Short answer: at Formula., the agent prospects on its own: it picks the prospects, writes each message in their tone, sends 28 LinkedIn invitations a day and follows up by email, with no human validation on that outbound side.
The only moment you step in is when a prospect replies. The agent drafts the reply, you approve it with a swipe from your phone.
Zero human in the loop on a commercial conversation that carries your word is a marketing claim, not a best practice.
What does a "fully autonomous" agent actually promise?
In 2026, more and more AI prospecting tools advertise "zero human in the loop" as a selling point.
The pitch is appealing: nothing left to do, the agent handles everything from the first message to the booked meeting.
The catch is that "everything" includes negotiation, replies to specific objections, and sometimes commercial commitments (pricing, timelines, exceptions).
An agent that replies alone to a prospect asking for a discount or a special case is making a commercial decision on your behalf, before you ever see it.
What is actually automated by an AI prospecting agent?
At Formula., the entire outbound side is fully automated. The agent builds its prospect list every day, reads each prospect's LinkedIn profile and recent posts, then writes a message adapted to each one.
A message template gets validated once with you (sometimes twice), then the agent runs its own A/B tests and improves its messages every day, without looping back through you.
It sends 28 LinkedIn invitations per day, the limit documented by LinkedIn, and follows up by email in parallel on the same qualified prospects. Over a month, that is roughly 840 prospects worked across two channels.
None of that goes through human validation. That is the genuinely autonomous part of the system.
Why keep a human in the loop at the reply stage?
Because that is the only moment the conversation becomes two-way, and a prospect can ask something with no generic answer.
A human sales rep spends less than 30% of their time actually selling, according to Salesforce, the rest absorbed by repetitive tasks.
Formula. flips that ratio: the agent absorbs all the repetitive work (targeting, writing, follow-ups), and leaves you only the high-value gesture, the one that carries your word.
"We made the opposite choice from the market: automate everything that is repeatable, and keep a human on the one step that truly puts your commercial credibility on the line. A three-second swipe costs far less than a poorly calibrated reply that loses a qualified prospect."
Laura Terriou, co-founder of Formula.
The swipe: what does it actually look like?
When a prospect replies, the agent reads the full exchange, drafts a reply in the tone of the conversation, and sends it to your phone.
You read it, edit it if needed, and approve or reject it with one gesture. The prospect gets your answer in under 5 minutes, which multiplies conversion by 100 compared to replying the next day.
On the Formula account, as of 07/23/2026: 587 invitations sent, 179 accepted (30%), and a 40% reply rate on the first message (283 messages sent, 114 replies). That is the real swipe volume over a given period, not an abstract number.
See the full mechanism in our full case study.
Full autonomy vs human swipe: the comparison
| Step | Semi-automated tool | "Zero human in the loop" agent | Formula. (swipe on the reply) |
|---|---|---|---|
| Prospect targeting | You, upfront | The agent, alone | The agent, alone |
| Outbound message writing | You (templates) | The agent, alone | The agent, alone |
| Replying to a prospect | You, entirely | The agent, alone, no review | The agent drafts, you swipe to approve |
| Daily human time | Several hours | Zero (and zero safety net) | A few minutes of swiping |
What is the real risk of a fully autonomous agent with no human validation?
The risk is not technical, it is commercial. A poorly calibrated outbound message costs one prospect among hundreds.
A poorly calibrated reply to a prospect already engaged in the conversation costs a meeting that already existed. That is a far more expensive loss.
This is why Formula. made a deliberate choice: automate broadly, but keep the human gesture exactly where it pays off most.
To see how the agent qualifies prospects upstream, before the first message even goes out, our AI prospecting glossary breaks down each step of the cycle.
The full mechanism behind the validation step
Sign up and I'll send you the exact breakdown of a day of swipes on an active Formula account, with the real volumes.
Frequently asked questions about AI agent autonomy
Can an AI prospecting agent be fully autonomous?
Technically yes, but it is not advisable. An agent can send invitations and messages with zero human validation.
The problem shows up when a prospect replies: without oversight, the agent can misread an objection. Formula. fully automates the outbound side and keeps a human swipe only on the reply.
What is the swipe validation at Formula.?
When a prospect replies, the agent drafts a reply matched to the conversation and shows it to you on mobile. You approve, edit or reject it with one gesture.
This is not a validation of outbound messages: those go out on their own, and the agent keeps improving them through continuous A/B testing.
Why not let the agent reply to prospects on its own?
Because a conversation moving toward a meeting involves your commercial word: pricing, timelines, exceptions.
A three-second swipe costs far less than a poorly calibrated reply that loses an already-qualified prospect.
How much time does human validation take per day?
A few minutes, spread through the day as replies come in.
With 28 invitations sent per day and a reply rate of 30 to 60% depending on sector and offer, the swipe volume stays manageable from a phone.
Want to see the autonomous part at work, and keep only the swipe?
Start the 2-week trialFirst replies usually land within the trial. Judge on evidence, not promises.