Starting on Sept 15th, SpaceXAI team members will build a company from the ground up using Grok @Bot and livestream it. Link to register for the stream is below.
Agenda:
"Starting with an idea and a blank slate, they’ll work through a business plan, feature decisions, and real engineering work to turn a concept into a reality. Over the course of this three-day livestream, you’ll be able to follow their building in real time and will see how Grok Bot fits into each stage of the development process.
Throughout the livestream, we’ll also hand off to teammates for department-specific deep dives. From the first line of code to setting customers up for success, you’ll learn how to use Grok Bot for real workflows across every stage of your business, and will walk away with actionable use cases for your function.
The livestream runs from 8:30 AM - 6:00 PM every day, with the following guest sessions:
Day 1: September 15
• 9:00 - 10:00 AM: Grok Bot 101
• 12:30 - 2:00 PM: Grok Bot for Engineering
• 2:30 - 4:00 PM: Grok Bot for Product Managers
• 4:30 - 5:30 PM: Grok Bot for Founders
Day 2: September 16
• 9:00 - 10:30 AM: Grok Bot for Sales Engineering
• 12:30 - 2:00 PM: Grok Bot for Sales
• 2:30 - 3:30 PM: Grok Bot for SDRs
• 4:00 - 5:00 PM: Grok Bot for Customer Support
Day 3: September 17
• 9:00 - 10:30 AM: Grok Bot for Marketing Operations
• 12:30 - 1:30 PM: Grok Bot for Post-Sales
• 2:30 - 4:00 PM: Grok Bot for Marketing"
Link to register for livestream: https://t.co/ncKCtdL3ui
My biggest takeaways from Grok @Bot product lead @RomanUgarte_:
1. Two key early product decisions made Grok Bot the success it has become. First, everything runs in the cloud. This means bots can be kicked off from a phone and run while the user sleeps. Second, each bot has access to its own computer. This allows it to operate on tools that have no API because they can simply use the computer the way a human would.
2. “Unshipping” is a big opportunity in product right now. One of the most counterintuitive things the Grok Bot team did in the weeks between internal beta and public launch was remove features. They initially exposed internal model thinking, memory traces, and developer-style observability tools inside the product. They decided to cut all that out. Their mental model for the product is a colleague: you wouldn't ask a teammate for second-by-second updates on every button they're pressing, so the bot shouldn't either.
3. The decision to make it a separate app (vs. inside Cursor) was not at all obvious. The team had the option to build Grok Bot inside Cursor, the way some competitors added knowledge-work tabs to their existing coding surfaces. Roman says the decision not to do that gave them complete control over every pixel and allowed them to build around a single, coherent vision of how an AI assistant should feel. The alternative, he argues, is "shipping your org chart"— and users can feel it.
4. Grok Bot went from the first line of code to a viral internal beta in four weeks. The team moved into a separate part of the office, set up private Slack channels, and made dozens of micro-decisions every day without needing buy-in from the larger org. This allowed them to go from internal beta to global launch in 3 weeks.
5. Roman and his team onboarded 200 to 300 people personally. This was one of the highest-leverage investments they made. When something broke or confused a user, the team felt it and fixed it the next day. Beyond the product improvements, the onboarding surfaced non-obvious use cases the internal team would never have imagined. A coffee shop owner found through a friend of a friend became one of Grok Bot’s most active users and a rich source of Shopify-integration feedback.
6. The Grok Bot team is “colleague-pilled” (vs. AGI-pilled). When the Grok Bot team hits a product decision with good arguments on both sides, they use a test: how would you want a human teammate to behave in this exact situation? Roman says the answer is almost always unanimous and clarifying. This is how Grok Bot landed on hiding internal mechanics from users, on building automations through natural language rather than UI, and on the huddle-style voice experience they’re building toward.
7. Grok Bot’s GTM team became one of its most enthusiastic internal user groups. Sales work depends on tools that often lack reliable integrations. Giving the bot a computer meant it could interact with those products the same way employees do. The resulting feedback forced them to level up Grok Bot’s computer use: the mouse needed finer control to click a particular part of a Salesforce dashboard. The infrastructure team would fix that invisible capability and hear the next day that the whole workflow finally worked.
8. Moats are discovered, not engineered. Roman said that if Cursor had tried to work backward from a moat strategy, it would not have created the product or the outcome it did. What worked was an obsession with building something genuinely useful today, pulling future capabilities forward even when they required significant engineering to stabilize, and then deleting that scaffolding as the models improved. The distribution advantages and data advantages followed from that.
9. Innovating requires the cultural willingness to abandon sunk costs. Roman’s explanation for why competitors haven’t made the same cloud-first, bot-gets-its-own-computer bets isn’t that they missed the insight. It’s that they’re stuck in an existing paradigm with real sunk costs. Most companies can see where the market is going; fewer have the stomach to act on it when acting means deprecating their current product. The advantage of building something new is that you get to be right about the future without the weight of having been right about the past.
10. Roman's advice for new users: connect Slack and email, then ask Grok Bot to go through them and suggest five things it could take off your plate. When he did this himself, two of the five were things no previous AI tool could have done, and he spun off two bots to handle them on the spot.
Built clawsweeper, which runs 50 codex in parallel around the clock, scans issues/prs deep and closes what is already implemented or what makes no sense.
Closed around 4000 issues today, a few thousand are in the pipeline. (rate limits are rough) https://t.co/AiNNDcvGke
We've reached the point in software development where it's far far easier for a small cracked team to actually build a product than for any team in a big co get the "approval" from "all the relevant stakeholders". This will have major consequences for most incumbents.
As coding agents start compiling English into code, the definition of WHAT to build becomes the highest leverage work.
That means teams need to collaborate deeply on requirements.
Some people write specs as markdown in Git repositories. But that isn’t a collaborative environment. PMs aren’t living in IDEs or PRs.
Software Factory brings a Cursor-like agent experience into a multiplayer document editor.
If you’ve wondered why Google Docs never built a true, agent-driven editor: it’s extremely hard.
We just shipped one.
Try it here: https://t.co/fkfTXgcI8c
Just gave a talk to 2,000 people today.
10 raised their hands; they've used agentic AI.
20% (maybe) knew what Claude was.
You are earlier than you think you are... still.
You can now push what you’re building in Claude Code directly into Figma.
With the latest updates to the Figma MCP server, build a working prototype in code, then send it to a Figma canvas to explore multiple versions.
The person who built Claude Code just mass-leaked the thinking behind it.
45 minutes of design decisions, mistakes, and where it's all going.
This is rare. Creators at this level don't usually talk this openly.