You're not going to lose your job to an

AI, but you're going to lose your job to

somebody who uses AI.

>> What jobs are going to be made redundant

in a world where I am sat here as a CEO

with a thousand AI agents. I'm thinking

about my CFO, my graphic designers, my

video editors. What jobs are going to be

impacted?

>> All of them.

>> So, we actually built a personal Jarvis.

>> No

>> Hey Chad, how do I solve merge conflicts

here?

>> This is my Jarvis agent clone system

that makes me 10 trillion a month and

swipes on Tinder at the same time. I'm

just joking. This is a flashy HTML file.

It has absolutely no logic behind it. If

you have the same for you page like me,

you see all of these flashy AI operating

systems going around. But in reality,

either has no backend system hooked up

or it's completely overkill and it

actually doesn't do anything. So, how do

you actually set up Jarvis? Well, you

could be using the voice feature inside

of Claude, but it's pretty, pretty

terrible. It can't even spell its own

name right. So, I set out to build an

actual useful system. And look at this.

We can close down Claude and close down

this. Then, we can click on this app

right here, click toggle sessions

overlay, and I can say, "Build me a

clean website using the premium website

skill and build a website for a bakery."

And this AI orchestrator is now going to

[music] dispatch this to Claude code.

>> Spinning up a fresh bakery site now with

premium website skill. I'll make it

clean and modern. You'll hear from me

when it's live.

>> can see how realistic it sounds as well

because it's using the new fish audio

2.1 pro model. And now Claude in the

background is going to be working on

this site. And this works with both

Claude or Codex. And look at the

results.

>> Golden Crumb Bakery site's live at

localhost 5173. Full React build with

hero menu, scheduler, contact form.

[music] Everything's ready to customize.

>> How insane is that? In the top right

corner, you can now see that bigger side

is finished. Claude is done. It has

built out this entire thing literally

just from our voice without stepping

inside of Claude code. And the thing

that's even crazier is I built it so

it's completely free to run. It's

completely open source. And later in

this video, I'm giving you the code so

you can run it as well.

>> And I'm going to give you my exact

prompts for how I built it out so you

can build your own version and customize

it to your exact workflow. [music] Let's

not waste any more time. Let's get into

it.

>> But first, before we build this entire

thing out, let's talk about how cheap it

actually is to run this because you need

to know the price. For the AI model,

that's the orchestrator, we need

something that's fast and cheap. And for

that, we're using Haiku, which we can

use through our Claude subscription,

which basically means it's free. We're

not paying anything extra for it. For

hosting, the app is going to be running

locally, so that's free as well. And for

the voice API, we need something that

sounds really, really, realistic without

being too expensive. And for that, we're

going to use Fish Audio 2.1 Pro. And at

the time of recording this, we can

actually try this model completely for

free as well. However, that is probably

temporary. It is, however, still much,

much cheaper than something like 11

Labs. Let's actually do a price

comparison. I'm just going to ask

Claude, what is the price difference

between Fish Audio 2.1 Pro and then the

11 Labs equivalent. Just going to

dictate that and hit enter. And what

we'll see is that 11 Labs flagship model

cost about $85 more per million

characters, which is around six to seven

x the price of Fish Audio. And for the

budget version, the Flash Turbo, it is

still three times the price. And that's

why I decided for this build to go with

the Fish Audio model. And then, of

course, for the actual coding task, it's

going to dispatch Babel and Opus, which

is also included in our already existing

Claude subscription. And I think I want

to set it up so we can do that with

Codex as well, so it can basically use

both models whenever I want it to do.

So, this will be a full-on AI

orchestrator where we can do everything

with just our voice, much better than

the current Claude dictation. And this

basically means that it barely cost us

anything also run this orchestrator. The

only thing that's going to cost

something at one point is the Fish Audio

API, but it's still pennies. It's really

not going to cost us that much to run.

Cool. Now comes the most important

thing, which is well, how do you build

this? This first step is crucial because

if you don't do this, it's going to take

much, much longer to build. We're going

to go inside of settings, then inside of

connectors, and then we're going to

click add and add a custom connector

because I want to connect with the Fish

Audio MCP before we even start building

this entire thing because then Claude

can grab a lot of information when we

are planning it out and when we are

building. So, this makes it much easier.

Then we're going to go to this docs

right here, which is the Fish Audio

docs. And what you can see is that we

have this URL right here, the https API

Fish Audio MCP. I'm going to drop a link

in the description to this page right

here so you don't have to search it up

yourself. And then we're going to call

the connector for Fish Audio, paste in

this URL right here, and click add. And

this is now going to check the

connection and connect Claude with Fish

Audio. It now says you're not connected

to Fish Audio yet, so we need to log in

with our account. It now says Claude

wants to access your Fish Audio account.

