The Vibe Scaling Playbook: One Agent, Every Follow-Up, Zero Leads Left to Rot

Your pipeline isn’t dying in the sales call. It’s dying in the silence after it.

Every service business I’ve opened up — and after six years and 300+ clients I’ve opened up a lot of them — has the same corpse in the same closet: the follow-up graveyard. Leads that replied once and never got a second touch. Proposals sent into the void and never chased. “Circle back in Q3” notes that nobody circled back on. Deals that didn’t say no. Nobody was there to hear the yes.

Founders love to blame lead quality. “The leads were trash.” The leads were fine. The follow-up was trash, because the follow-up depended on a human remembering to do the most boring, repetitive, emotionally unrewarding task in the entire company — at scale, forever, while also running the business.

That’s not a discipline problem. That’s a Cave problem: a function that lives entirely in the founder’s head, executed only when the founder has slack, which is never. Bus factor of zero is not a business model, and follow-up is where it shows up first.

So here’s the teardown of how one agent in my Fleet owns this function end to end. Not a concept. The actual playbook — steal it, build it, or have someone build it for you. This is what Vibe Scaling means in practice: vibe coding built products by riffing with AI on code. Vibe Scaling builds companies by riffing with AI on operations. Same energy. Different altitude.

The shape of the thing before the steps

One agent. One function. Five stages:

Trigger → Memory → Draft → Gate → Send.

Keep that skeleton in your head, because every working agent in an AI Fleet looks like this. When a vendor shows you something that doesn’t have all five, you’re looking at a demo, not an employee. (Boring is the moat. It’s boring on purpose.)

Step 1 — Trigger: the agent wakes up on a clock, not on a feeling

Every morning at 6:00, before anyone’s awake, the follow-up agent runs. Not when someone remembers. Not when the pipeline “feels” cold. On a schedule.

Its first job is a sweep: pull every open opportunity and ask one question of each — when was the last meaningful touch, and what’s the next committed action?

The rules are dumb by design:

• Replied lead, no response from us in 24+ hours → flag.
• Proposal out, no touch in 3 days → flag.
• Discovery call done, no recap sent same day → flag.
• Any open deal untouched in 7 days → flag, escalate.
• “Follow up in [month]” note whose month has arrived → flag.

That’s it. No machine learning. No lead scoring model. Five if-statements a competent teenager could write. I’m telling you this specifically because the industry wants to sell you intelligence, and what your pipeline is starving for is attendance. The magic isn’t that the rules are smart. It’s that they run 365 mornings a year without a single sick day.

Step 2 — Memory: why this isn’t a mail merge

Here’s where it stops being a drip sequence and starts being an employee.

For every flagged lead, the agent loads context before it writes a word: the full thread history, the call notes, what they objected to, what they said mattered to them, what was promised, who else is involved, and every prior touch — so it never sends “just checking in!” to someone who asked a pricing question we never answered. (Read that sentence again. Half of you have done exactly that this month. That’s how deals die of embarrassment.)

This is the part of the Agent Stack people skip because it’s plumbing: the agent has to be wired into the CRM, the inbox, and the notes, with memory that persists across runs. Plumbing is unglamorous. Plumbing is also the difference between “AI tool” and infrastructure. Everyone claims their AI works. The wiring is what makes it true.

Step 3 — Draft: your voice, not robot voice

The agent drafts every follow-up fresh, from context, in the owner’s voice — and the voice comes from a document, not from prayer. Real emails written by the owner, a short list of phrases they’d never say, and hard rules: short paragraphs, no corporate mush, no “I hope this email finds you well,” reference the actual last conversation, one clear ask per message.

Two things I’ve learned shipping this across businesses:

1. The draft that references reality converts. “You mentioned the Denver rollout was eating your ops team — did that settle?” outperforms any template ever written, because it’s not a template. It’s memory wearing your voice.
2. The agent should sound like your Tuesday self, not your launch-day self. Overwritten follow-ups smell like marketing. The agent writes the plain, direct, mildly imperfect email you’d send if you had time. That’s the whole trick.

Step 4 — Gate: the thirty-second approval that keeps you the Pilot

Nothing sends itself. Every morning the owner gets one surface — a queue — with the day’s drafts: lead, one-line context, the draft, and two buttons. Approve. Edit.

A typical morning is eight to fifteen drafts. Reviewing them takes the length of a coffee. And this gate is not a training-wheels phase you graduate out of — it’s a permanent feature, for three reasons:

• Blast radius. Client-facing words are reputation. Reputation doesn’t get delegated to a language model, full stop. The tribe distrusts full autonomy, and the tribe is right.
• The gate trains the Fleet. Every edit is a correction the system keeps. Month one you edit a third of the drafts. Month four you edit a few a week. You’re not doing the work — you’re coaching it, and the coaching compounds.
• It changes your seat. You stop being the engine that produces follow-ups and become the Pilot who commands the function — thirty seconds of judgment instead of ninety minutes of production. That seat change, repeated across every function in the business, is the entire Cave-to-Cockpit journey. It starts here, with the least glamorous function you own.

Step 5 — Send, log, repeat

Approved drafts go out inside a sane send window, staggered like a human sent them. Every send is logged back to the CRM with a timestamp and a next-action date — which feeds tomorrow’s 6 a.m. sweep. The loop closes itself. No lead can rot silently anymore, because silence itself is now a trigger.

And once a week, the agent files a receipt: touches sent, replies generated, deals reactivated, oldest untouched lead in the system (the number that should embarrass you toward zero). Fleet receipts, not vibes — you should be able to watch the function work.

What this replaces, honestly

Where the math consistently lands when this playbook goes in:

• The founder’s follow-up hours go to near zero. Five-plus production hours a week become ~15 gate-minutes a day. [VERIFY exact figures per client before citing]
• The “no committed next action” number collapses. In most pipelines that’s a horrifying share of open deals on day one. Watching it fall is the moment owners stop calling this an experiment. [VERIFY]
• Reactivated deals appear from the graveyard — not because anything got smarter, but because someone finally showed up. Revenue that was already paid for, left on the table, picked back up. [VERIFY]

Build cost? This is one agent, five stages, standard wiring into tools you already own. Weeks, not quarters. Costs less than one month of one SDR. Runs forever.

The jump most founders never make

Vibe coding built a generation of products. Vibe Scaling builds the company around them — and follow-up is where I tell every founder to start, because it’s the highest-ROI, lowest-risk, most boring function in the building. You don’t need a platform. You don’t need a “digital workforce transformation.” You need one agent that shows up every morning to do the job nobody wanted.

Then you do it again for reporting. Then content. Then finance ops. One function at a time, each one gated, scheduled, remembered — until one day you look up and the company runs on a Fleet you command instead of a to-do list you fear.

The flashy agent booked zero meetings. The boring one is in your CRM at 6 a.m., raising the dead.

Go build it. Or call the field medic.

— Josh Collier