What is an always-on AI agent?

In one sentence

An always-on AI agent is one that keeps running by itself, watching for new messages and acting on them without you typing anything, while a chat app such as ChatGPT, Claude or Grok only does something when you ask it in a conversation. Both can use the same connected accounts and the same safety limits; the difference is whether someone has to be there to start each job.

In more detail

Think of a chat app as a very capable member of staff who sits at a desk and waits for you to walk over with a task. You say "read my new WhatsApp messages and draft replies", and it does exactly that, then waits again. If you go home, the desk stays quiet until you come back.

An always-on agent is more like someone working a shift. Once it has its instructions, it keeps checking for new work, handles what it is allowed to handle, and leaves notes for you about the rest. It does not need you to open a window or type a request each time.

The tools are not tied to one or the other by magic. What decides it is where the agent runs and how it is set up. Chat apps such as ChatGPT, Claude on the web or desktop and Grok run when you have a conversation open. Agents such as Claude Code, Codex, OpenClaw and Hermes can be left running on a computer or a server, so they can keep going after you close the laptop lid, as long as the machine they run on stays on.

Being always on is not the same as being in charge. A well set up always-on agent follows the same notes you would give a chat app, works inside the same limits, hands awkward conversations to a person and can be paused in one step. It is simply there more of the time.

Most small businesses start with a chat app because it is easier to watch. You see every draft, you learn how the agent writes, and you build up the notes it works from. Moving to an always-on agent makes most sense once those notes are good and you trust how the replies read. The example below shows the same evening in a florist with each kind of agent. Names and details are examples.

An example

Chat app or always-on agent: the difference at a glance

Neither is better in every case. Plenty of owners use a chat app in the day, when they are around to read drafts, and leave an always-on agent to handle a short list of simple questions overnight.

What an always-on agent actually does all day

Behind the scenes, an always-on agent repeats a simple loop. It is worth knowing the steps, because each one is a place where you can set a rule.

1. It waits for something new, such as a message, a comment or an account needing attention.
2. When something arrives, it reads it and checks its notes: who this is, what they want and whether it is allowed to answer.
3. It replies, holds the reply as a draft for you, or hands the chat to a person, depending on your notes.
4. It records what it did, so you can read a summary later.
5. It goes back to waiting.

The waiting step matters more than it looks. An agent that keeps asking a platform "anything new?" every few seconds behaves like a bot and can put an account at risk. A good set-up waits for a signal that something has happened instead, so it can stay ready all day without pestering anyone.

Sample notes for an always-on agent

An always-on agent needs clearer notes than a chat app, because you are not there to correct it in the moment. Here is a short example for a small shop. Names and details are examples.

You answer customer messages for Bloom Florist on WhatsApp and Instagram.

Answer on your own:
- Opening hours, delivery days and delivery areas (see the list below).
- Whether we can do a bouquet for a date, using the order book notes.

Hold as a draft for the owner:
- Anything about prices over £80, discounts or refunds.

Hand to a person straight away:
- Complaints, damaged orders, anything about a funeral or illness.
- Anyone who asks for a person.

After 21:00, reply only to people who wrote first, keep it short and say the
owner will follow up in the morning. Never start new conversations at night.
Every morning at 08:00, write a short summary of what you answered and what
is waiting for the owner.

Notice how much of this is about what not to do. That is normal. The more an agent runs on its own, the more its notes should say where the edges are.

Is an always-on agent right for you? A quick check

• You get messages outside your opening hours that a quick, accurate answer would help.
• Most of those messages are simple, such as hours, prices on a set list, availability or directions.
• You have used a chat app with the same notes for a few weeks and the drafts rarely need changing.
• You, or someone helping you, can keep a computer or small server running and check it now and then.
• You are comfortable with the agent handing tricky chats to a person rather than trying to solve everything.

If most of these are true, an always-on agent is worth trying. If they are not, stay with a chat app for now, or use simple set replies for after hours while you build up your notes.

Common mistakes

• Switching on an always-on agent before the notes are tested. Run the same notes in a chat app first and fix what goes wrong.
• Letting it start new conversations overnight. Replies to people who wrote first are far safer than unprompted messages.
• Forgetting the computer it runs on. If the machine sleeps or restarts, the agent stops, so check it is still running.
• Never reading what it sent. For the first few weeks, read the day's conversations every morning.
• Giving it every permission. An agent that only answers messages does not need to post, comment or connect new accounts.
• Having no quick way to stop it. Know where the pause switch is before you need it.

Why it matters for your business

• Customers who write in the evening or at weekends can get a helpful answer instead of silence.
• You decide how much the agent does alone, from answering everything to holding every reply for you.
• The same limits and safety rules apply whichever kind of agent you use, so moving between them does not change the risk to your accounts.
• Starting with a chat app lets you learn how the agent writes before it works unwatched.
• Knowing the difference stops you expecting a chat app to answer while you sleep, which it will not do.

How Spaces handles it

Spaces works with both kinds. A chat app such as ChatGPT, Claude or Grok reaches your connected WhatsApp, Instagram, LinkedIn, Messenger, X and email accounts through the Spaces MCP server when you ask it to. An always-on agent such as Claude Code, Codex, OpenClaw or Hermes uses the same server and can keep working while it runs. The agents page shows how to add Spaces to each one.

For waiting, Spaces gives agents a way to hear about new messages without polling. The agent asks Spaces to wait for the next event, and Spaces answers the moment something new arrives on a connected account. That waiting never touches the platforms, so an always-on agent can stay ready all day without making your accounts look like bots. An agent running on a server can also have each event sent to its own web address.

Every agent gets its own permissions, its own share of each day's sending and its own pause. Every send is paced and counted against the account limits, whichever agent sends it. If you want an always-on agent to write but not send, ask it to hold replies as drafts, and nothing goes out until you approve it. Pause everything in Settings stops every agent at once.

In short

A chat app acts when you ask. An always-on agent keeps working while it runs. Both can be safe and useful, and the right choice depends on how much you want to be in the loop.

Start with a chat app, write good notes, and move the simple, well tested jobs to an always-on agent once you trust them. Keep the pause switch and the hand-over rules close at hand either way.

Will ChatGPT or Claude answer my customers while I am asleep?

Not on their own. Chat apps act when you ask them in a conversation, so messages wait until you next open the chat. For replies while you are away, use an always-on agent or simple set replies.

Which agents can run all the time?

Agents such as Claude Code, Codex, OpenClaw and Hermes can be left running on a computer or server. They keep working while that machine is on and the agent is running.

Do I need a server for an always-on agent?

You need something that stays switched on. That can be a computer in the shop that never sleeps, or a small server. If the machine turns off, the agent stops until it is started again.

Is an always-on agent riskier than a chat app?

It can be, because replies may go out before anyone reads them. You lower that risk with clear notes, fewer permissions, holding sensitive replies as drafts and reading the conversations every day at first. The sending limits are the same for both.

Can I use a chat app and an always-on agent together?

Yes. Many owners use a chat app during the day and an always-on agent for a short list of simple questions overnight. Give each agent its own access so you can see and pause them separately.

How does an always-on agent know a new message has arrived?

A good set-up waits for a signal rather than checking the platform over and over. In Spaces, the agent waits on an event stream that answers as soon as something arrives, and that waiting never touches the platforms.

How do I stop an always-on agent quickly?

Pause it. In Spaces you can pause one agent, hand a single chat to a person, or use Pause everything in Settings to stop every send at once. Reading keeps working while sending is paused.

Should my always-on agent start new conversations?

It is safer to keep it to replies, especially at night. New conversations carry more risk on every platform and have tighter limits. Leave those to times when you are around to check.