This will allow Claude to use all these

AI audio tools, which is exactly what we

want, so we're going to click approve.

Now we can open Claude, and then it is

connected. In order to set this up, you

need a Fish Audio account. I'm going to

drop a link in the description to where

you can sign up for that. The link does

give some kickbacks, so really

appreciate it if you use that. That's

the best way to support the channel.

When we are connected, I'll recommend to

set it to always allow on these settings

right here. This basically just means

that Claude doesn't have to ask for

permission every single time it does a

tool call. And then we can close it

down, and now I am just going to ask it,

"Do you have access to the Fish Audio

MCP?" Now check the connection, and

Claude can now check if it has the right

access. And there we have it. Claude

says that it has access, which is great.

You already saw that Fish is a lot more

affordable than something like Eleven

Labs, but I want to show you something

really cool and the reason that we are

using it. If I go inside of

text-to-speech, I can set it to the Fish

Audio S 2.1 Pro, that is right now free.

We then test it to see, okay, what voice

do we want to use for our orchestrator?

And the thing that's so cool is that we

can add these brackets right here. So, I

can write laughing, for example, and

then the next thing is going to have

that bracket. So, now we can click

generate speech. It's actually crazy.

Like, listen to this.

>> I just started two tasks.

>> [laughter]

>> You can sit back and relax.

>> We can also try this female voice right

here, Sarah, and then maybe remove the

dot right here, and then try and

generate that speech as well. Then we

choose Sarah up here, then generate that

too.

>> And then we can listen to it.

>> Just started two tasks.

>> [laughter]

>> You can sit back and relax. Um

>> It even says, um, it's crazy how

realistic the voices are now. The

technology is really moving insanely

fast. And at the time of recording this,

the new model, the Fish Audio S 2.1 Pro,

is actually free right now at the time

of recording this. So, great, go ahead,

create your Fish Audio account, again,

link in the description, and then we can

start building the actual orchestrator

app. It took me a while to figure out

this next part, but I've basically

prepared this starter prompt right here.

It basically explains exactly what it is

that we want to build and how it should

be built. And with this prompt, we can

literally build this out in literally 10

to 15 minutes. I'll leave a link to this

starter prompt in the description below.

But basically, what we'll do is that I'm

going to right click, create a new

folder on my desktop, and I'm going to

call this for the orchestrator. I don't

know if I spelled that correctly. Then,

inside of Claude code, it's very

important that this is running locally.

We're not going to build this in the

cloud. We're going to run it as a local

desktop app. So, make sure you have set

it to local. Then, for the folder, make

sure you open the folder that we just

created, which is the orchestrator. And

literally, all we need to do is just

copy this entire prompt, paste it in,

and say, "Please build this out for me."

And then hit enter. Claude will now have

all of the context that it needs, and

you can build this out in Codex as well,

by the way. It doesn't have to be

Claude. And it's now going to build out

this mini app for us. That's going to

run locally on our computer. And you can

see the first phase is that it detects

the machine, the hardware we are on, and

then builds the tech stack for our

specific computer. And stay around

because later in this video, I'm giving

you the source code as well. I'm going

to build this open source, so you can

grab this as well. This will take a

little bit of time to build out. Great,

it is now done. Took about an hour.

Let's test it. This part is crucial

because there will probably be some

bugs. The first thing we're going to do

is that we're going to set the API key

cuz we're going to use the Fish Audio

API key for the voice. So, I can say,

"Please open where I insert the API

key." Thank you. The Fish Audio API key.

This looks very cool so far. I think we

go through this onboarding right here.

We allow the microphone. That restarts

the app. We allow the speech

recognition. We might need to add

something like an API key to understand

and transcribe our words, but let's see

if it works. All right. Push to talk.

Shift command space. Hello. Release

captured. Continue. It works. Basically,

just make sure that we have both

authenticated Cloud Code and CodeX on

our computer. And now comes the field

where we have to insert the Fish Audio

API key. We get one of those by clicking

in the top left corner where it says

developer right here and then API keys.

And then we create an API key. Going to

call it orchestrator.

Click create. Never expire. I can see

that it doesn't allow us to paste in.

So, I'm just going to say, "Right now, I

can't paste the key. Gives me a error."

Let's have it fix that right quick.

