AI Agents Need Their Own Phone Numbers

Behind the codes.

Behind [music] the codes.

Behind [music] the codes. That's where the future grows.

That's where the future grows. Welcome to behind the code with Atomic

Welcome to behind the code with Atomic Build. My name is Emit and today I have

Build. My name is Emit and today I have with me Gal from Get Dial. I'm very

with me Gal from Get Dial. I'm very excited to have you here today, Gal.

excited to have you here today, Gal. Thanks for joining me. Um, and if you

Thanks for joining me. Um, and if you don't mind, just give a little

don't mind, just give a little introduction about yourself.

introduction about yourself. Absolutely. Thanks, Sam. Thanks for

Absolutely. Thanks, Sam. Thanks for hosting me. Um, so excited to be here.

hosting me. Um, so excited to be here. Um so to be honest um once we had that

Um so to be honest um once we had that that opportunity to jump onto on this

that opportunity to jump onto on this webinar this uh podcast sorry uh I was

webinar this uh podcast sorry uh I was very excited because I see what you're

very excited because I see what you're doing over there and the way that you're

doing over there and the way that you're breaking down companies and the way that

breaking down companies and the way that you are actually giving um the uh the

you are actually giving um the uh the ability for companies to show themsel

ability for companies to show themsel and you are asking very good questions.

and you are asking very good questions. So I'm very excited from the from that

So I'm very excited from the from that podcast and and upcoming questions. Um

podcast and and upcoming questions. Um maybe a quick intro introduction about

maybe a quick intro introduction about myself. uh I grew as um as a

myself. uh I grew as um as a professional game player. I was like a

professional game player. I was like a gamer uh on on CS 1.6

gamer uh on on CS 1.6 uh flow flew to through like the US and

uh flow flew to through like the US and the and the Europe uh to competitions

the and the Europe uh to competitions and until I I hit like 16 I realized

and until I I hit like 16 I realized that I have a lot of audience wants to

that I have a lot of audience wants to see me playing and even not only to see

see me playing and even not only to see him playing but also to basically play

him playing but also to basically play with me. Uh but I was only one and there

with me. Uh but I was only one and there was like thousands. So what I did, I

was like thousands. So what I did, I implemented a bot that um pretend myself

implemented a bot that um pretend myself to be the B the bot. So I just

to be the B the bot. So I just productized myself. Then I released it

productized myself. Then I released it as a paid plan uh like a self-s serve um

as a paid plan uh like a self-s serve um product and it reached to 50,000 like uh

product and it reached to 50,000 like uh uh daily active users. So on that time,

uh daily active users. So on that time, yeah, on that time uh my my my my mother

yeah, on that time uh my my my my mother told me that after the fourth uh air

told me that after the fourth uh air conditioner that I I I I purchased to my

conditioner that I I I I purchased to my room, she told me this and no, you you

room, she told me this and no, you you can't live at a server room, you need to

can't live at a server room, you need to sell everything. And then I realized

sell everything. And then I realized that I can't really do that. So I sold

that I can't really do that. So I sold it to uh to a gaming company and then

it to uh to a gaming company and then started to work on that the gaming

started to work on that the gaming company till I was 18. Um then I served

company till I was 18. Um then I served in the army in the IDF from the

in the army in the IDF from the intelligence scopes uh about 8 years uh

intelligence scopes uh about 8 years uh on the cyber security field mostly on

on the cyber security field mostly on the big data uh infrastructure and

the big data uh infrastructure and realtime infrastructure. After that I

realtime infrastructure. After that I realized that I want to pursue my

realized that I want to pursue my entrepreneurship career but I had no

entrepreneurship career but I had no enough experience. So I decided to join

enough experience. So I decided to join uh a company who was like a second

uh a company who was like a second timer. For the time it was a fa from

timer. For the time it was a fa from just I joined just as the uh first

just I joined just as the uh first employee first engineer and I h I hired

employee first engineer and I h I hired about like 40 engineers and basically uh

about like 40 engineers and basically uh grew the R&D department from zero to 40

grew the R&D department from zero to 40 people and the company grew in in an IP

people and the company grew in in an IP gold grow uh like from 1 to 10 to one to

gold grow uh like from 1 to 10 to one to 10 in two years. So it was an hyper

10 in two years. So it was an hyper growth and I the [clears throat]

growth and I the [clears throat] um I have the chance to be close to the

um I have the chance to be close to the farmers and see how how things are going

farmers and see how how things are going and what are the challenges of growing

and what are the challenges of growing so fast and once we hit the 10 million I

so fast and once we hit the 10 million I decided that I'm going to pursue my

decided that I'm going to pursue my career and I have enough experiences

career and I have enough experiences then we we Natan me and Omri who's a

then we we Natan me and Omri who's a child childhood friend of mine uh

child childhood friend of mine uh started the new journey which was

started the new journey which was January. Wow, that is you had quite a a

January. Wow, that is you had quite a a colored journey so far. That's amazing.

colored journey so far. That's amazing. So, so you essentially built a server

So, so you essentially built a server room in your room, your family home,

room in your room, your family home, and Wow, that's incredible. And only

and Wow, that's incredible. And only after,

after, I guess, scaling it to a certain point,

I guess, scaling it to a certain point, you got booted out like any true startup

you got booted out like any true startup founder getting booted out of their

founder getting booted out of their first workspace.

first workspace. Yeah,

Yeah, that's really that's really cool. Um so

that's really that's really cool. Um so when you were um building at Genway um

when you were um building at Genway um what was really like what did you learn

what was really like what did you learn in terms of the difference between what

in terms of the difference between what an agent can hold like a conversation

an agent can hold like a conversation and then actually operate in the real

and then actually operate in the real world?

world? So in January we optimized a lot for the

So in January we optimized a lot for the latency. It wasn't the early days on all

latency. It wasn't the early days on all of the speechtoech audio models. Back at

of the speechtoech audio models. Back at the time there was no audio to audio

the time there was no audio to audio speech to speech model at all. So it was

speech to speech model at all. So it was like combining parts like a puzzle and

like combining parts like a puzzle and try to see where you can minimize the

try to see where you can minimize the latency because we we wanted to make the

latency because we we wanted to make the conversation feel a humanike

conversation feel a humanike conversation and there was like a lot of

conversation and there was like a lot of fatigue of people that try humans try to

fatigue of people that try humans try to interact with the agents that time and

interact with the agents that time and it wasn't so natural. So one of the

it wasn't so natural. So one of the things that was an eye challenge for us

things that was an eye challenge for us it was latency but then also the quality

it was latency but then also the quality of the the conversation itself. how to

of the the conversation itself. how to make the AI agent, you know, following

make the AI agent, you know, following the guidelines. How to not uh make the

the guidelines. How to not uh make the agent say things it doesn't need to say.

agent say things it doesn't need to say. Uh how to make sure that the AI agent

Uh how to make sure that the AI agent ask the right questions, not making the

ask the right questions, not making the the human confused. So there was a lot

the human confused. So there was a lot of challenges bes around like speed uh

of challenges bes around like speed uh and quality.

and quality. And did you was it like driven just from

And did you was it like driven just from um in-house testing or a lot of customer

um in-house testing or a lot of customer feedback pointing to latency being a key

feedback pointing to latency being a key issue? So general AI uh was an AI

issue? So general AI uh was an AI moderation platform for user research

moderation platform for user research products. So um by definition the

products. So um by definition the product itself work worked with uh the

product itself work worked with uh the customer of the product was uh UXRS user

customer of the product was uh UXRS user experience researchers. So they own um

experience researchers. So they own um the onetoone uh interviews the in-depth

the onetoone uh interviews the in-depth interviews. So they knew how to judge

interviews. So they knew how to judge whether the interview is good or not. So

whether the interview is good or not. So basically it was feedback from the end

basically it was feedback from the end users but also from our customers who

users but also from our customers who are highly professionalized

are highly professionalized professionalized and out conducted and

professionalized and out conducted and effective interviews.

effective interviews. That is a tough audience.

