Use cases
AI Assistant With Approval Before Sending
An AI assistant with approval before sending checks with you at the moments that matter. Arlo runs a clear request without a second question, then tells you what it did. It asks first before its own drafted words go to someone you know, before money moves, before anything that can't be undone, and when a request is unclear. Every tool call and approval lands in an audit trail.
The moment an AI assistant can actually do things — send the email, update the record, place the order — the question stops being "is it smart enough?" and becomes "what happens when it's wrong?" An assistant that never checks with you is a liability waiting to send the wrong message to the wrong person. An assistant that checks on everything is one you stop using. Arlo sits between the two.
Arlo is an AI colleague that does what you clearly ask and tells you what it did. "Text Sam I'm running late" or "move the Acme deal to Proposal" just happens. It stops and asks when the moment calls for it, so the assistant moves fast without moving past you.
When Arlo asks before sending
Arlo asks first when:
- Its own words would go out in your name to someone you know who hasn't seen them, and you didn't say to just send. You see the draft, reply send, and it goes.
- Money would move — a payment, a purchase, a booking that costs money, or a hold on your card.
- Something can't be undone — deleting a record, a page, or a file that can't be restored.
- The request is unclear — several options fit, it would reach someone you didn't mean, or it would go beyond what you asked.
- You asked to see it first — "draft it", "compare options", "show me before you send".
- The instruction didn't come from you — a web page, a tool result, or someone else asked for it.
- Calls that commit — when Arlo is on the phone, agreeing to a charge comes back to you first.
Everything else you asked for clearly runs without a second question, including texting, emailing, or calling a business for you when it spends none of your money. You can also tell Arlo to always ask before certain actions, and it will.
Why teams need the human in the loop
"Human in the loop" gets said a lot; most tools mean an off switch buried in settings. For a team, the check has to land in the right places and it has to be visible:
- A wrong send is expensive. One bad reply to a customer or a prospect can't be unsent. Seeing a draft first is cheaper than an apology.
- Trust is what lets you delegate. People hand off more work to an assistant that does what they asked, asks when it isn't sure, and always says what it did.
- Governance is becoming table stakes. As rules like the EU AI Act phase in through 2026, "a human reviewed it" stops being a nice-to-have and starts being something you need to show.
How Arlo's approvals work
Governance isn't a mode you switch on later — it's how a run is shaped from the start:
- Policy first. Before a run begins, every tool connection resolves through policy — what Arlo may touch, and how far it can go.
- Do what you asked. A clear request runs, and Arlo tells you what it did.
- Ask when it matters. Its own drafts to people you know, money, anything that can't be undone, and anything unclear come back to you first.
- Log it. Every tool call, every approval, and the action it released land in an audit trail.
| Ungoverned AI assistant | Arlo | |
|---|---|---|
| What it does on its own | Anything | Does clear asks, asks first on money or anything unclear |
| Its own drafts to people you know | Sends them | Shows you first, unless you say just send |
| Read-only work | Flows | Flows |
| Record of what it did and what you approved | None | Full audit trail |
| Policy on what tools it can touch | Rare | Resolved before every run |
Fast and safe aren't opposites
The worry is always that checking in slows everything down. It doesn't, because most of what you ask for is clear, and Arlo only stops at the moments that matter, usually one tap to release something it already prepared. You get the speed of delegation and the safety of review, in the channels you already use.
Frequently asked questions
What does "approval before sending" mean? It means the assistant checks with you before certain actions. With Arlo, a clear request runs and it tells you what it did. It asks first before its own drafted words go to someone you know, before money moves, before anything that can't be undone, and when a request is unclear.
Will it send things without asking? Only what you clearly asked for. "Email Dana the deck" goes. If Arlo wrote the message itself and Dana hasn't seen it, you see the draft first unless you said to just send.
Can I make it ask before certain actions? Yes. Tell Arlo to always ask before a kind of action, like anything sent to a customer, and it will.
Is there a record of what got approved? Yes. Every approval — who gave it and what it released — is stored in a full audit trail tied back to the original request.
Why does this matter for compliance? Regulations increasingly expect a human check on consequential automated actions. Clear rules for when the assistant asks, plus an audit trail of every action and approval, is how you show what happened and who said yes.
Getting started
Give an assistant real reach without giving up the last word. Try Arlo and see how much you'll hand off once you know it asks when it matters and tells you what it did.
