Wink Pings

1.4 million people watched an AI bot team demo, then went back to working solo

Watching a demo is easy; turning a demo into your own workflow is hard. 200 agents often lose to 6. kozh./ breaks down the reasons and the role division, and the official team included free guides.

One video. 1.4 million views. In 30 minutes, Matt Palmer used Grok Bot to build a bot team that can operate a cloud computer on its own: move X followers into Notion, look up information, build software, and dispatch coding tasks to Cursor. In the demo, one bot completes its step, passes the result to the next, and keeps running after you close your laptop.

Most people watch it and treat the video as entertainment. A few go back and build a workflow for themselves. The difference isn't AI capability; it's whether you take action.

kozh./ pointed out a counterintuitive phenomenon in another long post: 200 Grok Bot agents may lose to 6. He described a scenario: you open the dashboard in the morning and 200 agents are running. Several are reading the same project brief, several are searching for information others have already found, three are editing the same file, one is retrying the same broken tool for the tenth time, and several mark their own work as "done." The dashboard looks busy, but the project hasn't moved.

How many agents you can run isn't important. What matters is how much verified work remains after they finish. Every agent consumes tokens. Send the same 12,000-word brief to 200 agents, and the input alone is 2.4 million tokens, not counting searches, tool calls, and retries.

His advice: don't start by assembling an imaginary 200-person company. Start with 6 roles—

  • Gatekeeper: Receives requests, defines goals, budget, deadline, and approval boundaries. Refuses to start if the task is too vague.
  • Scout: Finds relevant sources and returns a condensed evidence package.
  • Spec: Turns the request into acceptance criteria. What counts as done, and what is explicitly out of scope.
  • Builder: Produces artifacts: code, research briefs, landing pages, workflows.
  • Breaker: Checks artifacts against the original spec, looking for holes, edge cases, and claims unsupported by evidence.
  • Operator: Records what passed, what failed, and what needs a human decision. Publishes or makes changes according to the rules you set.

These 6 roles don't meet in one giant chat room. Each has its own inputs, outputs, and stopping points. At handoff, an agent should give the next agent a compact work order, not just "I looked into it, the next bot can continue." The work order should contain: goal, sources, decisions, existing artifacts, verification evidence, open questions, next owner, stop rules.

The core of this architecture is minimizing the distance from request to verified result. Parallelism is not the goal.

The official team also prepared several free guides for people who want to take action:

  • Grok Bot 101: Written by Matt Palmer, on building your first bot from scratch, connecting expert bots, plus four workflows he's currently using.
  • Grok Bot for Engineering: Lingxi Li shares how to run a fleet of engineer bots, manage 200+ cloud agents, and use a Notion database and an operations bot named "Jenny" for daily coordination.
  • How I run multiple teams of Grok Bots: Eric Zakariasson brings the project management model into bot teams: one channel per project, one Notion board, one project manager bot.

Matt Palmer uses a bot to move X followers into Notion

If you want to go from spectator to someone who takes action, the rule is simple: don't start with an imaginary company of 50 bots. Find one repetitive task and teach it to one bot. Once the first handoff is reliable, add a second. Bookmark this video and the guides, and go build the workflow that the other 1.4 million viewers only watched and never did.

发布时间: 2026-11-22 15:17