Big Idea

The argument in one line.

A working AI sales-calling agent can start as one unstructured, rambled voice memo describing the outcome you want, with the agent platform itself turning that ramble into the technical build plan.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run a business that gets website form leads who don't book a call, and you want a bot to follow up before they go cold.
You want a concrete order of operations for wiring a phone-calling AI agent to a voice model, a phone number, and a CRM.
You're comfortable describing what you want out loud in plain language and letting an AI agent draft the technical plan from that.

SKIP IF…

You want click-by-click setup instructions. This is a live walkthrough of one founder's build, not a course.
You need guidance on call-recording consent, opt-out handling, or telemarketing compliance. The video doesn't cover any of that.

TL;DR

The full version, fast.

A founder built an AI sales agent by recording one rambling voice memo describing what he wanted, then handing it to GrokBot, which turned the ramble into a written build plan. The resulting system chains three bots: one that notices a website form was filled out but never booked, one that decides what happens next, and a voice bot, built on ElevenLabs using the founder's own cloned voice, that calls the lead within moments, asks qualifying questions, texts a booking link, and logs the transcript into the CRM. In two weeks live, it produced $16,000 in booked revenue. The lesson: describe the outcome you want in plain language and let the agent draft the technical plan, rather than trying to spec every step yourself.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:18

01 · Cold open: an AI agent calls customers

Sponsor bumper (Zapier) followed by a live phone call where a voice agent, not the founder, is on the other end.

00:18 – 03:00

02 · The live GrokBot sales call, and the $16K reveal

The demo call plays out in full: qualifying questions about visa country and investment budget, a text with a booking link, then the reveal that the bot has already booked $16,000 in revenue in two weeks live.

03:00 – 04:21

03 · How the three-bot system works

Achint explains the chain: a follow-up bot notices the form was completed without a booking, wakes a master bot, which wakes the voice bot to place the call.

04:21 – 06:00

04 · From a rambled voice note to a build plan

The origin of the system: an unstructured voice memo describing the desired outcome, which GrokBot turned into a written, ordered build plan.

06:00 – 08:33

05 · Setting up ElevenLabs, Twilio, and a cloned voice

Cloning Achint's voice in ElevenLabs, buying and registering a phone number in Twilio, and testing the live call end to end.

08:33 – 10:03

06 · How the agents hand off, and saving calls to the CRM

Clarifying which system talks to whom (GrokBot orchestrates, ElevenLabs's own LLM handles the live conversation) and how every transcript lands in Attio.

10:03 – 10:41

07 · The full tech stack, and whether it sounds robotic

A recap of every tool in the stack, and an honest answer on how natural the calls sound this early on.

10:41 – 11:38

08 · Advice, other use cases, and GrokBot vs. coding agents

Closing advice on being explicit about the desired customer journey, a real example of the bot going wrong, and a comparison to coding agents like Codex.

Atomic Insights

Lines worth screenshotting.

A form that gets filled out but never submitted for a call is still a warm lead, and calling it back within minutes converts leads a next-day follow-up would lose.
Chaining a follow-up bot, a decision-making bot, and a voice bot outperforms one bot trying to watch for triggers, decide what to do, and hold the conversation all at once.
The voice agent runs on no fixed script. It gets a set of parameters and rules, then a separate LLM handles the live conversation itself.
Cloning a founder's own voice from a single recording, instead of using a generic voice, mattered because the product's pitch depends on sounding like someone the customer can relate to.
Two weeks of live calls produced $16,000 in booked revenue from leads that had already abandoned the signup form.
The build started as an unstructured voice memo. The AI turned that ramble into a written, ordered build plan before any actual building happened.
The orchestrating agent doesn't do the talking. It wires the website form, the phone number, and the CRM together, and a separate voice model handles the conversation.
Every call gets recorded and pushed into the CRM automatically, so a human salesperson picks up the thread with full context instead of starting cold.
An outbound call that interrupts a lead who is already booking on a calendar tool creates a worse experience than not calling at all.
Being vague about what you want an agent to build works against you, because the platform will build quickly in whatever direction you point it, wrong turns included.

Takeaway

How a ramble became a sales bot

RAMBLE TO REVENUE

Describing what you want in plain, unstructured language and letting an AI agent turn it into a build plan, then a working system, beat trying to spec every technical step yourself.

02The live GrokBot sales call, and the $16K reveal

A missed or abandoned form submission is still a live lead. Calling back within minutes, while the person still remembers filling it out, converts leads a next-day email would lose.
A qualifying phone call can gather the same information a sales rep would ask for, like which country someone's passport is from and how much they plan to invest, before a human ever gets on the line.

03How the three-bot system works

Splitting a workflow across a follow-up bot, a master bot, and a voice bot keeps each piece simple: one watches for the trigger, one decides what happens next, one has the conversation.
The orchestrating layer doesn't do the talking. It wires together the website form, the phone number, and the CRM, and hands the actual conversation to a separate voice model.

