Hey, can you tell me about the latest or

the upcoming call on Monday that I have

from cal.com?

>> Sure, let me pull that up from cal.com

for you.

>> I'm really excited about this. I've been

waiting for this for a long time. We

just got two huge improvements to voice

AI from both OpenAI in ChatGPT and Codex

and Anthropic in Claude. They both work

differently and are better at different

things. I'll explain everything. But

this is something I've personally been

very excited about for a long time

because until now the voice features in

ChatGPT or in Claude can't really do

anything. They can't really use tools.

You can talk to it, you can do a little

web search, but it can't use Connectors,

it can't control things. So I've been

building my own voice AI. I'll show you

that later, but I don't think it's even

relevant anymore because the new version

both OpenAI's version and Anthropic's

versions can now use tools depending on

the platform. And that means we can talk

to our AI and not just dictate. We can

talk, it can use tools, take actions,

and talk back to us. We don't even have

to hit the keyboard anymore. And that's

a huge deal. So first of all, OpenAI

made a huge deal about theirs. They made

a release video essentially bringing

voice back to ChatGPT desktop. But now

instead of ChatGPT desktop, whether it's

regular chat, work, or Codex, you can

have a voice agent that sits with you

and can control and take actions. This

will be very powerful moving forward

where it can actually read back its

responses and even control your

computer. One thing I want to point out

about this, to get it to work, you'll

see this chat icon, different from the

dictation, it will only show up in a

blank chat with no text entered. So hit

this, we see the famous animation, we

say, "Can you play Runaway by Kanye West

on Spotify?"

>> Yeah.

>> So right now it's controlling my

computer and still listening to me.

Let's see. So we're able to get it to

control our computer and play Spotify.

Now that's a very basic example, but

essentially what's going on here is

ChatGPT has computer use access and it

can run its voice agent as an

orchestrator to direct different agents

to do things on your computer. So it's

coding or it's work or it's anything

else. And I know a lot of us have been

using dictation tools like Aqua Voice or

Whisper Flow or Super Whisper or one of

the alternatives and that's really cool.

But being able to talk directly and hear

it talk back and not go back and forth

is really powerful. You can just

imagine. So ChatGPT and got this

advanced live voice on desktop, but

Anthropic did it a bit differently. But,

I just want to point out here, OpenAI

has the best voice model. They have this

real-time, which you can interrupt, it

will still process, and doesn't cut it

off completely. It's a great tool.

Anthropic is essentially the old type of

voice, which is speech-to-text and

text-to-speech. So, you talk, it

transcribes it, it sends it to the

model, it responds, and puts it back in

a voice for you. So, it's a bit slower.

But, Anthropic's release is nicer

because now it can use connectors, it

can use your MCPs in voice mode. They're

saying that the headline is now you can

choose different models. So, before

Claude Voice was using Haiku, their

smallest model. Now, you can choose if

it's Opus, Sonnet, or Haiku. Fable is

not an option, that's probably a good

thing cuz if you're like me, I'm already

at 100% of Fable usage. So, Claude Voice

is only available in chat. You have this

voice icon similar to ChatGPT's. Let me

just show you. First, we're going to use

ChatGPT Voice. I'm going to do it all on

my phone. And the reason we're going to

do that is I want to show you how it has

or doesn't have access to tools. So, I'm

going to open up ChatGPT Voice. Do you

have access to my Gmail?

>> Hang on. And not directly, no. If your

Gmail is connected through a supported

connector in this chat, then I can help.

Otherwise, I can't see your inbox.

>> It's connected to the connector.

>> All right. In that case, I can use it if

it's exposed to me here.

>> It's exposed to you here.

>> Checking. Okay, if it's exposed, the

quickest test is to ask something

specific like, "What's my latest

>> What is my latest email?"

>> Checking now. I can't see your Gmail

from here. The connector isn't actually

available to me in this chat.

>> Trust me, Gmail's connected here. I'm

even going to send it in the same chat.

What's my latest email? It's going to

use the connector. If it's not too

white, you'll be able to see it here.

Essentially, it just shows me my last

email that I sent to myself for this

video. It is a free 10-minute discovery

call between me and someone special.