That is a tough audience. Yeah. Yeah.

Yeah. Yeah. Yeah. Yeah. Yeah. But also like uh

Yeah. Yeah. Yeah. But also like uh really awesome for increasing product

really awesome for increasing product quality.

quality. Yeah.

Yeah. Yeah. Cool. Um so for you when did you

Yeah. Cool. Um so for you when did you realize that this was going to be a

realize that this was going to be a standalone infrastructure problem not

standalone infrastructure problem not just a feature that you're going to add

just a feature that you're going to add to an existing product? So since we

to an existing product? So since we spent a lot of time and the more we got

spent a lot of time and the more we got we spend the more we spent on the

we spend the more we spent on the understanding how the infra is being

understanding how the infra is being changed and evolves you know over time

changed and evolves you know over time it was like on point 23 when just the

it was like on point 23 when just the era started uh we realized that the

era started uh we realized that the problem is much bigger than we thought

problem is much bigger than we thought it's not only the latency and the speed

it's not only the latency and the speed but also that we implemented for one

but also that we implemented for one specific use case so for feedback but

specific use case so for feedback but then we realized that it's not only the

then we realized that it's not only the issue with um that we solved It was it

issue with um that we solved It was it wasn't only for that specific use case.

wasn't only for that specific use case. So people keep interacting with agents

So people keep interacting with agents on their phones, on their emails, on

on their phones, on their emails, on their WhatsApp, under SMS. Then we

their WhatsApp, under SMS. Then we realize that the issue which we try to

realize that the issue which we try to solve is something bigger. Then we

solve is something bigger. Then we decided that if the issue is so big, why

decided that if the issue is so big, why can't we implement it as a platform as a

can't we implement it as a platform as a standal platform? Then we thought that

standal platform? Then we thought that dial basically is the next um generation

dial basically is the next um generation product. That's why we pivoted to dial.

product. That's why we pivoted to dial. we understood that basically um the

we understood that basically um the communication between humans to agents

communication between humans to agents must be on the existing infrastructure.

must be on the existing infrastructure. So telephony uh and all of the

So telephony uh and all of the telecommunication

telecommunication uh uh protocols are so old and so you

uh uh protocols are so old and so you know behind and we thought that if we

know behind and we thought that if we can make the communication protocols

can make the communication protocols much efficient for agents at one day we

much efficient for agents at one day we would be able to even implement our own

would be able to even implement our own complication protocol and the next era

complication protocol and the next era of that is agents talking to agents

of that is agents talking to agents which we already see on dial that

which we already see on dial that basically agents are going to

basically agents are going to communicate with other agents and

communicate with other agents and right

right the telecommunication protocols are so

the telecommunication protocols are so non effective and so no so not optimized

non effective and so no so not optimized for AI agents. So for instance right now

for AI agents. So for instance right now a conversation of 10 minutes between two

a conversation of 10 minutes between two AI agents spends 20x more uh where it

AI agents spends 20x more uh where it should like sp I mean it spends 20x more

should like sp I mean it spends 20x more than it should should be spending in

than it should should be spending in terms of tokens but also in terms of

terms of tokens but also in terms of latency. So agents don't need an entire

latency. So agents don't need an entire context like people need.

context like people need. [clears throat] They don't need you to

[clears throat] They don't need you to remind what you talk like 1 minute ago

remind what you talk like 1 minute ago or like 5 minutes ago. They know how to

or like 5 minutes ago. They know how to uh they have better memory than us

uh they have better memory than us short-term memory but also long-term

short-term memory but also long-term memory. So uh the communication itself

memory. So uh the communication itself does not the protocol itself does not

does not the protocol itself does not really uh supports right now the relay

really uh supports right now the relay protocol. So this is one of the things

protocol. So this is one of the things that we found that if we are going to

that we found that if we are going to implement it's going to make that uh the

implement it's going to make that uh the communication much more effective.

communication much more effective. Right. Right. So, so I'm going to I'm

Right. Right. So, so I'm going to I'm going to ask you to step back for a

going to ask you to step back for a moment and just um because the audience

moment and just um because the audience a lot of people listening are also

a lot of people listening are also non-technical and if you're going to

non-technical and if you're going to explain dial just in in the simplest way

explain dial just in in the simplest way of what your business does. Can we start

of what your business does. Can we start there and then I have a second question

there and then I have a second question for you around the communication

for you around the communication limitations of a human-built system?

limitations of a human-built system? Yeah. Yeah. Perfect. So dial um is a

Yeah. Yeah. Perfect. So dial um is a phone carrier for AI agents basically

phone carrier for AI agents basically making AI agents accessible for all of

making AI agents accessible for all of the telecommunication protocols. So if

the telecommunication protocols. So if you say if you have a closed code or any

you say if you have a closed code or any other things like openflow nano and you

other things like openflow nano and you want it's to inter interact with the

want it's to inter interact with the real world through phone calls through

real world through phone calls through SMS through iMessage. Right now an AI

SMS through iMessage. Right now an AI agent won't have its own phone call

agent won't have its own phone call phone number sorry. What it will require

phone number sorry. What it will require you to do is to wire your personal

you to do is to wire your personal number uh into the agent. Then it will

number uh into the agent. Then it will be able to operate on your behalf. Now

be able to operate on your behalf. Now most of the times you don't want to do

most of the times you don't want to do so to do so. We want the AI agent to

so to do so. We want the AI agent to have its own phone call its own inbox.

have its own phone call its own inbox. So the only way to do to do so is go to

So the only way to do to do so is go to one of the BSPs

one of the BSPs um and start implementing and being a

um and start implementing and being a developer like weeks in order to make it

developer like weeks in order to make it happen. So what we make what we made is

happen. So what we make what we made is made it accessible but also compliant.

made it accessible but also compliant. So dial in one sentence is making the

So dial in one sentence is making the telecommunication protocols accessible

telecommunication protocols accessible and compliant for AI agents.

and compliant for AI agents. Got it. Yeah. No, that that's um I think

Got it. Yeah. No, that that's um I think the compliance part is pretty important

the compliance part is pretty important layer because uh the way communication

layer because uh the way communication moves today it's a afterthought I think

moves today it's a afterthought I think when people are building AI agents. So

when people are building AI agents. So in your um white paper you have on your

in your um white paper you have on your website uh you describe essentially like

website uh you describe essentially like a wrong species problem um and and what

a wrong species problem um and and what would you say like the infrastructure

would you say like the infrastructure today that we built for humans what are

today that we built for humans what are one of the key ways it won't work well

one of the key ways it won't work well for AI agents.