04From a rambled voice note to a build plan

Explaining an idea as an unstructured, out-loud ramble is a legitimate way to brief an AI agent. Language models are trained on exactly that kind of raw, unstructured text.
Asking the agent to propose a plan before building anything gives you a chance to review the approach and the tech stack before committing to it.

05Setting up ElevenLabs, Twilio, and a cloned voice

Cloning a founder's own voice from a short recording, instead of using a generic voice, matters when the product's whole value is feeling personal to a customer with a similar background.
Testing the bot by calling your own phone number first catches obvious problems before a real lead ever hears the agent.

06How the agents hand off, and saving calls to the CRM

The voice agent has no fixed script. It gets a set of parameters and rules, then a separate LLM carries the conversation, which is why it can handle unexpected answers.
Every call gets recorded and pushed into the CRM automatically, so a human salesperson who later picks up the thread already has full context instead of starting cold.

08Advice, other use cases, and GrokBot vs. coding agents

Being vague about what you want the agent to build works against you. The platform builds quickly in whatever direction you point it, wrong turns included.
Think about the customer's journey, not just the bot's logic. A call that interrupts someone already booking on a calendar tool can cost the booking instead of helping it.

Glossary

Terms worth knowing.

E-2 visa: A U.S. visa category for citizens of treaty countries who invest a meaningful amount of capital to start or buy a U.S. business.
Orchestrator agent: An AI agent whose job is to connect other tools and agents together, such as a website form, a phone system, and a CRM, rather than handling any single conversation itself.
Voice cloning: Technology that recreates a specific person's voice from a short recording, so an AI agent can speak in that person's voice on calls.
ICP check: A short set of qualifying questions an agent asks a lead to confirm they match the kind of customer worth pursuing, short for ideal customer profile.
CRM: Customer relationship management software that stores a contact's information and history so a sales team can follow up with context.

Resources

Things they pointed at.

00:18toolGrokBot (Grok) ↗

00:27productBelong ↗

03:02toolNetlify ↗

06:21toolElevenLabs ↗

07:21toolTwilio ↗

10:04toolAttio ↗

00:00productZapier ↗

Quotables

Lines you could clip.

02:56

“This bot has been live for maybe two weeks and we've booked 16 grand in revenue because of it.”

concrete, surprising result with a number→ TikTok hook↗ Tweet quote

04:57

“One of the things LLMs are best at is processing rambles of text. It's what they're trained on.”

permission-giving insight that stands on its own→ IG reel cold open↗ Tweet quote

10:33

“Be clear about what you want... and think a little bit about the customer journey.”

closing advice with a clear, practical takeaway→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphoranalogystory

Andrew, I'm calling because you wanted to buy. This is amazing. Your agent can now call people for you and get you customers.

We're going to show you exactly how to do it step by step. It's a lot easier than you would have imagined. And you're going to see it right now.

Presented by Zapier, the AI automation company. Achit, let me see how Grokbot calls customers and gets you closed sales. Yeah, so Andrew, this is Belong, belonglaw .com.

We help founders and... technical engineers move to the United States on legal work visas. We're just going to go ahead and fill out the form.

I filled it out. We're going to hit continue. But, you know, I'm a very finicky customer.

I don't end up booking a call here. I go and close out this tab. Grogpod is going to call me and try to get me booked.

Grogpod is calling me.

Yes, I'm good to talk right now. Awesome, thanks. I'm just reaching out because we have folks with E2 visas to move to the U .S.

and I wanted to see if that's something you're looking into. What country is your passport from? I'm a Canadian.

I have a Canadian passport. Perfect. That's a great start since Canada is a treaty country.

Roughly how much capital are you thinking of investing in the U .S. business? You know, maybe $50 ,000 to $70 ,000.

Got it. That's a solid range.

holistic business or start something new? I don't know. What's better?

I'm not that sure. My friend has an Airbnb business. We might consider just partnering on that.

That partnership angle could definitely work. We can explore those options together on a call. Would you like to grab 30 minutes on my calendar so we can map out the next steps with our lawyers?

Yeah. When do you have 30 minutes available? I can send you a link to my calendar so you can pick any time that works for you.

I prefer I text that link to your phone or email it over. Could you text me, please? No problem at all.

I'll shoot that text over to you right now just to make sure it goes to the right spot. Is this the number I'm currently speaking with? Yes, it is.

Perfect. I've just sent that text to you. Let me know once you get it and if you have any trouble booking a time.

I have the text. Awesome. Glad you got it.

Just pick whatever slot works best for you on there. Anything else you're wondering about before we hop on the call? No, this is great.

Look forward to talking to the attorneys.

