Two years ago, an "AI" prospecting tool meant ChatGPT plugged into your templates, varying the first lines of your emails. That was useful. That was not an agent.
An agent is not text that generates faster. It is a system that decides. It chooses who to contact, when, with what, and adjusts its decision based on the feedback it observes.
Here is how it works concretely, step by step. No buzzwords.
Step 1: Identifying the right prospects (the "learning targeting")
An agent does not receive a static list. It receives your offer, your positioning, and the characteristics of an ideal client: role, company size, sector, behaviors observable on LinkedIn.
From there, it scans LinkedIn continuously. Not once a month. Continuously. It watches who posts on topics related to your offer, who changes jobs, who announces a funding round, who is hiring. It weights these signals and builds a fresh list every day.
In practice for Formula.: a business coach who helps creative agencies structure their offer does not need to go to Sales Navigator. The agent identifies every day the agency founders who just posted about "too many projects and not enough structure" - that is an explicit signal that the pain is acute right now.
The difference vs Sales Navigator: Sales Nav shows you who matches criteria. The agent shows you who matches and has a buying signal this week.
Step 2: Writing messages that do not look like templates
This is where most "AI" tools fall short. Lemlist + ChatGPT generates 50 personalized messages per day, but they all share the same structure: "Hi {first name}, I saw your post about {topic}…". The market has learned to recognize this format in 30 seconds.
An agent works differently. It analyzes the prospect's entire profile, their last 10 posts, their tone (ironic, technical, measured), the comments they leave on other posts, the way they sign their messages. Then it writes in their own tone. Short or long. Direct or warm. With or without emojis.
The result: a message the prospect does not recognize as automated. Not because we are tricking anyone. Because it was genuinely written for them.
Step 3: Following up at the right time, not every 3 days
Classic tools have a fixed logic: Day 0, Day 3, Day 7, Day 14. Stop. That worked in 2022. It does not work in 2026 because everyone runs the same sequence.
An agent watches what the prospect does after your first message. Did they view it? Did they like one of your posts in the following days? Did they visit your profile? Based on the signal, the agent follows up at the right moment - not according to a dead calendar.
A prospect who visited your profile twice this week without replying does not need a generic follow-up. They need a message that opens a specific door: a comment on their latest post, a question, an immediately valuable resource.
Step 4: Preparing the reply you will send
What changes everything is what happens when the prospect finally replies. With Lemlist, you receive their reply, you think, you write, you send. That takes 10 to 15 minutes per reply. And you do this 5 to 20 times a day if things are going well.
With an agent like Formula., when the prospect replies, your response is already written. Matched to their profile, their message, their tone. You read it in the app, adjust one word if you want, swipe to validate, done.
The gesture we all know from dating apps - except at the end of the feed, there is a contract.
Comparison: AI agent vs classic prospecting tool
| AI Agent (Formula.) | Classic tool (Lemlist, Waalaxy) | |
|---|---|---|
| Identifies prospects | ✓ Continuously, intent signals | You import a static list |
| Personalizes messages | ✓ In the prospect's own tone | Variables inside a template |
| Decides when to follow up | ✓ Based on observed signals | Fixed schedule (Day+3, Day+7…) |
| Prepares your reply | ✓ Pre-written, ready to validate | ✗ You write manually |
| Books meetings | ✓ Automatically in your calendar | ✗ You schedule manually |
| Setup | 30-min onboarding | Several hours, multiple sequences |
| Daily time required | 3-5 min of swiping | 1-3h of management |
What an agent does NOT do (the honest part)
An agent does not guarantee a close. It identifies, contacts, qualifies, prepares. The final conversion depends on your offer, your price, your fit with the prospect. If your offer is unclear, the agent will not save it.
An agent also does not replace content. If you have zero LinkedIn presence, the agent works less well - because prospects it contacts will check your profile, and an empty profile kills conversion. You need a minimum of credibility signals.
Finally, an agent is not magic. It calibrates in seconds (ingesting your site, LinkedIn, conversations, posts, and competitor analysis). First results can arrive within the first week.
Want to see an agent in action?
Sign up for the newsletter and receive next Tuesday the full screenshot of a client case (Camille, design agency founder, 11 meetings in 6 weeks).
Who it is for
An AI prospecting agent makes sense when:
- Your average deal size is above €1,500 (below that, the agent cost does not justify itself)
- You sell B2B, primarily on LinkedIn
- You have a clear offer and an identifiable promise
- You no longer want to spend 10+ hours per week on prospecting
If you check those 4 boxes, the math is simple: the monthly agent cost is recouped with the 1st or 2nd meeting closed. And you recover 30 to 50 hours per month to focus on your offer, your product, your existing clients.