for AI agents. So the thing with compliance in like few

So the thing with compliance in like few sentences is that it's updated so

sentences is that it's updated so frequently and in order to be complied

frequently and in order to be complied with the regulations you need to uh make

with the regulations you need to uh make it your own business dayto-day. So

it your own business dayto-day. So [clears throat]

[clears throat] um say you're going to call to a

um say you're going to call to a different like country other than the

different like country other than the states. Now you need to comply not with

states. Now you need to comply not with only the destination but also with the

only the destination but also with the origin. So there's a combination. It

origin. So there's a combination. It depends from where you call to the

depends from where you call to the destination you call and see what are

destination you call and see what are the regulations over there. So for

the regulations over there. So for instance not many people know but in

instance not many people know but in order to send SMS to like a different

order to send SMS to like a different destinations there are regulations not

destinations there are regulations not only on the message itself but also on

only on the message itself but also on the hours that you can you can send SMS.

the hours that you can you can send SMS. So if you are not going to comply with

So if you are not going to comply with it, you might be in a risk that your

it, you might be in a risk that your phone number is going to be blocked and

phone number is going to be blocked and the time you realize that it's blocked,

the time you realize that it's blocked, it's too late. So you need to prevent it

it's too late. So you need to prevent it and you need to implement your own guard

and you need to implement your own guard railways. That's what we are doing

railways. That's what we are doing behind the scenes and making sure we are

behind the scenes and making sure we are compliant with everything uh and you're

compliant with everything uh and you're not hurting your um core product,

not hurting your um core product, right? So, so what is so that that seems

right? So, so what is so that that seems like a pretty key or at least one key

like a pretty key or at least one key reason to have have an agent have its

reason to have have an agent have its own phone number using its own

own phone number using its own protocols. What is another that it's a

protocols. What is another that it's a good way to have it instead of being

good way to have it instead of being inside the company's existing telecom

inside the company's existing telecom infrastructure?

infrastructure? Um,

Um, that's actually a good question. Um, so

that's actually a good question. Um, so I think one thing is to separate things.

I think one thing is to separate things. I mean to make sure that everything is

I mean to make sure that everything is in its own. But if you want to support

in its own. But if you want to support uh like use cases that are massive in

uh like use cases that are massive in terms of scale and you want to send a

terms of scale and you want to send a lot of SMSs and you want to make a lot

lot of SMSs and you want to make a lot of voice calls, so you might even need

of voice calls, so you might even need you know to maintain um a larger bucket

you know to maintain um a larger bucket of phone numbers and you you want your

of phone numbers and you you want your AI agents to control all of them on your

AI agents to control all of them on your behalf. I'm not sure if it answers.

behalf. I'm not sure if it answers. Yeah. So like essentially um with dial

Yeah. So like essentially um with dial an an AI agent can have its own phone

an an AI agent can have its own phone number to create actions in the world

number to create actions in the world and not lean into the company's existing

and not lean into the company's existing like telecom infrastructure. So one

like telecom infrastructure. So one thing that comes to my mind because uh I

thing that comes to my mind because uh I built products in the supply chain

built products in the supply chain industry um specifically in trucking and

industry um specifically in trucking and freight and there's so much

freight and there's so much communication outbound

communication outbound from like a shipper or a freight broker

from like a shipper or a freight broker going out and to your point um like most

going out and to your point um like most protocols are broken after hours.

protocols are broken after hours. there's no rules being followed etc. So

there's no rules being followed etc. So I think if you could communicate to a

I think if you could communicate to a company why it's important to have uh an

company why it's important to have uh an AI agent with its own number acting in

AI agent with its own number acting in the world um that's the core like the

the world um that's the core like the root of my question like how it would

root of my question like how it would not uh basically cross over with

not uh basically cross over with existing systems and instead be its own

existing systems and instead be its own independent actor and why that would be

independent actor and why that would be impactful.

impactful. Yeah. So the thing with AI agents that I

Yeah. So the thing with AI agents that I believe that today um every company

believe that today um every company implement its own dedicated solutions

implement its own dedicated solutions that are not part of the core product

that are not part of the core product itself. Now it's good for separation of

itself. Now it's good for separation of data and also a separation of concerns.

data and also a separation of concerns. So once you plug in your knowledge base

So once you plug in your knowledge base your entire company knowledge base into

your entire company knowledge base into the AI agents it get messed and you want

the AI agents it get messed and you want to control the way that the agent is

to control the way that the agent is going to operate. So the same thing

going to operate. So the same thing applies with phone numbers. uh you're

applies with phone numbers. uh you're going to separate virtually separate the

going to separate virtually separate the data that is being accessible for your

data that is being accessible for your AI agent does you will be able to

AI agent does you will be able to control what are the things that the

control what are the things that the agent could potentially risk and the

agent could potentially risk and the thing with the phone number for the

thing with the phone number for the agents once once it becomes so simple so

agents once once it becomes so simple so companies can you know start

companies can you know start implementing and and PC things without

implementing and and PC things without making their entire core product going

making their entire core product going wrong and we do believe that if you have

wrong and we do believe that if you have a company if you have like a team a

a company if you have like a team a dedicated team that is the owner uh uh

dedicated team that is the owner uh uh like owning the telecommunication

like owning the telecommunication protocol inside the company. So we might

protocol inside the company. So we might not need dial. We might need to go to

not need dial. We might need to go to the other BSPs, but you need to make

the other BSPs, but you need to make sure that that that team this is the the

sure that that that team this is the the bread and the butter, right? You need to

bread and the butter, right? You need to make sure that that that team is the

make sure that that that team is the owners are the owners, sorry, of the um

owners are the owners, sorry, of the um of the entire like um entire info,

of the entire like um entire info, right? Got it. Yeah. I think it's um I

right? Got it. Yeah. I think it's um I think because you're in like a category

think because you're in like a category uh like you're creating a category and

uh like you're creating a category and defining a new way people kind of

defining a new way people kind of approach this. Um, one thing that I I

approach this. Um, one thing that I I came to before I met you and just when I

came to before I met you and just when I got introduced to your company as a

got introduced to your company as a whole is that and including my our

whole is that and including my our engineering team, they came to this

engineering team, they came to this conclusion like is this a voice agent?

conclusion like is this a voice agent? This is a telecom API company like what

This is a telecom API company like what is it? When people kind of approach and

is it? When people kind of approach and define your company before actually

define your company before actually getting in and reading through what you

getting in and reading through what you do, what do you think defining it as a

do, what do you think defining it as a voice agent or a telecom API company?

voice agent or a telecom API company? What does that fundamentally mis like

What does that fundamentally mis like misunderstand about what dial does? So

misunderstand about what dial does? So it's actually [clears throat] a good

it's actually [clears throat] a good question. Um so the way we see dial is

question. Um so the way we see dial is that you don't have to handle um not the

that you don't have to handle um not the the the the bones itself and not also

the the the bones itself and not also the metal itself. So you just want to

the metal itself. So you just want to give us the mission and you want us to

give us the mission and you want us to accomplish that. Now yes there are teams

accomplish that. Now yes there are teams that are working on a different mode

that are working on a different mode what we call the self-hosted mode.

what we call the self-hosted mode. Basically we are just giving the infra

Basically we are just giving the infra and the compliance and they owning the

and the compliance and they owning the um the actual conversation and what are

um the actual conversation and what are the logic. Now the way we think about it

the logic. Now the way we think about it and we are going to talk a lot about it

and we are going to talk a lot about it uh in the the next talks the next

uh in the the next talks the next questions it should be battery included.

