Agents that manage agents: one engineer, five bots and 2,000 pull requests a month
SpaceXAI engineer Lingxi Li builds Grok Bot with Grok Bot. Five bots, each owning a domain, spin up cloud agents and verify proofs; the team now ships major work daily rather than every few weeks.
The most concrete field report of the agent era comes from SpaceXAI. Engineer Lingxi Li published a long write-up on building Grok Bot with Grok Bot, which developer Netto Farah shared as "the best engineering article you will read this month". Li's framing is plain: think of Grok Bot as a highly capable engineering intern, with its own computers, that manages coding agents and learns from how you work. The teammate that keeps things moving while you are away, asleep or in meetings.
The numbers carry the claim: one teammate shipped more than 2,000 pull requests in the past month; two engineers built the foundation of Grok Bot in four weeks, using Grok Bot; Li built the iOS version in three weeks on his own. Everyone on the team, he writes, now delivers major work every single day rather than every few weeks.
Five bots, five domains
Li runs five engineer bots, each owning an area: the mobile shared layer and iOS, the desktop client and CI/CD, infrastructure plus user issues with unclear ownership, Android, and the harness itself. The bots can work across each other's areas, but Li explains his choice: each has a different memory system and limited context, and they perform best focused on a single domain, because the specs and design principles they carry are much sharper inside the areas they own.
The workflow is agentic end to end: bots create Cursor cloud agents, read transcripts, review proofs attached to pull requests, queue follow-up messages or interrupt a run when something goes wrong. Tasks arrive from Li or from Slack; the bot kicks off an agent with a thorough prompt detailing what must be done and what proof is expected. Personal skills get invoked along the way: /lingxi-design for visual work, /react-native-best-practices for code quality, /lingxi-review for judging architecture.
One hardware detail matters too: Grok Bot can start cloud agents on your own machine, a spare Mac mini for instance. As Li puts it, you no longer need a dedicated machine running at home around the clock. If your workflow needs VPN access or a special configuration, that machine becomes a private worker, which unlocks things like running the iOS Simulator and getting screenshots back from your agents.
Context: this week we covered Hermes, where agents DM each other, what a harness actually is, and OpenClaw running on your own devices. Li's report completes the picture: the question is no longer what a single agent can do, but how agents managing agents reshape an engineer's day.
