What Grok Bot Galaxy showed about AI teammates that actually work

Last week, SpaceXAI ran Grok Bot Galaxy, a three-day hybrid event at The Howard in San Francisco (Sept 15–17, 2026) with a free parallel livestream. The pitch was simple: live demos and hands-on build sessions with the team behind Grok Bot, framed as “your team of extraordinarily capable AI agents.”

We didn’t organize Galaxy. We watched it the way we use the product: as a small software shop that already runs Grok Bot teammates for marketing, sales, website, and ops. The useful question for us wasn’t “is AI cool?” It was: what does a public, multi-day build reveal about giving agents real work?

What the event actually was

• Where: The Howard, 661 Howard Street, San Francisco, plus livestream
• When: Tue–Thu, Sept 15–17, 2026: in-person roughly 8:45am–6:00pm PT; livestream windows listed 8:30am–6:00pm PT
• Watch / hub: x.ai/galaxy, Luma livestream, X broadcast

The livestream’s through-line asked whether three people could build a company in three days with Grok Bot. The builders named on the official Luma page were Matt Palmer, Lauren Tan, and Roshan Sadanani. Alongside that narrative, Galaxy ran role-specific sessions (engineering, product, founders, sales, SDRs, support, marketing ops, post-sales, and marketing) with hosts listed on the official Galaxy page.

What we’re taking into Viking Labs work

Three habits show up again and again in how Grok Bot is positioned (product pages and docs, not just the stage):

1. Give the bot real work, not a chat toy. A named bot, a goal, a style, and access to tools beats a generic prompt box.
2. Specialize by role. Galaxy’s schedule was basically an org chart: eng, PM, founders, sales, support, marketing. That’s how we’ve been staffing our own agents: one owner per lane, clear handoffs.
3. Keep a human in the loop. Our client work and our own products (including ResearcherFlow) assume review checkpoints for anything nuanced or risky. Agents accelerate the boring middle; people still own judgment.

If you’re evaluating AI teammates, Galaxy’s useful artifact isn’t a slogan. It’s a public schedule of department-shaped use cases, plus a multi-day build under a camera. Start with one bounded workflow, measure the return, and expand only when version one is earning its keep.

Links

• Event: x.ai/galaxy
• Product: x.ai/bot
• Docs: cursor.com/docs/grok-bot
• Viking Labs: vikinglabs.com

CTA: Bring us one repetitive workflow. We’ll map the smallest useful first build. Or try ResearcherFlow’s free paper summariser if you’re in academic research: researcherflow.com/tools/paper-summary