← All posts
AI agents

What should an AI agent be allowed to do without asking me first?

An AI agent is only useful if it can act, and only safe if it knows where to stop. A simple starting setup for a small business: what to let it do alone, what it should ask first, and what to keep for yourself.

WakatoNovember 21, 2026
What should an AI agent be allowed to do without asking me first?

Short answer: let an agent do on its own anything you could undo in a minute, such as drafting, sorting, looking things up and answering routine questions. Make it ask first before anything that spends money, speaks to someone new in your name, or cannot be undone. Give it a spending cap either way.

Why this question matters

An agent that asks about everything is just a slower way to do the work yourself. An agent that asks about nothing will one day send the wrong email to your whole list. The useful setup sits in between, and it is easy to get right if you sort tasks into three groups.

Group 1: let it work alone

  • Reading and looking things up: contacts, orders, your calendar, your documents.
  • Drafting: emails, posts, pages, replies. A draft changes nothing until it is sent.
  • Filing and tidying: tagging contacts, sorting receipts, updating a task.
  • Answering routine questions from customers using your own information.

The test is simple: if it gets one wrong, can you fix it quickly and cheaply? If yes, it does not need to ask.

Group 2: make it ask first

  • Spending money: launching an ad, buying a phone number, a paid service.
  • Writing first to someone who has not written to you.
  • Sending anything to many people at once, like a campaign.
  • Publishing: putting a page, a post or a product live.
  • Deleting things, and refunds.

Asking should take you one tap. A good request says exactly what will happen, such as "Send these 10 follow-up emails from hello@yourdomain.com?", so you can say yes without opening anything.

Group 3: keep for yourself

Some things an agent should not see at all. If it only handles customer messages, it has no reason to open your books. Hiding an area is safer than trusting an agent not to touch it.

Always set a spending cap

Agents do their work with AI models, and that work costs money per use. A cap by the day, week or month means a mistake or a loop stops by itself. Start low. An agent that reaches its cap tells you, and raising it takes a moment.

How this works on Wakato

Each Wakato agent has one screen called "What it may do". Area by area (contacts, email, pages, shop, ads and so on) you pick how far it goes, from hidden to full, and tick the actions it must ask you about first. You can also name the exact tables, pages or inboxes it may use.

When it reaches something it must ask about, a card appears in its chat: Yes, Yes and don't ask again, No with a note, or Not now. If it needs a permission it does not have, it asks for that with a card too. Every agent has its own daily, weekly or monthly spending cap.

A starting setup you can copy

  • Pick a role that matches the job, such as Email manager or Bookkeeper, instead of full access.
  • Leave the suggested "asks first" items ticked for the first two weeks.
  • Set a small daily cap.
  • After two weeks, turn off the questions you always answer yes to.

See how agents work on Wakato, or start free.

The only app you need, run by AI agents.

Website, shop, inbox, phone, ads, social, email, calendar and books in one place. Free to start.

Start free