All use cases

Work

Ten concepts before you build a Grok Bot team

New users spawn a fleet first — then learn plugins, gates, and routines the expensive way.

Share

What hurt

Michael (@Michael_Fenech_) posted a beginner thread: if starting with Grok Bot today, learn ten concepts before a huge team — (1) agent = job/objective/context/tools/boundaries/done; (2) plugins as ready-made connections; (3) MCP as the connection framework; (4) when browser/computer use beats plugin/API; (5) business context; (6) permission ladder read→draft→recommend→change→send→buy→publish; (7) routines only for proven workflows; (8) event triggers vs timers; (9) handoffs/one owner; (10) feedback loops after important tasks. Tip/thread — complementary to five-qs-before-routine.

What to install or create

One useful agent first. Connect with plugins. Gate consequential actions. Schedule only what already works. Name one owner per job.

Bots

  • Concept-one desk

    Single useful agent while you learn the ten concepts. No fleet yet.

Plugins

  • Browser

    Optional

Routines

  • None on the first run. Watch one manual pass first.

The prompt

Author-quoted ten starter concepts (thread summary)

Michael (@Michael_Fenech_) posted this 10-part thread starting at the cited tweet. Summarized from his note_tweet/thread text; not invented.

1. An agent isn’t just a chatbot with a name. Give it a clear job, objective, context, tools, boundaries and definition of done. Start with one useful agent before creating a team.
2. Plugins are packaged integrations (email, files, calendars, CRMs). Ready-made connections.
3. MCP is the underlying connection framework; plugins are usually what you install first.
4. Browser/computer use when there is no clean integration — prefer plugin, MCP, or API when available.
5. Teach business context: who you serve, what you sell, priorities, constraints, success.
6. Permissions ladder: read / draft / recommend / change / send / buy / publish — clearer gates for more consequential actions.
7. Routines: schedule only a proven workflow — every morning, Monday, month-end — not a broken one.
8. Triggers: some jobs start because something happened (email, form, lead status, new file), not only on a timer.
9. Handoffs: one job, one owner; say what is done, what the next agent needs, when to escalate.
10. Feedback loops after important tasks: what worked, failed, missing, change next time, permanent rule.

First run

  1. 1Read the thread from https://x.com/Michael_Fenech_/status/2099260874045395194.
  2. 2Create one bot with a real job, objective, tools, boundaries, and done definition.
  3. 3Connect needed plugins; only fall back to browser when no connector exists.
  4. 4Write the permission ladder into Description; gate send/buy/publish.
  5. 5After one clean manual win, save a skill — then consider a routine or a second bot with a clear handoff.

Approval fence

Map send/buy/publish to explicit yes before any routine goes live.

What to tweak

Steal the ten-concept order. Fleet size is not a beginner metric.

Source

Michael | AI + Agents @Michael_Fenech_ · Original post

Not affiliated with xAI. We did not write this prompt. Third-party template terms still apply.

Browse templates

Browse team blueprints

Pro pack

Want the full setup, not just the idea?

Pro packs give you the job text, prompts, schedules and pitfalls for popular use-cases.

See Pro