Why the Agent asks first

The Agent in SalexHub can create and update records and send email from your mailbox. By default, it asks before it does any of that. Here's why we made asking the default, how you can give the Agent more room, and what always waits for your click.

It works with your permissions#

The Agent uses the same tools as the rest of SalexHub, and it uses them as you. It sees what you can see and changes only what you can change. The autonomy setting doesn't give it any new rights. It only decides which of its steps still need your click.

The settings page puts it this way: "Every action runs with YOUR permissions either way. This only decides what still needs your click."

Four modes#

The organization's mode is the ceiling. You can set a narrower mode for a role, for one person or for a single conversation, and nothing goes past the organization's mode. Out of the box, only the Owner can change it.

Each property has its own rule#

Each property says what AI may do with it: Never change it, Propose a change or Change it. The default is Propose a change. When the Agent creates or updates a record and writes to such a property, that step comes to you as a card, even in Full autonomy. A property set to Never change it, the Agent can't write at all, not even on a card.

So in a new organization, the values the Agent writes to records come to you as cards, whatever the mode. If you want it to keep one property up to date on its own, an admin sets that property to Change it. In a mode that allows record changes, the Agent then writes it without asking, and every other property still asks.

An admin sets it on the property's AI tab, under What AI may do. When the Agent proposes that change itself, it's a card in every mode, Full autonomy included.

What always waits for you#

Some changes wait for your approval in every mode, Full autonomy included. Changes to who can see and do what, such as sharing rules, teams and field access, always come to you as a card.

Some things the Agent can't do at all, even with your approval: erase a person's data, change someone's role, decide an approval, create API keys or change billing. You do those yourself in the app.

Your approval rules apply to the Agent too. If invoices above a set amount need a manager's approval, an invoice the Agent prepares waits for that manager like any other.

What a card looks like#

A card shows what the step will change, with Approve and Decline. When the Agent proposes several changes, they wait together under "3 changes waiting for your approval", with Approve all 3.

In the conversation, each change shows whether you approved it or it ran without a click. In the record's activity, the Agent's changes carry a "via the agent" badge.

If an answer goes wrong, Undo takes back its record changes in one step. It archives the records the Agent created, restores the values it changed and brings back what it archived. It skips any record someone has edited since, and it can't recall an email that has already been sent.

Why asking is the default#

Your team and your customers rely on your records. A wrong amount on a quote or a deal in the wrong stage costs more than a click, and often a customer is the one who notices. Asking first is slower on day one. After that it saves time, because you don't have to go back and check what happened while you were away.

As you come to trust the Agent, you can give it more room, one mode or one property at a time. Whatever it does on its own is still recorded, in the conversation and in the record's activity.

The full rules, with every mode and every exception, are in Autonomy and approvals.