There we go. Cloud had to fix some kind

of input bug, but now we can paste in

the Fish Audio API key. We can click

continue. And for our first project, we

can just click finish for now. Okay, so

how does this work? We have a couple of

agents. We have a voice. We have our

projects right here. So, I wonder if we

can dispatch an agent now. So, if I say,

"Please create an agent for Actually,

create a website for a plumbing

company." And there we go. After a bit

of back and forth, look at how cool it

is now. We now have the app that's

called Conductor. And I can click on

toggle sessions overlay right here,

which basically just shows our active

sessions. And look at this. Now I can

say build out a website or a car

mechanic and run it on 4,520

and open it in my browser when you're

done. The coordinator is now going to

give this task to either Claude or Codex

and now you can see the task is running

right now in the right hand corner and

it's going to say

>> Running car mechanic site on 4,520 and

opening it now.

>> How cool is that? And there we go, then

it opens and says

>> Car mechanic site successfully built and

running on localhost 4,520

All sections complete with animations.

>> On the top right corner you can see this

session is now complete, but I could

also say the car mechanic site that we

just built, please make that blue

instead of yellow. And then it's going

to look at the past sessions we have had

and it's going to update the session

>> mechanic site to blue on it.

>> How cool is that? And now it says right

here that it's working again and it's

going to update the theme of this site.

>> Car mechanic site theme successfully

changed from yellow to blue. Dev server

hot reloaded at localhost 4,520.

>> So let's give it a refresh. It's still

yellow it seems, the car mechanic site.

>> Which site is still yellow? Mechanic,

salon or plumber?

>> The mechanic.

>> Should I debug why the theme didn't

change or fix something specific?

>> Please pass this task to Claude and tell

it to fix it.

>> Sending fix to Claude. Comic and sidebar

turning blue now.

>> And there we go, now the website is

blue.

>> Car mechanic site theme conversion to

blue completed. Server restarted. CSS

updated. Hard refresh localhost 4,520 to

clear yellow cache.

>> Ah, it was a cache it said apparently.

And if we go inside of settings and have

all connect is connected, for example I

have Google Calendar connected, I can

now say please check my calendar for

what events I have tomorrow specifically

just answer how many calls I have. And

just like before it sends out the

session, but this time it asks Claude to

check our MCP into Google Calendar and

it's now going to return what events I

have.

>> Tomorrow has three calls scheduled.

Stefan Bujic at 1:00 p.m., Volker Hilmer

at 4:00 p.m., Thomas at 6:30 p.m.

>> Thank you very much for that.

>> You're welcome. Ready for the next task.

>> How cool is that? We can control

everything that Claude can do without

being inside of the chat. I can also

tell it to what are the latest emails

that I've received today.

>> Checking your latest emails from today.

Today's emails, cold pitch from Pavya,

blood clot login code, two strike

notices including a lost $79 dispute.

>> Pretty cool. And if I don't want to see

this overlay, I can just click this off.

I can also click shift command O, that

opens it, toggles it on and off, but I

can also do that in this window right

here, toggle off the session overlay.

And inside of settings, we can change a

couple of things. We can say, "Okay, do

we want to launch it login?" We can set

it to bypass permissions on Claude and

Codex. I'll probably recommend doing

that. We can set the speaking rate for

how fast the speaker actually speaks,

and we can see the projects that has

been created. I want this to sound even

more human, however. So, what I'm going

to tell Claude is that I want to make

the voice sound even more human. Do that

by including brackets like these that

fits like laughing or serious.

Basically, what this does is that this

gives the tone of the voice. So, we can

prompt Haiku, the AI orchestrator, to

give it in this voice, which is going to

make it sound even more human. Cool. I

tuned it quite a bit. So, now I can add

something like, "Hello, how's your day

going?"

>> Going pretty smoothly, thanks for

asking. How about you? What's on the

agenda?

>> That sounds a lot more human. And that's

because we're using these brackets that

basically gives it like a mood or way to

speak, which is really cool that we can

do that. So, now we have a full-on voice

AI system that can dispatch Claude code

sessions. So, we can basically build

anything we want. And in my opinion,

this is a lot cooler way to build rather

than inside a regular Claude session. If

you want to set this up yourself, then I

have left this prompt right here in the

description. I'm also going to upload

the code for the app that I've built, so

you can just grab that if you are on

Mac. You can also take it and just

transform it into a Windows app just by

taking my code and telling Claude to

turn it into a Windows app. That will

definitely work as well. For the

official audio account, I've left the

link right in the description below. And

if you want to join one of the biggest

AI communities in the world that takes

you from a complete beginner to building

an AI business in 7 days, and where you

get all of my Claude skills, then I'm

also going to leave a link in the

description for that. You can start with

a completely free trial. Thank you so

much for watching. Albert out.