How I use Claude Code as my own bot: a crew of AI assistants in the terminal

Short answer: Grok Bot runs my daily briefing in the cloud. My other "bot" runs on my Mac: Claude Code, Anthropic's AI agent that works in the terminal, set up as a crew of assistants, each with its own role. They read my real files and calendar, argue from their roles, then do the work. Nothing leaves without my yes.

What I mean by a crew

• Three assistants, each with one job:
  • A Chief of Staff, who oversees the others and makes the call.
  • An Executive Assistant, who executes across my work and chases follow-ups.
  • A Personal Assistant, who looks after my personal life only.
• Other crews can join when the topic needs them. A Speaking Coach from my coaching crew joins when a talk is involved.
• Each role is written down once: its job, what it never does, how it speaks. Claude Code loads that when I call the crew. (Skills are explained in my Projects, Skills and Agents post.)
• Each role has its own voice. I select a role's text and my Mac reads it aloud in a different voice for each role, so it sounds like a meeting, not a wall of text.

One evening with the crew

• On the night of 31 October I brought the crew a personal question: is my professional presence holding back my earnings?
• In scope: how I dress, how I speak on stage, and a regular training routine.
• They argued from their roles. The Executive Assistant pushed to act this week. The Personal Assistant guarded my personal time. The Chief of Staff weighed both and decided.
• Then they did the work, not just the talk:
  1. Read my real calendar and placed two weeks of training blocks around the meetings and talks already in it.
  2. Added speaking practice sessions that rehearse the talks I am actually giving next.
  3. Sent a research agent off to build a clothing shortlist, every item with a source and a price checked live, while our conversation carried on.
• Nothing was booked, bought or sent. I got proposals I could say yes or no to.

The four rules that make it safe

1. Nothing is sent, bought or published without my yes.
2. Every fact has a source. If a role cannot point to one, it says so.
3. One memory, shared with everything else. The crew reads the same notes as the rest of my AI system, so I never explain myself twice.
4. A short state log at the end. The crew writes down where we stopped, so the next session picks up from there.

Why a crew here, and one bot in Grok Bot

• My Grok Bot lesson was: one agent, one output. That is for a daily routine.
• A decision is different. A few roles that disagree show me the trade-offs before I choose.
• The difference that matters: the crew only speaks when I open it. It never pings me.

How to start your own crew

1. Write one short role description per assistant: its job, what it never does, how it speaks.
2. Give them one shared notes file to read and update.
3. Bring a question you have been avoiding.
4. Ask for actions with sources, not advice.
5. Keep the final yes for yourself.

FAQ

Do I need to know how to code?

No. The roles are plain-language instructions. Claude Code runs in the terminal, but I talk to it in sentences.

Can the crew act on its own?

It reads, plans and prepares. Anything that sends, buys or publishes waits for my yes.

Why give each assistant a different voice?

Hearing the roles in different voices helps me follow who is arguing for what, without reading long blocks of text.

---

Sam Wong (黃力桐) is a Hong Kong AI adoption trainer who has trained 11,000+ people.