Seven AI Employees Sound Great—First Look at How They Hand Off Work

Morty posted Lauren Tan's Grok Bot setup, and at first glance it looks like an org chart for a small company. The workflow has only five steps: goal, delegation, specialists doing the work, merging results, and final reply.

The seven roles are divided very specifically:

• Chief of Staff is responsible for planning and assigning work
• Researcher is responsible for finding information and making comparisons
• Engineer is responsible for building and testing
• Reviewer is responsible for checking quality and edge cases
• Doc Writer turns work into reusable knowledge
• Operator keeps the workflow running continuously
• Memory Manager preserves context across sessions

This diagram is the easiest thing to copy as a template. What’s really hard is how the roles hand off results to one another. One commenter said: seven specialists, one shared inbox, zero excuses—the merge step is the most likely to break, and it’s best to add a decisive tie-breaker rule; otherwise the Reviewer and Doc Writer can argue forever. That reminder is more useful than the role chart.

Another underrated position is the Chief of Staff. One user has managed about 20 agents and said the specialists were never the problem; the problem was that everything landed on him at the same time. Once one agent prioritized things first and then gave him a clear yes or no, the situation changed. Regular users are the same: there are too many entry points, and in the end they’re still manually moving things around in the chat box.

The Memory Manager and Doc Writer don’t sound cool, but they determine whether this whole thing can survive overnight. In a reply, Ricker said the Memory Manager is one of the most important roles. Klonzu said the Doc Writer is underrated—many setups are forgotten by the next day. Camila was even more direct: she said she only needs the Memory Manager, because every new chat forgets what was said ten minutes earlier. There’s no mystery here—it’s just context preservation and knowledge accumulation.

Costs have to be counted too. Adam James Hill said seven roles mean seven bills. The Reviewer is worth paying for; the other roles can run on the cheapest models that can pass the checks. That judgment is very pragmatic. Seven roles look like a small company—and so does the bill.

Others poured cold water on it. Kimblee said good AI fleet managers were already doing similar things a year ago, and she is just promoting Grok Bot. Peter Church put it more interestingly: at first he thought it was a token-burner, but later he applied the bot pattern to other things and the results were unexpected—after a year of building micro-agents, it still wasn’t as good as this setup. Put the two comments together: the number of roles can be smaller, but the process, handoffs, and merge rules cannot be missing.

Ricker’s complete playbook is here: article link, and the original tweet is here. If you want to copy Lauren Tan’s seven-person team, don’t rush to spin up seven bots. First define one entry point, write down clearly who hands results to whom, and then prepare a rule for the merge step that can make the final call. Otherwise, all you get is seven chat windows.