Short answer: no, outgoing messages from an AI prospecting agent aren't approved one by one. The Formula. agent prospects on its own from a template you approve once, adapts it to each prospect, and keeps improving it.
Human control shows up somewhere else: when a prospect replies. That's where you approve or edit the reply with a swipe, in under 5 minutes. It's a shift in where control sits, not a removal of control.
This is one of the questions we get asked the most, and it makes sense: most prospecting tools people already know work the other way around.
They let you qualify, draft, then click "send" yourself, one message at a time.
An agent isn't that. Here's why, and what it actually changes for you.
Why would you assume every message needs approval?
Because that's the model most people learned prospecting tools by.
Semi-automatic tools built around scheduled sequences leave the user in the loop on qualification and drafting. You write or approve a text, then you click.
That feels reassuring on paper. It's also what makes those tools slow at scale: the limiting factor stays your own time, message after message.
An autonomous agent changes that mechanic at the root. The question stops being "who clicks send" and becomes "who sets the strategy".
So how do outgoing messages actually go out?
You approve a prospecting message template once, sometimes twice if you want to test two angles.
From there, the agent runs its own A/B testing on that template: it improves every day by picking the best message for the right prospect.
It doesn't send the same text to everyone. It reads each prospect's LinkedIn profile and recent posts, and adapts the message accordingly.
Concretely, the agent sends 28 LinkedIn invitations per day, alongside a parallel email follow-up on the same qualified prospects. Over a month, that's roughly 840 prospects worked, none of them waiting on your individual approval.
Measured results on active client accounts: a 30 to 60% reply rate depending on sector and offer, versus 7 to 11% for classic manual outreach.
So where does the swipe actually happen?
The swipe happens when a prospect replies, not before.
At that point, the agent drafts a reply. You approve or edit it with a swipe on your phone, before it sends.
The prospect gets your reply in under 5 minutes, which multiplies conversion odds by 100 compared to a reply sent the next day.
"The agent prospects on its own, message after message. The swipe isn't there. It shows up when a prospect replies, to approve the conversation that's starting."
Laura Terriou, co-founder of Formula.
For the full picture of what happens between a prospect's reply and a meeting landing on your calendar, we broke it down in this article on what happens when a prospect replies.
Why this setup instead of approving everything?
Because approving every outgoing message would kill the one lever that makes the whole system worth it: speed.
A sales rep already spends less than 30% of their time actually selling, according to Salesforce. Adding manual approval on every opening message would eat further into that time, without improving reply rate at all.
Opening outreach doesn't need your case-by-case judgment. It needs a good template, real automatic personalization, and consistency in sending.
Your judgment, on the other hand, matters a lot the moment a real person writes a real reply. That's exactly why control sits there, and not upstream.
Approving every send vs. swiping on replies: the comparison
| Criteria | Approving every send | Swipe on replies (Formula.) |
|---|---|---|
| Human time required | Several hours per week | A few swipes a day, on replies |
| Monthly volume reachable | Capped by your available time | ~840 prospects/month across 2 channels |
| Reply speed to a prospect | Depends on your availability | Under 5 minutes, 24/7 |
| What you control | Every message, one by one | The starting template + every reply sent |
A concrete example: agencies and consulting firms we work with went from $20,000 to $55,000 in monthly revenue while cutting about 2 hours of prospecting per day, precisely because nothing was waiting on their click on outgoing messages anymore.
The full breakdown of that case is in our complete case study.
What the agent does, step by step
Sign up and I'll send you the full breakdown of a message's journey, from the approved template to a swipe on a reply.
Do you actually stay in control?
Yes, on what actually matters. You choose and can change the starting template anytime.
You see your stats continuously: invitations sent, reply rate, ongoing conversations.
And most importantly, no reply goes out without passing through your swipe. This isn't a full autopilot system. It's a system that only removes tasks where your case-by-case judgment added nothing, and keeps you exactly where it matters: the conversation with a real person who just replied.
Want to see the template + swipe running on your own prospects?
Start the 2-week trialFirst replies usually land within the trial. Judge on evidence, not promises.
Frequently asked questions about approving AI agent messages
Does an AI prospecting agent send every message without approval?
Yes, for outgoing messages. The Formula. agent prospects on its own from a template you approve once, adapts it per prospect by reading their LinkedIn profile and posts, and keeps improving through ongoing A/B testing.
So what is the swipe actually for?
The swipe kicks in when a prospect replies. The agent drafts the reply, you approve or edit it with a swipe, in under 5 minutes.
Why not approve every message just to be safe?
Because it would kill reply speed, the system's most profitable lever. Sales reps already spend less than 30% of their time actually selling: adding manual approval on every send would eat further into that time without improving reply rate.
Do I still stay in control of what gets sent?
Yes, on what matters: you approve and can change the starting template anytime. Message-by-message control happens on inbound replies, where every swipe stays editable before it sends.