Rémy Touzard recently shared something that caught my attention: "Manual outreach = 1 meeting/week, 15 hours.

CLAUDE = 2-5 meetings/DAY, 0 hour." And then he doubled down on the part most people miss when they hear "AI outreach": it is not just a tool that sends a first message and a few canned follow-ups.

Rémy Touzard argued that "Most people think AI can only send the first message or basic follow-ups. They're WRONG." What changed everything for him was building an AI system that manages entire LinkedIn conversations - not sequences, but real back-and-forth that handles questions, objections, and qualification until a meeting is booked.

In this post, I want to expand on what Rémy described, because the difference between "automation" and "autopilot conversations" is exactly where most outreach efforts win or die.

The real breakthrough: from sequences to conversations

If you have ever used a classic outreach tool, you know the pattern:

Pull a list
Send a templated connection note
Push prospects into a fixed follow-up sequence
Hope they reply

That workflow can produce meetings, but it also produces a lot of silence, awkward interactions, and brand damage when your "automation" clearly does not understand what the prospect said.

Rémy Touzard's claim is more ambitious: an agent that runs the conversation end to end.

The moment outreach becomes a conversation, your system stops "sending" and starts "selling".

A conversation-capable system does three things a sequence cannot:

It listens and adapts.
It qualifies dynamically instead of asking the same questions to everyone.
It navigates objections in context.

What Rémy Touzard's system is actually doing (step by step)

Rémy outlined a simple set of capabilities that, when combined, produce the "2-5 qualified meetings every day" outcome:

"Scans prospect profiles for personalization signals"
Sends connection requests
"Writes unique first messages"
"Handles ALL replies - questions, objections, qualification"
"Follows up uniquely for each person"

Let me unpack each of these, because the details matter.

1) Scanning profiles for personalization signals

Personalization is not inserting {FirstName}. It is choosing a reason to reach out that makes sense for that specific person.

Examples of usable signals on LinkedIn:

Role and scope (VP Sales at a 50-person SaaS is a different world than VP Sales at a 2,000-person enterprise)
Recent posts and comments (what they care about right now)
Hiring signals (open roles often mean a priority or a growth push)
Tech stack hints (tools mentioned, integrations, job descriptions)
Geography, industry, and buyer context

An AI agent can summarize these into a short "why them" note that guides the first message and the rest of the conversation.

2) Connection requests that do not feel like spam

The goal of the connection request is not to pitch. It is to start a relationship with minimal friction.

A strong connection note is short and specific:

One relevant reason for connecting
No links
No long paragraph
No pressure

This matters because your acceptance rate sets the ceiling for everything that follows.

3) Unique first messages that earn a reply

Rémy emphasized "unique" first messages. That word is doing a lot of work.

A good first message:

References a real signal (post topic, role, hiring, product focus)
States a clear value hypothesis in plain language
Asks a low-effort question

The fastest way to kill responses is to lead with a generic claim like "We help businesses grow". The second fastest is to write something that sounds like it was generated. Ironically, the best AI outreach often reads more human because it is grounded in real context.

4) Handling all replies: questions, objections, qualification

This is the centerpiece of what Rémy Touzard is saying. "All replies" means the system needs to recognize intent and respond appropriately.

Common reply types your agent must handle:

Curiosity: "What do you do exactly?"
Skepticism: "We already have a process for this."
Timing: "Not a priority right now."
Delegation: "Talk to my SDR" or "Email me"
Confusion: "Why are you reaching out?"
Price fear: "Sounds expensive"

A conversation agent needs a playbook for each, plus guardrails so it does not hallucinate or promise things you cannot deliver.

Qualification is similar. Instead of rigidly asking BANT questions, a good approach is progressive qualification:

Confirm relevance (are they the right profile?)
Confirm pain or priority (is this worth a conversation now?)
Confirm ability to act (can they influence or decide?)
Confirm next step (is a meeting the right next move?)

5) Follow-ups that are unique to each person

Most follow-ups fail because they pretend the previous message never happened. A proper agent uses conversation memory:

What the prospect said
What they ignored
What question was left unanswered
What timing constraint they mentioned

A useful follow-up can be as simple as: "You mentioned X last week - curious if Y is still a priority this quarter?" That is a human move. It is also something an agent can do consistently if the system stores and retrieves the right context.

"No more spam automation tools" - what that really implies

Rémy Touzard positioned this as an alternative to spammy tools. I think that is the right framing, but it comes with a responsibility: if you automate conversations, you are scaling your reputation.

Here are practical guardrails that keep this approach ethical and effective:

Limit daily volume so personalization stays real
Use a strict "no-link" rule early in the conversation
Never fake personal experiences (no "I loved your talk" unless you actually did)
Add an easy opt-out ("If this is not relevant, tell me and I will close the loop")
Keep a human review step for edge cases (angry replies, compliance concerns, enterprise procurement)

The goal is not to sound human. The goal is to be helpful enough that a human wants to reply.

Why the meeting numbers can jump so dramatically

Rémy's headline contrast (15 hours for 1 meeting per week vs 0 hours for 2-5 meetings per day) is intentionally provocative. The reason it can happen is not magic. It is leverage.

A conversation-capable agent creates leverage in three places:

Speed to response: replying quickly increases conversion.
Consistency: every lead gets a thoughtful next step.
Coverage: you can run more parallel conversations without losing context.

The limiting factor becomes targeting and offer clarity, not time spent typing messages.

If you want to replicate this, focus on the inputs

Tools matter, but inputs matter more. If your agent is underperforming, it is usually one of these:

Your audience definition is too broad

If you target "founders" or "sales leaders" generically, no agent can save you. Narrow by industry, size, and a specific trigger.

Your offer is not concrete

"We help you grow" is not an offer. A concrete offer includes:

Who it is for
The problem
The outcome
The mechanism (what you actually do)

Your qualification logic is unclear

Decide what "qualified" means before you scale. Otherwise you will book calls that go nowhere and assume the system is broken.

You are not measuring the right metrics

Track:

Connection acceptance rate
Reply rate
Positive reply rate
Meetings booked per 100 new connections
Show rate and close rate by segment

Where CLAUDE + KAKIYO fits in

Rémy Touzard closed with an invitation: "Wanna see how CLAUDE + KAKIYO can book 2-5 meetings per day for your business?" Whether you use that exact stack or not, the blueprint is clear:

A model that can write naturally and reason about intent
A workflow layer that can fetch profile data, store memory, and route conversations
A qualification and scheduling path that ends with a booked meeting

If you are evaluating solutions, ask one question: can it handle the messy middle of the conversation (objections, timing, nuance), or is it just a better sequence tool?

The takeaway

Rémy Touzard made a point that really resonated: the future of LinkedIn outreach is not blasting more messages. It is building systems that can carry real conversations at scale.

If you can get the targeting right, keep the messaging honest, and add guardrails, "autopilot" does not have to mean "spam". It can mean consistent, helpful outreach that earns replies and turns into meetings.

This blog post expands on a viral LinkedIn post by Rémy Touzard, Let our AI Agents prospect, qualify & book meetings for you. View the original LinkedIn post →