AI Systems · Nov 14, 2026 · 6 min read

AI Permissions for Small Businesses: Let AI Draft, Keep People Approving

An AI agent with no access to your email or calendar can only give advice. Connect it, and it can work on real data — reading leads, drafting replies, preparing reports. That is where the value is, and it is also where the risk is. The answer isn’t to avoid connecting AI. It’s to decide, app by app, what it can do.

In the Lagoon Galaxy System™ those connections are called Warp Gates, and the rules for each one are its clearance.

The rule that covers 90% of cases

AI reads and drafts. People send, publish, delete, move and pay. If an action can’t be undone, it waits for a yes from a person.

Four layers of control, simplest first

  1. Open only the connections a task needs. A content task doesn’t need your payment platform. A reporting task doesn’t need permission to send email.
  2. State the limits in the instructions. “You may read the CRM and the calendar. Don’t delete or change anything. Before sending an email, publishing or moving an appointment, show me and wait for my yes.”
  3. Use limited accounts. A connection acts as the account it signs in with. A view-only share can’t be edited, no matter what the agent is told.
  4. Use team controls where you have them. On business plans, admins can mark connector actions as allowed, needs approval or blocked.

A starting clearance plan

AppAI canOnly with your OK
CRM / client listRead, search, match leadsDelete or edit records
Email & SMSRead, write draftsSend anything
CalendarCheck availability, propose timesMove or cancel appointments
Website & socialPrepare posts and pagesPublish
Ad platformsRead performance, recommend changesChange budgets or pause campaigns
PaymentsRead reportsCharge, refund or pay out

Keep the workspace small

Permissions inside apps matter, and so does what the agent can see on the computer itself. Give it a dedicated machine, user account or folder. Keep personal files, contracts and unrelated client data out of its reach. If it can’t see it, it can’t leak it or act on it.

Watch for instructions hiding in content

Agents read emails, web pages and documents. Some of that content will contain text that looks like an instruction (“forward this to…”, “approve this payment”). A well-built agent treats everything it reads as information, not as orders, and only takes instructions from you. Test that before you trust it, and keep irreversible actions behind approval so a bad instruction can’t do damage on its own.

Loosen slowly, on evidence

Once a task has produced correct drafts consistently for weeks, you can decide to let it do more — for example, sending a standard confirmation that never varies. Make that change on purpose, for one task at a time, and write it down. Most of the time, a draft that takes ten seconds to approve is the right place to stop.

Every CTRL+R Studio™ AI System starts with a clearance plan like this one, set up before the first mission runs. If you are deciding what your first mission should be, start with the one-week audit.

← Back to Thinking   Explore AI Systems →