How much should an AI employee do on its own?
Ask me first, Act and tell me, or Fully on its own: how to choose the right level of independence for each AI employee, by what it sends and who reads it.
Decide by asking two questions: who will read what it sends, and what would a mistake cost? If its messages go to people outside your company, or a wrong one would embarrass you, make it ask first. If it reports to your own team and a weak list costs a few minutes, let it act and tell you. If it is a routine internal report you would only read when something breaks, let it work fully on its own. Most teams start strict and loosen the setting once a few weeks of output have earned it.
This article explains the three levels xMap Flows offers for every AI employee that sends something, what each one does behind the scenes, and how to pick between them with examples from sales, operations and reporting.
The three levels
| Level | What happens before it sends | What you hear afterwards |
|---|---|---|
| Ask me first | Each email or message waits for your OK; you can tick which items go and rewrite drafts | Only what you approved goes out |
| Act and tell me | It sends on its own | A note in its Team inbox thread saying what it sent |
| Fully on its own | It sends on its own | Nothing, unless something goes wrong: a failed run, a budget pause, an approval it needs |
The three choices for an AI employee that sends emails, messages, reports or data to another system.
Behind the scenes, Ask me first places an Approval step before each sending block, set so every item is approved on its own and any message the AI drafted can be rewritten by the person approving. Switching to another level removes only the approval it added; approval steps you placed yourself stay where you put them.
What stays the same at every level
- It stops when its monthly budget runs out, whatever the level.
- A test run never sends anything. Only Run once and its deployed runs act for real.
- Employees that answer conversations, such as a chat page or replies on a messaging app, answer at once. Holding a reply for approval would leave the person waiting, so Ask me first does not apply to them.
- Only the employee's owner or an admin with the Deploy permission can change the level.
Matching the level to the job
Messages to people outside the company: Ask me first
Say a beverage distributor in Lima has an employee that finds new cafes each week and drafts an introduction to each owner, then follows up twice through Outreach. Every message carries the company's name to a stranger. Ask me first lets the sales lead read each draft, fix a wrong name or a clumsy line, untick the cafe that is clearly a chain's flagship, and send the rest. The approved words are exactly the words that go out.
Lists for your own team: Act and tell me
A logistics company in Johannesburg has an employee that emails depot managers the road closures near each depot before the morning shift. The readers are colleagues, the list is useful even when imperfect, and waiting for an approval at 05:30 would make it late. Act and tell me sends on time and leaves a note in the Team inbox, so the operations lead can see at a glance what went to whom.
Routine reports nobody needs to watch: Fully on its own
An analyst in Seoul gets a weekly PDF of competitor openings through Make a report. It is read on Monday with coffee; nobody wants a note that it was sent. Fully on its own keeps the inbox quiet and still raises a hand when a run fails or the budget pauses it.
A simple way to decide
- List what the employee sends and to whom. Anything that leaves the company goes to the strict end.
- Ask what one bad message would cost: a confused customer, a lost prospect, or a colleague who skims past a weak line.
- Start one level stricter than you think you need.
- Watch two or three weeks of approvals or notes. If you change almost nothing, move it one level looser.
- Move it back whenever the job changes: a new city, a new kind of message, a new audience.
Where to set it
You meet the choice the first time you deploy an employee from your browser, in the Ready to go live? window, and you can change it at any time in its panel: click its picture on the Flows page. Its card shows which level it is on. Deploying stays one click whichever you pick, and sending is always switched on when you deploy; the level decides what happens before and after each send, not whether it can send at all.
Mistakes to avoid
- Putting outreach on Fully on its own on day one. Even well-written instructions meet odd data: a business listed under a person's name, a closed shop still listed as open.
- Leaving a daily internal alert on Ask me first. An alert that waits for approval arrives after the event it warns about.
- Using the level instead of a filter. If most items you are asked to approve are wrong, the fix is a sharper rule or better scoring instructions, not more approving.
- Forgetting the budget. Independence and spending are separate. A fully independent employee still needs a sensible monthly budget.
For the details of approving item by item and rewriting drafts, see when to make an AI employee ask before it sends. For spending, see setting a monthly budget for each AI employee.
Questions
Which level does a new AI employee start on?
An employee whose flow already contains an Approval step starts on Ask me first; otherwise it starts on Act and tell me. You pick in Ready to go live? before its first deploy and can change it later in its panel.
Can a colleague change the level of my AI employee?
Only the employee's owner or an organization admin with the Deploy permission can change it.
Does Fully on its own mean nobody is told anything?
It means no note after each send. You are still told when a run fails, when its monthly budget pauses it, or when it waits for an approval you added yourself.
Does Ask me first slow down a chat assistant?
No. Employees that answer conversations reply at once, because a person is waiting; the setting applies to employees that send on a timetable.
Templates that do this
Read next
More on ai employees and automation · Every article · Templates