BH Technology Book a consultation

Back to Capacity

Department of 10 · Setup 2 of 5

Connected tools

What makes a connected workflow different from owning several tools? Work crosses a real permission line, a named person approves, and each step leaves a log.

Example service · the company builds and owns this queue
  1. Bot zone · restricted access

    1QueueReads approved fields and removes duplicates

    The service reads approved HubSpot fields and removes duplicates by stable deal ID. It holds the work. It does not decide what is correct.

    In this example, 3 builders use Codex through ChatGPT Business to maintain and test the service.

  2. Bot zone · assigned work only

    2Prepared packA draft and next task for one assignee

    Each bot gets only its assigned work through a restricted app or export. Across 10 Grok Bot Teams seats, the lead gets a morning brief, reps draft follow-ups from their queue and operations gets handoff notes.

    Limit to know: app permissions sit on each member’s identity, so bots under one member share that member’s access.

  3. Approval line · own login

    3Person checks and approvesNothing goes out without this step

    The assigned person checks the pack and approves it, corrects it or rejects it. There are no shared logins and no blanket auto-send.

    In this example, 3 reviewers using Claude Team check results and update playbooks, so a correction can improve the next run.

  4. Record · logged

    4Logged record updateDraft and approval logged separately

    Separate draft and approval entries show what was prepared and who accepted it. For example, a reviewed proposal can trigger an approved record update and create the next follow-up task.

    The records in this example are HubSpot, Google Workspace and Supabase Postgres.

When it does not go to plan

  • Fails a checkMoves to Exception for a manager

    The item leaves the normal path, and a manager decides. Managers also approve templates, so a repeated exception has an owner who can change the template.

  • Permission failureThe run stops and is logged

    It never borrows wider access. ChatGPT Business and Claude Team logins stay with their users, and bots get only the restricted app access.

  • Integration failsWork falls back to manual

    The queue keeps moving by hand until the connection is fixed. That only works if people have been trained on the manual path.

The test: software alone is not a team workflow. Without permission, logging and ownership, it is only moving information around. Enterprise-only controls are not assumed here.

The company stack and the example serviceSeats, roles, ground rules and product sources

Connected tools

The company stack

  1. Grok BotRuns jobs
  2. QueueHolds work
  3. PeopleApprove
  4. HubSpotRecords
Grok Bot Teams
10 company seats. The lead gets a morning brief, reps draft follow-ups from their queue, operations gets handoff notes.
ChatGPT Business
3 builders use Codex to maintain and test the queue service.
Claude Team
3 reviewers check results and update playbooks.
Records
HubSpot, Google Workspace and Supabase Postgres.
Code
VS Code and a shared repo.

Ground rules

One example service

  1. ReadsApproved HubSpot fields
  2. ChecksRemoves duplicates by stable deal ID
  3. AssignsGives each bot only its assigned work through a restricted app or export
  4. RecordsDraft and approval logged separately

The company builds and owns this queue service.

  • App permissions sit on each member's identity. Bots under one member share that member's access.
  • ChatGPT Business and Claude Team logins stay with their users. Bots get only the restricted app access.
  • No shared logins and no blanket auto-send.
  • If the integration fails, work falls back to manual.
  • Enterprise-only controls are not assumed.

Back to Capacity