Now, let's go to Claude. And you're

going to see why I'm so excited about

this. One of the MCPs or connectors I

built was a cal.com MCP. So, I could

have whatever agent I'm using do

research or change or send replies to

people that book calls with me. So, now

we're going to open up Claude on my

phone, and we're going to press the

voice. Hey, can you tell me about the

latest or the upcoming call on Monday

that I have from cal.com.

>> Sure, let me pull that up from cal.com

for you. It's just listed as a free

10-minute discovery call. So, there's no

specific topic attached to it. The only

details I have are that it's with Steve

Rogers at his email steve@avengers.com.

Want me to dig into anything else like

check your email for context on him?

>> Can you do research and tell me who he

is with Bright Data?

>> Sure, let me see what I can dig up on

him through Bright Data. Well, here's

the thing. That name and email are

basically a placeholder. Steve

Rogers@avengers.com

is Captain America. So, either someone

booked with a fake name or it's a test

booking. Uh there's no real person I can

dig up here.

>> So, I'm editing this video as I ride my

indoor bicycle. I realized I didn't do a

one-to-one test, right? So, we're just

going to open up Claude again on my

phone and I'm just going to ask it the

same question I asked ChatGPT.

Can you look at my email and tell me the

last email I got?

>> Sure, let me take a look. Your most

recent email came in today from cal.com.

It's a booking confirmation for a free

10-minute discovery call between you and

Steve Rogers scheduled for Monday, July

27th at 10:45 a.m.

>> So, apples to apples, it was able to use

the same Gmail connector that was

connected to Claude that ChatGPT wasn't

able to use even though it was able to

use it in its text interface. So, the

point is both Claude's new voice mode

and ChatGPT's voice mode have the

ability to take actions. They can

actually do stuff for us. The voice

technology for ChatGPT is a lot better,

but it's only running on your desktop,

only running on the computer for now.

But, Anthropic finally added connectors

or MCP use to their voice mode so that

it can actually take actions from your

voice. So, I don't have to sit at the

computer. I can have it use my

connectors either on the computer or on

my phone when I'm on the go to take

actions, to do research, to tell me

what's going on. And that is huge.

That's something I've been waiting for.

And I'm sure you can imagine all the use

cases where you want to direct your AI

to do something without having to type

on your phone or type on your computer

and just do it by voice. And now that is

possible. And I really believe voice is

the interface of the future for better

or for worse. I know that this will be a

huge productivity unlock and I'm so

happy that both Frontier Labs released

this on the same day. It It not be a new

model or a usage reset, but it's pretty

cool. Now, I just want to show you my

voice model. It's pretty funny. I'm

probably going to take it offline cuz I

won't need it anymore. Essentially, this

is connected to Better Claude. This is

my own personal version of Open Claude

that is running in Claude code on my Mac

mini. I'll link to that video above. I

use my Better Claude every day via

remote control, but what I was really

missing was the ability to talk to it,

similar to what ChatGPT or Claude didn't

have until today. I wanted to be able to

take actions or look at my databases,

look at my schedule, look at my emails.

So, I built it myself. And I modeled it

after ChatGPT's voice where it could

spawn multiple threads. So, I have an

orchestrator running on a small model,

and then it's able to send tasks to

Better Claude. So, let's just see. So, I

have it as a call, right? It's like a

phone call. What's up, man?

>> Not much, just standing by. What's on

your mind?

>> Can you tell me about my latest email?

>> Let me grab that for you. That's a

Netlify bot notification about a PR.

Looks like feedback on an editor

feature. Want me to pull the full thread

on that one? I can have Better Claude

check that for you.

>> So, yeah, basically, I built what I

could build, similar to what OpenAI was

offering. It's speech-to-text to speech

running on my Mac mini where I could

call into it and have it run tasks, have

it do actions. It's obviously slower and

not as efficient as anything that OpenAI

or Anthropic can do. And the creepy

thing is it was using my voice that I

created in my 11 Labs video like a year

ago. So, essentially, I was literally

talking to myself. Super weird. I don't

think I need that anymore. So, yeah, I'm

really into this about this. I like

doing research and getting updates while

I'm walking or on bike rides or driving.

This is huge. The voice interfaces from

OpenAI and Anthropic now have the

ability to take

So, I hope you found this video helpful

or insightful. If you have any questions

or feedback, drop them in the comments

below. Thank you guys for watching and

have a great weekend.