Proposals you accept daily without changes can go to the AI. On the “Trust” tab in “Pulse”, “Management” sets a type of action to “Act on its own”. The AI then does it at once, without a proposal, and everything shows in the log.
When you need this:
- You accept contacts from calls every day without changes: let the AI add them by itself.
- A manager wants the AI to schedule follow-ups without them: “Management” turns trust on, for the whole company.
- An agent keeps proposing the same product edits: trust goes to that agent only.
Where to start: “Pulse” → “Trust”.
What trust gives you
- The AI does routine work by itself: for the chosen type of action, without a proposal.
- Shows why it is safe to trust: 30-day counts of accepted and rejected proposals of the type.
- Holds itself back: no more than 20 identical actions per hour, and the Controller checks emails.
- Switches back with one toggle, and what was done can be undone.
How to trust the AI with a type of action
The “Trust” tab lists types of actions: “Create a deal”, “Plan a follow-up”, “Add a contact to the card” and others. Each is in “Ask” mode (the default) or “Act on its own”. In the second mode, the AI acts at once, without a proposal; the action appears on “Done by AI” labeled “Done by AI”.
1
Company-wide trust note2
AI action3
Act on its own switch (off)- Trust applies to the whole company; acting on its own means the AI works without confirmation and shows it on the “Done by AI” tab.
- Each row is one AI action with a description and 30-day accepted/rejected counts.
- The switch in the Mode column; while it is off, the AI only asks.
- Trust applies to the whole company, not to one manager: the AI performs a trusted type by itself, whoever assigned the work. Only “Management” can switch it; everyone else sees grayed-out switches and the caption “Only Management can change trust.”
- Next to each type: proposals accepted and rejected over 30 days. When a type has 10 or more accepted and none rejected, “Management” sees a hint above the list: “…accepted and no rejections in 30 days. You can trust AI with it.” and a “Trust” button. It is shown for one such type, the one with the most accepted; actions that go outside get no hint.
- Turning trust on asks for confirmation, and every switch-on and every action the AI performs by itself is written to the operation log.
The double check mark on a proposal does the same. It runs the proposal and turns the mode on for its type (proposal buttons).
You can switch back to “Ask” with the same switch at any time. What the AI has already done by itself stays; undo it on “Done by AI”.
Customer emails: separate rules
Even with trust, the AI sends only complete email replies by itself: no “I will check”, confidence 0.8 or higher. Complaints, returns, payment for services, account access, bug reports, and emails about partnerships, vacancies or supplier offers always wait for a person. The Controller checks every reply before sending; if it is off, nothing goes out.
1
Reply to a customer email row2
Dialog text: no confirmation, cannot be undone3
Trust button4
Cancel button- The switch of the “Reply to a customer email” row opens the dialog.
- The dialog warns that the AI will act without confirmation for the whole company and that a sent email cannot be undone.
- “Trust” enables the mode; it was not pressed.
- “Cancel” closes the dialog without changes; it was pressed after the shot.
What the AI replies and why it writes “I will check”: AI replies to customers. How the AI closes support tickets by itself: Autonomous support.
Record edits: trusting one agent
Edits to products, customers and documents cannot be turned on from the “Trust” tab. “Management” accepts the agent’s request with the “Accept and do it yourself from now on” button, and this applies to that agent only.
Safeguard: a second such change within an hour goes back to being a request, and “Pulse” shows a proposal to restore the mode. What an agent sees and what it can change: AI employee rights and security.
Safeguards
- Even in “Act on its own” mode, one AI employee performs at most 20 actions of one type per hour. After that, proposals wait for a person again, with the note “Too many identical actions in a row”.
- Customer emails are checked by AI · Controller first; if it is off or out of budget, the email does not go out.
- Markups and promotions change prices in bulk, so “Act on its own” does not apply to them.
Scenarios
- Contacts from calls. You accept contacts every day without changes. “Management” switches “Add a contact to the card” to “Act on its own”; anything unwanted is removed with “Undo” on “Done by AI”.
- Follow-ups without confirmation. For a month, the owner accepts every follow-up from the AI and turns on “Act on its own” for “Plan a follow-up”. Now the AI schedules follow-ups by itself.
- Markups stay with a person. “AI · Pricing” proposes a new markup for a category. The request waits for a person: “Act on its own” cannot be turned on for it.
What to do when…
| What you see | Why | What to do |
|---|---|---|
| The “Act on its own” switches are grayed out | Only “Management” can change trust | Ask your manager, or use “Discuss with AI” |
| There is no “Act on its own” for a markup | A person always accepts markups and promotions | Accept each one separately in “Pulse” |
| “Too many identical actions in a row (20 per hour) — a human decides from here.” | The Controller’s bulk-action safeguard was triggered | Review the proposals and accept them manually (Controller) |
| Emails are trusted, but a reply is waiting for a person | The reply contains “I will check”, the confidence is below 0.8, the reason is on the “always for a person” list, or the Controller is off | Read it and accept it manually |
FAQ
Can a manager trust the AI with actions just for themselves?
No. Trust applies to the whole company, and only “Management” can change it: users with the right to manage AI trust (group 1 by default). A manager can ask their supervisor or discuss the proposal with the AI.
Does trust apply to one employee or to everyone?
Trust on the “Trust” tab applies to the whole company: the AI performs a trusted type of action by itself, no matter who assigned the work. Trust for record edits applies to one agent.
Can the AI reply to customers by itself?
Yes, if “Management” has turned on “Act on its own” for “Reply to a customer email” and confirmed “The AI will send emails to customers without review”. Even then, only a complete email with confidence 0.8 or higher, passed by the Controller, goes out automatically.
Can I turn on trust from the chat?
Yes, with the phrase “let the AI add contacts by itself”. “Management” gets a confirmation of the trust change; others are told by the consultant that trust is changed by “Management”.
How do I switch back to “Ask”?
With the same switch on the “Trust” tab. What the AI has done stays; undo it on “Done by AI”.
Where to find it and who has access
- Where
- Directories → Contacts → AI Pulse → “Trust”; for record edits, “Accept and do it yourself from now on” on the agent’s request
- Who turns it on
- “Management” only: employees with the “AI Pulse: trust” right, group 1 by default
- What cannot be trusted
- markups and promotions: a person always accepts them, one by one
For the overall picture of the window, see “Pulse”.
Related topics
- For the overall picture of the window: “Pulse”.
- To reverse what the AI did: Done by AI: the action log and undo.
- To decide on proposals manually: How to decide on an AI proposal.
- If it is unclear what the AI can change without you: AI employee rights and security.
- To see how the AI checks emails: Controller.
