What are AI agent permissions?
When you connect an AI agent to your business accounts, you are deciding what it is allowed to do. Can it only read? Can it reply? Can it post on your Instagram, or connect new accounts? Those choices are its permissions. Here is what they are in plain words, how to set them sensibly, and why giving each agent its own is one of the simplest ways to stay in control.
In one sentence
AI agent permissions are the list of things an agent is allowed to do with your accounts, such as reading messages, replying, posting or connecting new accounts, and everything not on the list is refused. Giving each agent its own permissions, accounts and daily share means you can see what each one does and stop one without stopping the rest.
In more detail
Think of keys to a building. The cleaner has a key to the front door and the store cupboard. The manager has keys to every room. Nobody gets the safe key unless they need it. Permissions work the same way: each agent gets the keys for its job and no more.
Permissions are usually grouped by kind of action. Reading is the gentlest: the agent can see messages but not send anything. Replying lets it answer conversations. Posting and commenting lets it publish things everyone can see. Connecting accounts lets it add new ones. Some services add others, such as running ads.
An agent that tries to do something it is not allowed to is simply refused. Nothing is sent, and the agent is told it needs a permission it does not have. That is useful in itself: if you see an agent being refused something, it is trying to do more than you planned.
Permissions are not only about what an agent can do, but where and how much. You can limit an agent to certain accounts, so your shop helper cannot touch your personal LinkedIn. And you can give it a share of each day's sending, so one busy agent cannot use up an account's whole allowance.
The reason to give each agent its own access, rather than one set shared by all, is visibility and control. If something goes wrong, you can see which agent did it, pause that one alone and change its access without disturbing the others. The table below shows a small business with three agents. Names and details are examples.
An example
| Agent | What it is allowed to do |
|---|---|
| Shop agent (an always-on agent on the shop computer) | Read and reply on the shop WhatsApp and Instagram. No posting, no new accounts. Half of each day's sending. |
| Owner's chat app (Claude on her laptop) | Read, reply, post and comment on every account. Used when she is at her desk. |
| Bookkeeper's helper (ChatGPT) | Read only, on the business email account, to find invoices. It cannot send anything. |
The main permissions, in plain words
| Permission | What it allows | Who usually needs it |
|---|---|---|
| Read | See conversations, messages and account status | Every agent |
| Reply | Answer conversations and start new ones | Agents that handle customers |
| Post and comment | Publish posts, comment and react, which everyone can see | Only agents you trust with your public voice |
| Connect accounts | Add new accounts to your workspace | Rarely; usually you do this yourself |
| Ads | Work with your ad accounts | Only an agent set up for advertising |
Start with less than you think you need. It is easy to add a permission when an agent is refused something you do want it to do, and much harder to undo a public post it should never have made.
Setting up permissions, step by step
- List your agents and what each one is for, in one sentence each.
- For each, tick only the permissions that sentence needs. A helper that finds invoices needs reading, nothing more.
- Choose which accounts each agent may use. Leave out accounts it has no reason to touch.
- Decide each agent's share of the day's sending, if more than one agent sends from the same account.
- Test it. Ask each agent to do its job, and check it cannot do something outside it.
- Review once a month. Remove agents you no longer use, and trim permissions that turned out to be unneeded.
A worked example: adding a second agent
Here is how one owner added an always-on agent alongside the chat app she already used, one step at a time. Names and details are examples.
| Step | What she did | Why |
|---|---|---|
| 1 | Gave the new shop agent its own access instead of copying her own | So she could see and pause it on its own |
| 2 | Ticked reading and replying only | Its job is answering customers, not posting |
| 3 | Chose the shop WhatsApp and Instagram, not her LinkedIn | It has no reason to touch her personal network |
| 4 | Gave it half of each day's sending on those accounts | So she still had room to send from her chat app |
| 5 | Asked it to hold replies as drafts for the first week | So she could read how it wrote before it sent alone |
| 6 | Checked its activity every evening, then weekly | To catch surprises early, then less often as trust grew |
After a fortnight the drafts rarely needed changing, so she let the agent send on its own. When it once tried to like a customer's post, it was refused because it had no permission to react, which was exactly what she wanted.
What if...
- An agent says it is not allowed to post? It was refused because it lacks that permission. Add it only if you want that agent to post.
- An agent runs out of sending for the day before lunch? It has used its daily share. Give it a bigger share, or move some work to another time.
- You stop using an agent? Remove its access entirely, rather than leaving it unused.
- An agent should manage other agents? Some set-ups allow one trusted agent to see and pause others. It still cannot give itself more access than you gave it.
- You think an agent's access has leaked? Pause that agent straight away, then remove its access and set it up again.
Common mistakes
- Giving every agent every permission "just in case".
- Sharing one key or sign-in between several agents or people.
- Letting a read-only job, such as finding invoices, have permission to send.
- Forgetting old agents, which keep their access long after you stopped using them.
- Not checking what an agent actually did. Look at its activity now and then, especially in the first weeks.
Why it matters for your business
- A mistake by one agent stays small, because it can only do what you allowed.
- You can see which agent did what, which makes problems quick to fix.
- One agent can be paused or removed without stopping the others.
- Public posts and new account connections stay with agents you trust most.
- Daily shares stop one busy agent from using up an account's allowance for everyone.
How Spaces handles it
In Spaces, every AI agent that signs in or uses a key has its own permissions, its own accounts, its own daily share and its own pause. You set them under Your agent in your dashboard. The permissions are reading, replying, posting and commenting, connecting accounts, and ads. If an agent tries to post, comment or like without permission, it is refused and nothing happens.
An agent's daily share is a part of each account's daily sending. When an agent has used its share, it is told to wait until the next day, and you can raise its share under Your agent if you want it to do more. The account limits always apply on top, so all your agents together can never send more than an account's safe limit.
A paused agent is refused everything, reads included, and its scheduled messages and routines wait. If you give one trusted agent permission to manage the others, it can see them, check their activity, change or pause them, and remove them. It cannot change its own access, give another agent anything it does not have itself, or hand on the power to manage agents.
In short
Permissions decide what an AI agent may do with your accounts. Give each agent its own, keep them as small as the job allows, and you stay in control even as you add more agents.
When something surprises you, check which agent did it, pause that one, and adjust its access. That is far easier when every agent has its own.
FAQ
What permissions should a new AI agent have?
Only what its job needs. For most customer reply agents, that is reading and replying on the accounts it looks after. Add posting or other permissions later if you find you want them.
Can I give an agent read-only access?
Yes. A read-only agent can see conversations and account status but cannot send, post or change anything. It suits jobs like finding information or writing summaries.
What happens when an agent tries something it is not allowed to do?
It is refused, and nothing is sent or changed. The agent is told which permission it needs, so you can decide whether to give it.
What is a daily share?
It is the part of an account's daily sending that one agent may use. It stops one agent from using up the whole allowance when several agents share an account.
Can I limit an agent to one account?
Yes. In Spaces each agent has its own accounts, so you can keep, for example, a shop agent away from your personal LinkedIn.
Can an agent give itself more permissions?
No. Even an agent you allow to manage other agents cannot change its own access or give others more than it has.
How do I see what an agent has done?
Look at its activity. In Spaces, an agent with permission to manage others can show you each agent's activity, and the conversations themselves show what was sent.
Do permissions replace the sending limits?
No. Permissions decide what an agent may do. The account limits decide how much and how fast. Both always apply.