questions it should be battery included. So you just need to basically provide um

So you just need to basically provide um the mission itself and um like a budget

the mission itself and um like a budget then we are going to do that in your on

then we are going to do that in your on your behalf. Telecommunication APIs are

your behalf. Telecommunication APIs are accept accessible for agents for a long

accept accessible for agents for a long time. That's not that's not the way that

time. That's not that's not the way that we uh like um um make make a huge um

we uh like um um make make a huge um like make make a huge success but making

like make make a huge success but making AI agents works natively with

AI agents works natively with telecommunication by providing them the

telecommunication by providing them the error codes that are accessible for a

error codes that are accessible for a agents to know how to solve that issue

agents to know how to solve that issue by itself but also to make sure that the

by itself but also to make sure that the agents uh like building the honest of

agents uh like building the honest of the agents of when things get wrong uh

the agents of when things get wrong uh prevent the action. itself and not

prevent the action. itself and not making you know the the the ice ball

making you know the the the ice ball keep rolling.

keep rolling. Got it. Okay. So, so can in in like your

Got it. Okay. So, so can in in like your experience with deployments can can um

experience with deployments can can um an agent perform really well in the

an agent perform really well in the application layer and then start to fail

application layer and then start to fail because carriers, reputation systems or

because carriers, reputation systems or users start flagging it as as spam.

users start flagging it as as spam. Yeah. So that's also like we have models

Yeah. So that's also like we have models in in in set in our product that we know

in in in set in our product that we know uh based on your initial use to predict

uh based on your initial use to predict whether your use case is going to be a

whether your use case is going to be a huge failure or not. So we give you a

huge failure or not. So we give you a lot of notifications not you but your

lot of notifications not you but your agents that's why you don't need to be

agents that's why you don't need to be in the loop but we give a lot ofations

in the loop but we give a lot ofations and true web books of what they need to

and true web books of what they need to change in line to be uh in line with the

change in line to be uh in line with the regulations. So if the the the call

regulations. So if the the the call itself has a lower like um for instance

itself has a lower like um for instance response rate or the agent talks much

response rate or the agent talks much more than the the other side. Those are

more than the the other side. Those are flags that we find makes the phone

flags that we find makes the phone carriers

carriers um flagging uh that that that specific

um flagging uh that that that specific phone number is is uh going to be like a

phone number is is uh going to be like a spam or something like that. So it's

spam or something like that. So it's much more complicated to that.

much more complicated to that. the [clears throat] response rate and

the [clears throat] response rate and the Russia between the token and talker.

the Russia between the token and talker. This is something that we find um makes

This is something that we find um makes the agents in a loop that it can um can

the agents in a loop that it can um can ill itself in the way that is acting on.

ill itself in the way that is acting on. Okay. Right. So, so basically like uh

Okay. Right. So, so basically like uh creating flags that uh show conversation

creating flags that uh show conversation patterns that are not uh well it's

patterns that are not uh well it's interesting that you say that because uh

interesting that you say that because uh conversations with humans are often like

conversations with humans are often like human to human. There are so many

human to human. There are so many interruptions and some people talk over

interruptions and some people talk over others and there's no flags, right? Like

others and there's no flags, right? Like it's just that particular human operator

it's just that particular human operator continues in that way perhaps

continues in that way perhaps indefinitely until they get coaching.

indefinitely until they get coaching. But it sounds like what you've created

But it sounds like what you've created essentially is a coaching layer inside

essentially is a coaching layer inside to have agents learn from what is good

to have agents learn from what is good and what is bad to keep moving forward.

and what is bad to keep moving forward. Is that is that how also how your

Is that is that how also how your customers interpret it? Like they can

customers interpret it? Like they can actually have these agents being self-

actually have these agents being self- coach, self-healing and moving forward.

coach, self-healing and moving forward. Correct. Correct. However, we do have as

Correct. Correct. However, we do have as a human to human, you know, um flags.

a human to human, you know, um flags. It's most on the nuance of you know the

It's most on the nuance of you know the the non non-talked um like um um where

the non non-talked um like um um where we desri describe things. So it can be

we desri describe things. So it can be like a nonverbal cues that we can

like a nonverbal cues that we can basically infer from the conversation.

basically infer from the conversation. So if someone say like again as as I

So if someone say like again as as I said before you realize that you

said before you realize that you basically did not follow the context the

basically did not follow the context the previous context that you shared with

previous context that you shared with you and there are clues that you can

you and there are clues that you can take and take the advantage of and make

take and take the advantage of and make sure that next time your prompt is going

sure that next time your prompt is going to be better you are missing some MCPS

to be better you are missing some MCPS and if you start asking questions so we

and if you start asking questions so we had like one use case um that the

had like one use case um that the customer um called to his customers and

customer um called to his customers and asked questions the there there was so

asked questions the there there was so obvious like when was the last time that

obvious like when was the last time that you used my product the agent asked

you used my product the agent asked that. So um the company itself has this

that. So um the company itself has this data and when you're asking that

data and when you're asking that question to your customer so basically

question to your customer so basically if it feels not natural and feels not

if it feels not natural and feels not not like personal. So what he had to do

not like personal. So what he had to do is to implement its own uh MCP and

is to implement its own uh MCP and connect it to dial then basically it

connect it to dial then basically it made the conversation much more

made the conversation much more healthier and filthy in terms of you

healthier and filthy in terms of you know making the conversation tailor made

know making the conversation tailor made to that specific customer. And to to

to that specific customer. And to to your point earlier, because agents have

your point earlier, because agents have uh very sharp short-term, long-term and

uh very sharp short-term, long-term and context memory, that must have been a a

context memory, that must have been a a big change probably immediately, I

big change probably immediately, I imagine, for customer interactions in

imagine, for customer interactions in the company itself.

the company itself. Correct.

Correct. Um so humans, you know, can connect up

Um so humans, you know, can connect up to 10 dots. Uh like the complexity is up

to 10 dots. Uh like the complexity is up to 10 dots. Agent can connect up to

to 10 dots. Agent can connect up to 1,000 dots. And sometimes, you know, I'm

1,000 dots. And sometimes, you know, I'm asking my GL code questions and he

asking my GL code questions and he immediately knows what I'm talking about

immediately knows what I'm talking about and connects the dots that I was I I'm

and connects the dots that I was I I'm so like always so surprised of how

so like always so surprised of how intelligent it is it is and we al we

intelligent it is it is and we al we always like underestimates AI agents

always like underestimates AI agents because we don't feed them with the with

because we don't feed them with the with the right data. So the way that dial is

the right data. So the way that dial is working is by giving um seamlessly

working is by giving um seamlessly integration to your product uh with a

integration to your product uh with a really simple like MCP that you can

really simple like MCP that you can implement for dial than the conversation

implement for dial than the conversation and we see it a lot of on the cold

and we see it a lot of on the cold outreach. We see in the cold outreach

outreach. We see in the cold outreach use case on dial about like 20 to 30

use case on dial about like 20 to 30 higher percentage of conversions. That's

higher percentage of conversions. That's only because it's so obvious to connect

only because it's so obvious to connect your Apollo or your MCP or your actual

your Apollo or your MCP or your actual product or even play to dial. It's like

product or even play to dial. It's like 5 minutes from the time that you want to

5 minutes from the time that you want to do that until it's in production and

do that until it's in production and then the conversations is being better

then the conversations is being better in terms of the feedback loop because it

in terms of the feedback loop because it knows from conversation to conversation

knows from conversation to conversation how to implement itself with the flags

how to implement itself with the flags that we are giving them on real time.