Sounds good. The team will be ready to dive into those details with you. Have a great rest of your day and I'll chat with you soon.

Thank you. You're very welcome. Talk soon.

Does that lead to customers where somebody will book a meeting when they wouldn't have otherwise and then actually sign up and pay? You know, this bot has been live for maybe two weeks and we've booked 16 grand in revenue because of it. Wow.

All right. Right on. I want to know the logic of it.

I want to see how you built it from scratch. All right. We saw the form.

What platform is that form on? That form is on my website on belonglaw .com. We use Netlify to service the forms.

Okay. So now when the form is filled out, show me how it goes into Grokbot. Yeah.

Before that, I'm going to show you, this is my Grokbot screen, if you can see that correctly. I can see it. I have some different bots set up.

I have a sales agent, I have a follow -up bot, etc. It all starts with the follow -up bot. So every time a quiz gets completed on the form, it comes into the follow -up bot and it's like, you know, this is a thing happening.

You should call this lead because they didn't book the form. It wakes up the voice bot and it says, voice bot woken up with the call ASAP line. That goes to the voice bot.

It gets a message from the master bot. And then it calls me, ringing this number. This is the conversation ID.

And it says pick up when you can. And then it just calls me and tries to book me for a call. When you say book, when it says pick up when you can, who is that to?

That's to you or to the customer? So in this particular instance, because it knows that this number is my number, it knows that it's calling me. So it says pick up when you can.

Otherwise, it wouldn't say that. It would just have the call, and then it would send out. It would say, we're waiting for the guest.

We're waiting for the customer to pick up. Got it. Okay, so that's what happens behind the scene.

That's how the bots do it. Let's take a look at how you created it, that initial ramble that you showed me yesterday when you showed me how you set this up. Yeah, so this is like, just look at this.

This is a voice note that I just rambled into. the speakerphone here. And I was like, look, here's what I want you to do.

I want to create a bot that learns from what our follow -up bots, et cetera, are doing. So we have other bots. They're sending emails.

We're having certain conversations. And I asked it, how do we give it a voice? I'd like to give it my voice.

Why don't you first make a plan? You know, these are some more details about what I want you to talk to them about. And let's see what happens.

So it took that, you know. The thing that stands out for me is just how much of a ramble that is. It's, I know there's something here.

I'm not exactly quite sure what it is. I want you to figure it out. You make a plan.

Let's see from there. Got it. And then you tell me what to build.

So let's see what it came back with. Yeah. One of the things LLMs are best at is processing like rambles of text.

It's what they're trained on. So I would encourage you to just talk to us as much as possible and not worry about being very good with syntax when you talk to it. Okay.

So what did it say to do? So it was like, okay, we're going to do these things. This is a tech stack.

11's lab agent, a phone number, and this chat as the trigger. How we give it your voice, you record this number of speech. This is how we give it the phone number.

And by the way, this is what we're going to learn from the follow -up bot. This is the tone. These are some hard rules.

This is the ICP check. So as you heard, it was like, where's your passport from? You know, how much money do you want to invest?

And then it gave me a build order. Okay, and then here it is. Connect to 11 Labs.

Scroll up a little bit. Let me see the order at the very top. It said here, build plan in order.

Connect to 11 Labs and Twilio if we need a number layer. Clone your voice from a short recording. Create the belong outbound agent.

Reading ICP questions, et cetera. Wire a from number and outbound. Okay, that's a lot of stuff to do.

I hate when they give me homework like this. What'd you do with that? So I first said, okay, you go ahead and add 11 Labs.

I kind of responded to it here. It opened 11 Labs on its own. It made me log into it.

I have an 11 Labs account. Then I had to go into 11 Labs. You'll see my 11 Labs screen here.

Go into voices, just to create new voice. And I just dropped in an audio recording that I made that I wanted the bot to use as the base for these conversations. Let me ask you this.

Why did you want your voice and not one of these generic American Midwestern television announcer type voices? Belong is created in my image and to solve a problem that my family faced not long ago, five years ago. They want somebody that talks like them and sounds like them when they get on the phone.

I see. Okay. And the difficulty was, how do I get into the US when it's so difficult?

Got it. Okay. So you said, I don't want that generic voice.

I want me, I want them to feel like they're talking to someone like them. Got it. All right.

So now you have it in here. The telephone number from Twilio, you took its recommendation. How hard was that to set up?

It was nothing. It was just I'm already a couple of steps in here. But if I scroll back, this is your Twilio homepage, you know, whatever your IDs and tokens.

We just go to numbers and shenders. We set up a new phone number. And you just choose whatever number you want of what's available.

Are there any delays in getting a number or being able to activate it immediately? You get it. You're good to go.

You get it. You're good to go. You register it for voice.

I mean, I'm doing this live in front of you now. And you just bought a phone number. That's it.

