← All articlesORCHESTRATION

Lauren Tan’s Chief of Staff Bot

A coordinator bot finds the right engineering work, routes it to specialist Grok Bots, and returns evidence for one human review queue.

The pain point the bot solves

A team of agents creates a routing problem. Each specialist may be capable, but someone still has to inspect the incoming work, decide who owns it, pass along the context, and collect the results. If the human stays in the middle of every handoff, the agent team has not removed much coordination work.

Lauren Tan’s Chief of Staff bot owns that routing layer for an engineering workflow. It looks for relevant bugs, assigns them to Grok Bots responsible for different product areas, and brings the finished work back into one review queue.

About Lauren Tan

Tan works at Cursor and has moved between individual-contributor and engineering-management roles. In her public explanations of agentic engineering, she keeps returning to the same constraint: generating more code is easy; verifying and managing the work is the harder system to design.

Her Chief of Staff demo reflects that management lens. The coordinator does not try to become the best agent at every engineering task. It keeps ownership clear and hands the work to specialists that already have the right product context.

How the Chief of Staff works

In Tan’s demo, the Chief of Staff first identifies interesting bugs from the day’s feedback. She then tells it to dispatch the work. The bot messages a Grok Bot team whose members are responsible for different features, so each task lands with a specialist instead of a generic queue.

Those specialists can write code directly and can also start Cursor cloud agents. Tan showed an engineering bot managing a cloud agent on its own computer. The result returns as a pull request plus screenshots and video for review.

  • Scan the agreed intake source for work that meets the routing criteria.
  • Match each item to a named specialist with a clear area of ownership.
  • Pass the report, context, and required proof into the specialist thread.
  • Track the work without pulling the human into every handoff.
  • Collect pull requests and evidence in one review queue.

Where this setup breaks

A coordinator only works when the specialist roster is explicit. Overlapping ownership creates duplicate work. Missing review rules let a fast handoff become a fast mistake. The Chief of Staff needs a routing map, a definition of done, and a short list of actions that always stop for approval.

The first safe version should coordinate one recurring queue with two or three specialists. Add another bot when a stable job exists, not because the org chart looks more impressive with another avatar.

Sources