that we are giving them on real time. So, so what what do you think stop? So,

So, so what what do you think stop? So, so I've noticed of course being the

so I've noticed of course being the recipient of uh everyone is the

recipient of uh everyone is the recipient of inbound calls, cold calls,

recipient of inbound calls, cold calls, right now some companies elect to not

right now some companies elect to not give agent systems contacts. They're

give agent systems contacts. They're just they're just literally cold calling

just they're just literally cold calling with no context in a general format.

with no context in a general format. What do you think prevents a company

What do you think prevents a company from connecting their Apollo or Clay to

from connecting their Apollo or Clay to enrich um the agent performance and a

enrich um the agent performance and a company that does it naturally? So it's

company that does it naturally? So it's about complexity

about complexity [clears throat and cough]

[clears throat and cough] some most of the time when you want to

some most of the time when you want to connect dots you need um a tech guy you

connect dots you need um a tech guy you need like an engineer without engineer

need like an engineer without engineer you just need a plot code subscription

you just need a plot code subscription and you can connect our NCP our

and you can connect our NCP our documentation NCP which is doc first NCP

documentation NCP which is doc first NCP we call it like that

we call it like that um to your CL code and just describe

um to your CL code and just describe what you want to do and your cloud code

what you want to do and your cloud code is going to figure it out what it needs

is going to figure it out what it needs to do in your in Dial in your dials

to do in your in Dial in your dials environment. I think that is a very

environment. I think that is a very powerful um setup note that that's what

powerful um setup note that that's what people yeah people because like uh

people yeah people because like uh anyone who's building in the space that

anyone who's building in the space that is uh not a software engineer you can do

is uh not a software engineer you can do this now on your own as a strong ICP

this now on your own as a strong ICP inside a company that's been tasked with

inside a company that's been tasked with an outcome like this and um speaking of

an outcome like this and um speaking of of that [clears throat]

of that [clears throat] you've discussed like how using AI

you've discussed like how using AI agents across product engineering design

agents across product engineering design QA is is pretty important um I think

QA is is pretty important um I think many engineering leaders like yourself

many engineering leaders like yourself has have been saying this now for the

has have been saying this now for the past year or so. Um but inside dial what

past year or so. Um but inside dial what is being built by agents if anything at

is being built by agents if anything at all?

all? So it's more than if anything at all

So it's more than if anything at all it's 99% of the code built by agents.

it's 99% of the code built by agents. Okay. So the way that we are

Okay. So the way that we are implementing it's not the old you know

implementing it's not the old you know um you know the old path to implement

um you know the old path to implement products that when you describe um the

products that when you describe um the documentation but what we are doing uh

documentation but what we are doing uh we are spending a lot of time on the

we are spending a lot of time on the description of what we what we want the

description of what we what we want the the agent to achieve not the user it's

the agent to achieve not the user it's not even anymore user story it's agent

not even anymore user story it's agent story that's the way we call it so we

story that's the way we call it so we describe what is the agent what is its

describe what is the agent what is its identity um describe it I mean There are

identity um describe it I mean There are different identity for the agent of

different identity for the agent of agents. There are codex and plot code.

agents. There are codex and plot code. We are optimizing for both of them. Um

We are optimizing for both of them. Um now the way we describe it is the way

now the way we describe it is the way the agent wants to um to act. Meaning

the agent wants to um to act. Meaning that it's not only what is the um the

that it's not only what is the um the what what is the definition of done but

what what is the definition of done but also how much tokens how many tokens we

also how much tokens how many tokens we want the agent to spend on that specific

want the agent to spend on that specific action. Then that the that requirement

action. Then that the that requirement basically feeds into our engineer agents

basically feeds into our engineer agents that basically engineer agents then in

that basically engineer agents then in parallel to parallel to their QA agent

parallel to parallel to their QA agent and at the end there is the requirement

and at the end there is the requirement that is built and optimized specifically

that is built and optimized specifically for that AI agent and then is being

for that AI agent and then is being tested by by our clo code and also codex

tested by by our clo code and also codex agents to see whether they ac accomplish

agents to see whether they ac accomplish the task or not and if not It's become

the task or not and if not It's become it revert back to the engineer the

it revert back to the engineer the architect and come back until the codex

architect and come back until the codex agent our codec agents and and cloud

agent our codec agents and and cloud code agents approve that it works for

code agents approve that it works for them in the way that the the human

them in the way that the the human meaning our team described.

meaning our team described. I think you just gave um every product

I think you just gave um every product manager on earth a very hot tip to start

manager on earth a very hot tip to start writing agent stories instead of user

writing agent stories instead of user stories as well as the breakdown of what

stories as well as the breakdown of what that should look like. Um it's

that should look like. Um it's interesting because like uh naturally I

interesting because like uh naturally I think teams have conformed to trying to

think teams have conformed to trying to write uh for agents when they're heavily

write uh for agents when they're heavily used inside a team but I think for other

used inside a team but I think for other teams this is still very foreign. It's

teams this is still very foreign. It's just like a hack together a hybrid of

just like a hack together a hybrid of like some teams are using it some teams

like some teams are using it some teams are not. What do you think like a a team

are not. What do you think like a a team an engineering team that is just kind of

an engineering team that is just kind of introducing the idea of having coding

introducing the idea of having coding agents involved in their uh environment

agents involved in their uh environment versus a team that is um very strong and

versus a team that is um very strong and using it? What would the strong team

using it? What would the strong team refuse to do that the junior team is

refuse to do that the junior team is probably doing often now? So I believe

probably doing often now? So I believe that the Gen Z has no barriers and I

that the Gen Z has no barriers and I would say that jourers right now has a

would say that jourers right now has a huge advantage

huge advantage um to be on the market and you know to

um to be on the market and you know to go viral with products with PFC's and

go viral with products with PFC's and and you know go beyond

the teams that I see right now are

thinking about the one percentage that

thinking about the one percentage that it does not work and you know can can

it does not work and you know can can fail and are optimizing for that thing.

fail and are optimizing for that thing. Now it's a thing between velocity and

Now it's a thing between velocity and quality. I believe that there is a

quality. I believe that there is a balance between them. So we also had

balance between them. So we also had that conversation in our team when we

that conversation in our team when we have one senior engineer who is like 15

have one senior engineer who is like 15 years old, 15 years uh experience as

years old, 15 years uh experience as experience of 15 years and the other one

experience of 15 years and the other one is three years experienced and this um

is three years experienced and this um uh friction between them create a lot of

uh friction between them create a lot of like knowledge of how one was one is

like knowledge of how one was one is optimizing for velocity the other one is

optimizing for velocity the other one is opting for quality and they found a

opting for quality and they found a balance. one one specific uh example we

balance. one one specific uh example we have um we have the community that we

have um we have the community that we are maintaining of Dial builders that

are maintaining of Dial builders that want to build on Dial and everything

want to build on Dial and everything over there is is like automated. We have

over there is is like automated. We have no community manager and we have about

no community manager and we have about like hundreds of of of community members

like hundreds of of of community members and a lot of feature requests a lot of

and a lot of feature requests a lot of bugs that is being applied. Hopefully it

bugs that is being applied. Hopefully it will reduce the the amount of bugs but

will reduce the the amount of bugs but um we man we we found ourselves that we

um we man we we found ourselves that we can basically we can't manage the

can basically we can't manage the community efficiently. So we implemented

community efficiently. So we implemented a community manager agent that basically