And now your agent would have that because of the tokens that you showed earlier that you asked me to hide. Exactly. So I did that and I gave the model to the agent and it was like, okay, cool.

By the way, I didn't know how to set up those numbers. That thing that looks very easy, Rockwalk walked me through every step of it. From there, you should see this and that.

This is what you do. I shared with it some of my information. And then it was like, great, this is awesome.

We have that phone number. We wanted to test it. So I gave it my phone number and it says, awesome.

I'm calling this phone number from belong outbound E2, our bot right now. You know, the thing that I've wondered with this is when a customer asks a question, it goes to what computer? Is it 11 labs computer or does 11 labs get it, then send it back to Grokbot, Grokbot responds and then comes back to 11 labs and 11 labs sends the voice or how do you get all of it together?

For this one, I would think of Grokpark as the orchestrator agent. So all it's doing is getting all of these other agents to work together. When you're talking, you're only talking to the 11 Labs agent.

What Grokpark enables me to do is connect it to my website, to that form, to that phone number. And by the way, as you saw, it calls very quickly after you fill out the form. That part is Grokpark.

It sees that Achimp signed up. He didn't book a call. If I had, by the way, it wouldn't have called me.

And then it goes to the 11 Labs agent and says, hey, you know, call Achimp and run the script. There is no script. There's a set of parameters, and then it can talk to you whatever you want, because 11 Labs also is using an LLM to process these conversations.

I see. So it's giving the instructions to the 11 Labs LLM, and at that point, it's not watching or controlling the conversation. It's just letting 11 Labs have the conversation with the directions.

With one correction. It's recording all those conversations and putting them into my CRM. Which, oh, I see.

Because then Grokbot gets the conversations and then Grokbot sends it to your CRM. Is that right? Exactly.

Because when the sales agent finally gets on the call with the customer, I need all that context of what was talked about, what are the questions, et cetera. So it's very seamless. It's like, I see you want to invest in this case in an Airbnb business in Hawaii.

Let's talk about that. Okay. So it's 11 Labs, Twilio.

What other software is involved? That's all. 11labs, Twilio, Brockbot.

We have Netlify for my website hosting and we have Atio for my CRM. Do people feel that it's robotic, that it feels fake, or are they accepting it as just a regular conversation? The early versions did feel robotic.

I think by now it's fine. We keep changing it. So I imagine in a week or two from now, it'll be like exactly talking to a person.

All right. Any other advice if someone's setting this up that they should know before setting it up? Be clear about what you want.

Just because RockBot is fast and can build a lot of things really quickly, you don't want it going down the wrong rabbit hole. And also think a little bit about the customer journey because having them call at wrong times or like an example of an issue we faced was a customer was trying to book a call on Calendly, but RockBot called and interrupted that flow.

The customer is annoyed. It didn't talk to the bot. It also didn't book on Calendly.

So be cognizant about that stuff. Being clear about the experience is the whole thing. Everything's so easy to build with AI.

All right. Fantastic. And if you are into Grokbot, I've got a video here with case studies that I think you're going to want to watch and see more things that you can do with Grokbot.

And frankly, don't you agree, Achint? Anything that's done with Grokbot can basically be done with other agents that people like. It can.

The thing about Grokbot is that it acts very quickly on stuff and is more iterative in testing versus a, you know, a code or a codex. It's different. It interacts with your computer to build tools.

GraphMod has its own computer. It's much easier to work with. Yeah, I've been loving it.

It's so easy, so quick. And you'll see the use cases right here. See you in that video.

The Hook

The bait, then the rug-pull.

The video opens mid-call. A voice that sounds like founder Achint Singh is on the phone with Andrew Warner, following up on a form Andrew filled out and abandoned. It isn't Achint. It's an AI agent built on GrokBot, and by the time the walkthrough ends it has already closed $16,000 in booked revenue on its own.

Frameworks

Named ideas worth stealing.

04:57list

The ramble-to-build-plan workflow

Say what you want to happen, in plain language, without worrying about syntax
Let the agent draft a build plan and a tech stack before you touch anything
Approve the plan, then execute it step by step
Test with your own phone number before going live
Assume a specific customer scenario to stress-test the qualifying questions

The approach behind the build: record a rambling explanation of the outcome you want, let the AI turn it into a structured plan, then execute the plan piece by piece.

Steal forany AI agent build where you don't already know the exact technical steps

03:30model

The three-bot handoff chain

Follow-up bot notices a form was completed but no call was booked
Master bot decides to wake the voice bot
Voice bot calls the lead within moments, in a cloned voice, and asks qualifying questions
Voice bot texts a booking link and logs the transcript to the CRM

The system splits the job across specialized bots instead of one bot doing everything, with a master bot deciding what happens next.

Steal forany lead-followup automation, not just outbound calling

CTA Breakdown