him personally. VentureBeat names him among Grok Bot's early testers.

How many users did the Grok Bot team onboard manually?

Three sources give three numbers:

  • Lenny's question: "I heard you and your team onboarded two to 300 people manually, including me."
  • Ugarte's answer: "there was about a two week period where we were in that mode and onboarded a couple hundred people."
  • The episode show notes: "Why the team personally onboarded nearly 300 of its first users."

The TikTok caption presents "200 to 300" as the founder's own words. The order of magnitude, a few hundred, is the only part all three agree on. The full transcript is behind a paywall, so a precise count could not be checked.

Why did the team onboard users by hand?

To watch failures happen and fix them fast. Ugarte describes having to "sit on a call for 20 minutes when the computer isn't spinning up or when someone's in onboarding and they're just incredibly confused." Those sessions "were pretty painful," and they set up a same-day fix loop: "that can never happen again. We need to solve this tomorrow because tomorrow I'm onboarding this person and it needs to go better."

"The computer isn't spinning up" describes the product itself. VentureBeat reports that "Each Bot operates through its own computer environment, can continue working when the user's laptop is closed." A remote machine had to start before a new user could do anything, and when it did not, the founder watched it fail live. The episode's chapters include "Cloud-first architecture" and "Long-lived agents, persistent memory, and the computer abstraction." Wikipedia describes the agents as sharing "a cloud-based computer," so sources differ on whether each Bot gets its own.

Was Grok Bot really built in a month?

The internal build was. The episode description says a small team "took it from first line of code to a working internal product in four weeks, and to a hugely successful public launch just three weeks later." That is seven weeks in total.

VentureBeat dates the beta to 11 August 2026, "available beginning today, August 11 in beta for SuperGrok Heavy, Cursor Ultra and Cursor Premium Teams subscribers," at $300, $200 and $120 a month respectively. The two-week onboarding sprint falls in the weeks around that launch, though no source gives exact dates.

What was happening at SpaceXAI and Cursor at the time?

The launch came during a corporate transition. SpaceX announced on 2 February 2026 that it had acquired xAI, which Wikipedia says was rebranded SpaceXAI in July 2026. According to Wikipedia's Cursor article, SpaceX secured a right to buy Cursor for $60 billion on 21 April 2026, exercised it on 16 June, and closed the deal on 14 August 2026. Grok Bot's beta launched behind Cursor subscription tiers three days before that close.

The name moved again soon after. Calcalist reported on 4 October 2026 that Musk answered a question about renaming SpaceXAI to SpaceXSI with "Yes, we will make that change," and the company's X account was renamed SpaceX Super Intelligence on 10 October, two days before the TikTok posted.

Where does the manual onboarding idea come from?

Paul Graham's essay "Do Things that Don't Scale," from July 2013, says "The most common unscalable thing founders have to do at the start is to recruit users manually. Nearly all startups have to." His Stripe example is close to what Ugarte describes: "When anyone agreed to try Stripe they'd say 'Right then, give me my laptop' and set them up on the spot."

Ugarte's version is narrower. Graham argues for manual recruiting to grow and to learn what users want. Ugarte uses manual onboarding to find defects and fix them the next day.

Lenny has covered the approach before. His March 2025 post on Superhuman, with founder Rahul Vohra, describes manually onboarding every new user as a long-running program. Grok Bot's team ran it as a two-week burst before automating.

Bottom line

The Grok Bot team hand-onboarded a few hundred early users over about two weeks, and the founder's own figure is "a couple hundred." The useful lesson is the loop: sit through each failed session, then fix it before the next day's onboarding.

Sources

Published October 12, 2026. Based on a TikTok by @fieldnotesfl. Full video notes (markdown)