a community manager agent that basically goes over the request, connect the dots,

goes over the request, connect the dots, see whether it's a bug or not. It's

see whether it's a bug or not. It's going to um open a ticket on notion and

going to um open a ticket on notion and then the engineers wanted to have more

then the engineers wanted to have more control on things that are like on our

control on things that are like on our core product or things that are are for

core product or things that are are for instance on the CLI level. So we managed

instance on the CLI level. So we managed to uh basically productize two different

to uh basically productize two different agents, two different um tasks

agents, two different um tasks um tasks definition. One we call auto

um tasks definition. One we call auto ship, meaning that the AI agent is going

ship, meaning that the AI agent is going to close the loop itself and merge into

to close the loop itself and merge into production and the other one is manual

production and the other one is manual ship where the AI is going to implement

ship where the AI is going to implement everything um launch a feature branch

everything um launch a feature branch send to the developers for their check

send to the developers for their check with all of the the automated checks

with all of the the automated checks that we ran before

that we ran before and only the engineer is going to be the

and only the engineer is going to be the one who's going to push the production.

one who's going to push the production. So that gave a lot of control

So that gave a lot of control once uh but but also maintaining um the

once uh but but also maintaining um the decision whether it's a specific whether

decision whether it's a specific whether it's like auto ship or manual ship with

it's like auto ship or manual ship with the engineering itself.

the engineering itself. And do you are you are you tracking uh

And do you are you are you tracking uh which group naturally gravitates to

which group naturally gravitates to manual versus auto ship because you

manual versus auto ship because you mentioned like a junior is all about

mentioned like a junior is all about velocity someone more experiences all

velocity someone more experiences all about like like is there a natural

about like like is there a natural gravitation like I'm going to go into

gravitation like I'm going to go into auto versus manual mode.

auto versus manual mode. [laughter]

[laughter] So frankly uh our senior engineers are

So frankly uh our senior engineers are much are using much more uh like

much are using much more uh like autoshipip versus the gener who are

autoshipip versus the gener who are using the manual chip. We thought

using the manual chip. We thought Oh nice nice

Oh nice nice we thought it did the the opposite but

we thought it did the the opposite but that the senior engineers because they

that the senior engineers because they built the auto ship and the manual ship

built the auto ship and the manual ship infrastructure they feel much more

infrastructure they feel much more confident with auto ship things.

confident with auto ship things. Okay got it because they built the

Okay got it because they built the system the confidence grow. Yeah that

system the confidence grow. Yeah that makes a lot of sense. Um so for you like

makes a lot of sense. Um so for you like now that agents can produce code, test

now that agents can produce code, test systems, work continuously, how does

systems, work continuously, how does your role change as like the the leader

your role change as like the the leader of the company um on the technical and

of the company um on the technical and business side, how does your role

business side, how does your role change?

change? So I think it's not even my role. I

So I think it's not even my role. I think it's it's everyone's roles. Uh we

think it's it's everyone's roles. Uh we find our engineers that they are less

find our engineers that they are less focused on building specific requests

focused on building specific requests rather than building infrastructure and

rather than building infrastructure and processes, automated processes for our

processes, automated processes for our product. So instead of spending two days

product. So instead of spending two days on a specific feature, we are thinking

on a specific feature, we are thinking ourselves how can we implement how we

ourselves how can we implement how we can spend like 3 days on building a

can spend like 3 days on building a process that next time that we are going

process that next time that we are going to have the same feature on the same

to have the same feature on the same like you know area uh and spend like one

like you know area uh and spend like one hour spending on the specific feature

hour spending on the specific feature but more than that I feel that each one

but more than that I feel that each one of the employees right now are focused

of the employees right now are focused more than 50%age on the go to market

more than 50%age on the go to market because we we sell to builders um We

because we we sell to builders um We highly known and highly interaction. We

highly known and highly interaction. We have high interaction with our

have high interaction with our customers. We know the we know the

customers. We know the we know the customers because we are the customers

customers because we are the customers of of ourselves. We are using dial to

of of ourselves. We are using dial to call outreach uh to potential prospects.

call outreach uh to potential prospects. So the employees is also a go to market

So the employees is also a go to market engineer and thinking about how can we

engineer and thinking about how can we implement a better approach. Uh so one

implement a better approach. Uh so one of the things that we implemented we are

of the things that we implemented we are not using anymore clay lamb list and all

not using anymore clay lamb list and all of the instantly products because those

of the instantly products because those called outreach um as a so low you know

called outreach um as a so low you know reply rate between one to three

reply rate between one to three percentage and we built our own that

percentage and we built our own that basically goes to the um uh all of the

basically goes to the um uh all of the products and the vendors that we know

products and the vendors that we know where our prospect uh would be for

where our prospect uh would be for instance Gita Bing face discord and we

instance Gita Bing face discord and we basically knows um I We take that

basically knows um I We take that context and we feed it into our go to

context and we feed it into our go to market product that we built and we know

market product that we built and we know how to uh surface the opportunities to

how to uh surface the opportunities to the specific pro prospect not to make

the specific pro prospect not to make sure that he's not thinking of how can I

sure that he's not thinking of how can I integrate dial into my stack but we are

integrate dial into my stack but we are giving him exactly how we think he

giving him exactly how we think he should integrate dial into the stack and

should integrate dial into the stack and also the instruction itself that takes

also the instruction itself that takes like 30 seconds to test.

like 30 seconds to test. Wow. So, so like this this basically

Wow. So, so like this this basically like this is one of the I think themes

like this is one of the I think themes that a lot of businesses talk about is

that a lot of businesses talk about is that how would you unlock teams uh from

that how would you unlock teams uh from task like where would their time

task like where would their time otherwise go. So in this case um every

otherwise go. So in this case um every staff member essentially becomes a go to

staff member essentially becomes a go to market engineer and taking that you've

market engineer and taking that you've built your own system to target your

built your own system to target your prospects and increase conversion rate

prospects and increase conversion rate which is really uh phenomenal and was

which is really uh phenomenal and was that was that one of your team members

that was that one of your team members that came to that conclusion that they

that came to that conclusion that they should pursue this or was that you? So I

should pursue this or was that you? So I think once you know you release time for

think once you know you release time for your team when you have like um when you

your team when you have like um when you have like um brilliant team and you your

have like um brilliant team and you your higher talent team uh they know what to

higher talent team uh they know what to do with the next time. So I think

do with the next time. So I think our engineers and the the staff members

our engineers and the the staff members always want to be connected and much

always want to be connected and much more close to the to the customers. So

more close to the to the customers. So the natural thing was start sending

the natural thing was start sending them, start hunting them, start know

them, start hunting them, start know where they are, start being a part of

where they are, start being a part of more communities and see the feedback

more communities and see the feedback yourself. You don't need me to, you

yourself. You don't need me to, you know, uh, forward the feedback. You can

know, uh, forward the feedback. You can just talk,

just talk, right? Yeah. No, that's great. Um, so

right? Yeah. No, that's great. Um, so one thing in I saw in the white paper

one thing in I saw in the white paper again is this the ratio like 100 agents

again is this the ratio like 100 agents to one human which is just uh you hear

to one human which is just uh you hear about it like often but it's sometimes

about it like often but it's sometimes it's hard to hold that concept in your

it's hard to hold that concept in your mind. Um, a question I had for you is

mind. Um, a question I had for you is that are we do you think we're building

that are we do you think we're building a world where uh agents will behave like

a world where uh agents will behave like humans or will are we building a world

humans or will are we building a world where networks will eventually recognize

where networks will eventually recognize agents are fundamentally different

agents are fundamentally different participant um and and how do you

participant um and and how do you approach that as you're building out a

approach that as you're building out a dial?

dial? So I believe that the new world is going

So I believe that the new world is going to be operators and workers. The humans

to be operators and workers. The humans are going to be the operators and the

are going to be the operators and the workers are going to be the humans. Now

workers are going to be the humans. Now today the operator and the worker is the

today the operator and the worker is the same one is the human

same one is the human right

right believe that the differentiation is

believe that the differentiation is going to be clear it's not going to it's

going to be clear it's not going to it's going to be so clear but also I believe

going to be so clear but also I believe that um humans are going to do to know

that um humans are going to do to know uh I mean managing a lot of people this

uh I mean managing a lot of people this is like um a muscle that not a lot of

is like um a muscle that not a lot of people know how to do and managing

people know how to do and managing agents who are so brutal in terms of you

agents who are so brutal in terms of you know going and always positive to do So

know going and always positive to do So we're not challenging you about things

we're not challenging you about things that's create a lot of of challenge. Um

that's create a lot of of challenge. Um so I think people are going to be much

so I think people are going to be much more educated on how to manage people

more educated on how to manage people but also then need to shift on how to

but also then need to shift on how to manage agents

manage agents and um in terms of how we approaching

and um in terms of how we approaching that on building dial we starting with

that on building dial we starting with the um with with the trust layer to give

the um with with the trust layer to give the humans the control of what they want

the humans the control of what they want to do without thinking on the complexity

to do without thinking on the complexity the techity of what happened behind the

the techity of what happened behind the scenes while not um while showing them

scenes while not um while showing them and being transparent with all of the

and being transparent with all of the what happened behind the scenes. So in

what happened behind the scenes. So in terms of the voice call, we are storing

terms of the voice call, we are storing and recording all of the voice calls to

and recording all of the voice calls to make them able to hear and see what

make them able to hear and see what happened over there. In the SMS, if

happened over there. In the SMS, if there was any replay SMS, meaning that

there was any replay SMS, meaning that it was declined and then it recent

it was declined and then it recent automatically, we're going to show them

automatically, we're going to show them that we're giving them all of the tools

that we're giving them all of the tools to feel confident, but also not to

to feel confident, but also not to thinking on how things should be

thinking on how things should be implemented uh in place.

implemented uh in place. Got it. So you've touched on like the

Got it. So you've touched on like the idea in today or in the very near future

idea in today or in the very near future it'll be like a operator worker uh like

it'll be like a operator worker uh like the economy will be ruled by this this

the economy will be ruled by this this kind of system um and today that is a

kind of system um and today that is a new skill people are picking up right

new skill people are picking up right like how do you actually deploy the

like how do you actually deploy the these workers in large numbers and

these workers in large numbers and orchestrate and manage them what do you

orchestrate and manage them what do you think in I don't know why we use 5 years

think in I don't know why we use 5 years as the deacto amount project in the

as the deacto amount project in the future but let's just use it uh 5 years

future but let's just use it uh 5 years from now what will feel less strange

from now what will feel less strange that feels rather uncomfortable today

that feels rather uncomfortable today talking to to an AI agent right now it

talking to to an AI agent right now it still feels not natural it still feels

still feels not natural it still feels that you're talking to a robot

that you're talking to a robot I believe that in [clears throat] five

I believe that in [clears throat] five years from now people will want to speak

years from now people will want to speak to to AI agents not to humans in terms

to to AI agents not to humans in terms of work related tasks while maintaining

of work related tasks while maintaining the friendliness I mean I I believe that

the friendliness I mean I I believe that right now even right now IRL's invite in

right now even right now IRL's invite in in real life events uh getting a lot of

in real life events uh getting a lot of traction and people wants the connection

traction and people wants the connection between people with other people to be

between people with other people to be um to be perceived. So I believe that

um to be perceived. So I believe that all of the tech so the the the work task

all of the tech so the the the work task related um will be much more natural and

related um will be much more natural and it's going to be uh more more and more

it's going to be uh more more and more on voice rather than on communications

on voice rather than on communications channels like Slack and WhatsApp and

channels like Slack and WhatsApp and things like that. So you just mean like

things like that. So you just mean like uh just speaking naturally into like a

uh just speaking naturally into like a whatever recipient, phone, computer,

whatever recipient, phone, computer, microphone. Okay.

microphone. Okay. I mean even more than that, think about

I mean even more than that, think about it like orchestrating a fleet of agents

it like orchestrating a fleet of agents right now requires you to be the

right now requires you to be the orchestrator. Meaning that one agent

orchestrator. Meaning that one agent going with going to you asking a

going with going to you asking a question, you're going to ask your other

question, you're going to ask your other other agent. Think about it like a voice

other agent. Think about it like a voice call with 15 agents that knows how to

call with 15 agents that knows how to communicate together where you're on the

communicate together where you're on the line. just making sure that if someone

line. just making sure that if someone askked you something, you have the right

askked you something, you have the right answer to get. But then we believe that

answer to get. But then we believe that most of the communication the the um the

most of the communication the the um the tactic communication is going to be

tactic communication is going to be agent to agent and you run only on the

agent to agent and you run only on the strategic questions that has no enough

strategic questions that has no enough context. You would just need to you

context. You would just need to you would just need to uh add integrations

would just need to uh add integrations and add more relevant context to make

and add more relevant context to make the the workers work efficiently.

the the workers work efficiently. And and much like um in a fully human a

And and much like um in a fully human a human organization, the workers collect

human organization, the workers collect context through operations that a

context through operations that a strategy leader would sometimes miss in

strategy leader would sometimes miss in that like future you're describing. I I

that like future you're describing. I I envision a world where the agent workers

envision a world where the agent workers will collect context that I don't have.

will collect context that I don't have. So I think that'll be a very interesting

So I think that'll be a very interesting exchange that will will start to occur.

exchange that will will start to occur. Yeah. Yeah. I would say that there

Yeah. Yeah. I would say that there [clears throat] is a risk you know

[clears throat] is a risk you know because um it can be like a rolling

because um it can be like a rolling mistake that if the agent got the wrong

mistake that if the agent got the wrong context and is going to act on that and

context and is going to act on that and you don't you are not even aware what

you don't you are not even aware what was the context that he received during

was the context that he received during the steps.

the steps. Mhm.

Mhm. Won't be able to judge whether it's good

Won't be able to judge whether it's good or not. So I think context this is I

or not. So I think context this is I mean harnessing your AI agent to work on

mean harnessing your AI agent to work on the real data versus like um an

the real data versus like um an imaginate data. This is something that

imaginate data. This is something that would need uh I attention from your

would need uh I attention from your side. This is what why I think the

side. This is what why I think the operator should be the only one who's

operator should be the only one who's going to add additional um knowledge

going to add additional um knowledge bases into the AI agents.

bases into the AI agents. Right. Yeah. Yeah. I I think um it's

Right. Yeah. Yeah. I I think um it's just it's just very it's for me it's

just it's just very it's for me it's fascinating and to to see a a point in

fascinating and to to see a a point in the world will come where the operators

the world will come where the operators context will be accelerated by the

context will be accelerated by the workers in this like strange rapid

workers in this like strange rapid symbiotic relationship that we sort of

symbiotic relationship that we sort of see today but we don't fully see it

see today but we don't fully see it because like you said memory like human

because like you said memory like human beings memory is it's often changes as

beings memory is it's often changes as time from the event goes forward right

time from the event goes forward right like if you're on a we're on this call

like if you're on a we're on this call today tomorrow my recall will be reduced

today tomorrow my recall will be reduced from like right after we hang up the

from like right after we hang up the call and it'll just continue to reduce

call and it'll just continue to reduce as we go forward. So I I'm excited for

as we go forward. So I I'm excited for that world and like you like having

that world and like you like having seeing Claude and um Chat GPT's memory

seeing Claude and um Chat GPT's memory like it persists in such a more elegant

like it persists in such a more elegant way than it did last year and it's made

way than it did last year and it's made my my work much better. Um I think your

my my work much better. Um I think your your lens and your vision for the future

your lens and your vision for the future from a communication standpoint is is

from a communication standpoint is is fantastic. So, um, you paid me a

fantastic. So, um, you paid me a compliment earlier in the conversation.

compliment earlier in the conversation. That's my compliment, uh, to you. And I

That's my compliment, uh, to you. And I think what you're building is really

think what you're building is really cool. And just shifting as we close out

cool. And just shifting as we close out our conversation. Um, which business

our conversation. Um, which business leader inspires you the most. And, um,

leader inspires you the most. And, um, what what about what they operate you

what what about what they operate you kind of bring into your own day-to-day?

kind of bring into your own day-to-day? Well, [clears throat] I would say that I

Well, [clears throat] I would say that I won't go to the big names. I will go to

won't go to the big names. I will go to someone who is like up to date and, you

someone who is like up to date and, you know, um, he's still on on the bush, you

know, um, he's still on on the bush, you know. So I would say that Jake Cooper,

know. So I would say that Jake Cooper, the CEO of railway, I think he's doing a

the CEO of railway, I think he's doing a great job. I think, you know, he's

great job. I think, you know, he's making the all of the IAS infrastructure

making the all of the IAS infrastructure service much more accessible for

service much more accessible for developers and you can see that by their

developers and you can see that by their numbers. They also have like a brilliant

numbers. They also have like a brilliant page, a website that show their stats,

page, a website that show their stats, how many users, how other

how many users, how other um so I this is my inspiration. I'm

um so I this is my inspiration. I'm always showing my team listen and this

always showing my team listen and this is how things going to look like. I mean

is how things going to look like. I mean uh once you hit a point it's going to be

uh once you hit a point it's going to be like um it's going to blow. Now I think

like um it's going to blow. Now I think the same things the same thing he did

the same things the same thing he did for AI uh for like cloud infrastructure

for AI uh for like cloud infrastructure we want to do for the telecommunication

we want to do for the telecommunication protocols. We want to make it accessible

protocols. We want to make it accessible for everyone and we want people not to

for everyone and we want people not to think about the complexity and the

think about the complexity and the compliance.

compliance. Yeah. No, I think I think the railway

Yeah. No, I think I think the railway example is is is phenomenal and it's

example is is is phenomenal and it's often cited as inspiration by many

often cited as inspiration by many engineers who use the product

engineers who use the product day-to-day. Um, and so when you're when

day-to-day. Um, and so when you're when you're not building Dial um, and just

you're not building Dial um, and just like having a very busy life, what helps

like having a very busy life, what helps you quiet your mind? [laughter]

you quiet your mind? [laughter] I would say vacations that I have now

I would say vacations that I have now for the last years vacation.

for the last years vacation. Um, no. So, so my my my wife and my

Um, no. So, so my my my wife and my daughter um spending time with them,

daughter um spending time with them, this is like uh priceless, right? Um we

this is like uh priceless, right? Um we just talked about it, but yeah, I have a

just talked about it, but yeah, I have a newborn like one week ago. We had a

newborn like one week ago. We had a newborn and doing the the birds itself.

newborn and doing the the birds itself. I had a phone call from one of our

I had a phone call from one of our investors. So, I had to do I had to go

investors. So, I had to do I had to go outside and told my wife, listen, I'm

outside and told my wife, listen, I'm sorry, I'm going I'm so I'm going to

sorry, I'm going I'm so I'm going to miss the bird itself. Um don't be mad on

miss the bird itself. Um don't be mad on me so much. You know, I have nothing to

me so much. You know, I have nothing to do with it.

do with it. Right. Yeah. But yeah, I mean time with

Right. Yeah. But yeah, I mean time with my daughter. This is the prices. It's

my daughter. This is the prices. It's like an a small machine learning model

like an a small machine learning model like small LLM that is being learning

like small LLM that is being learning every day and and the the process of you

every day and and the the process of you know thinking and process of like

know thinking and process of like getting approved for me the HF it's the

getting approved for me the HF it's the RHF sorry the the reinforcement learning

RHF sorry the the reinforcement learning this is something for me that feels you

this is something for me that feels you know the joystick itself I'm not we're

know the joystick itself I'm not we're not a company who's making foundation

not a company who's making foundation models but with my I feel that I have my

models but with my I feel that I have my own foundation model and

own foundation model and group So crazy. One day you are saying

group So crazy. One day you are saying something, the other day you see how she

something, the other day you see how she reacts with your fillet.

reacts with your fillet. Yeah. Yeah. And they and they pick up

Yeah. Yeah. And they and they pick up that so quickly. It's unbelievable.

that so quickly. It's unbelievable. Yeah. Yeah. Wow. Well, I'm I'm so happy

Yeah. Yeah. Wow. Well, I'm I'm so happy to hear that you get those moments with

to hear that you get those moments with your with your newborn and and your

your with your newborn and and your other daughter and your wife. That's

other daughter and your wife. That's amazing. I do hope you get that vacation

amazing. I do hope you get that vacation soon, even if it's a mini vacation.

soon, even if it's a mini vacation. Um, so for you, just just final

Um, so for you, just just final question, what does success ultimately

question, what does success ultimately look like? that we are going to have

look like? that we are going to have 100% coverage that each one of the

100% coverage that each one of the agents out there are going to have its

agents out there are going to have its own phone number and ability to interact

own phone number and ability to interact with the real world.

with the real world. Amazing. Yeah, that's that's very

Amazing. Yeah, that's that's very concrete. I think you can put that

concrete. I think you can put that publicly internally and externally. So,

publicly internally and externally. So, that's awesome. Uh Gal, thanks a lot. I

that's awesome. Uh Gal, thanks a lot. I I really appreciate I know your

I really appreciate I know your schedule's super busy, so I really

schedule's super busy, so I really appreciate you joining me today. Behind

appreciate you joining me today. Behind the throw. [music] Behind the throw.

the throw. [music] Behind the throw. One move and the whole thing goes. Yeah.

One move and the whole thing goes. Yeah. We build. We [music] build. We build it

We build. We [music] build. We build it real behind the