Enjoying this issue?
Get tomorrow's AI & engineering digest in your inbox — hand-picked, summarized, and always spam-free.
TLDR
Grok Bot's new voice mode lets users speak with individual AI agents, delegate tasks, and use computer use across connected apps via MCP, but its voice quality and speed lag behind ChatGPT's voice mode.
Key points
Grok Bot voice mode allows speaking with each AI agent individually.
Voice speed can be adjusted up to 1.5x.
Grok Bot can use computer use to open apps like Instagram.
Grok Bot connects to external apps via Zapier MCP.
The presenter notes ChatGPT voice mode has better voice quality.
Tools mentioned
Techniques
- voice mode
- multi-agent orchestration
- computer use
- MCP integration
Stop scrolling. Start reading smarter.
Receive the day's most important AI & engineering updates in one concise email. No spam.
Transcript (captions)
Grockbot just released a brand new voice mode inside of the platform. You can now speak with each of your individual AI agents inside of Grockbot. And I am extremely excited for this release. So
in this video, I'm going to break it down, see exactly how it works, and I'm going to compare it to Chatabt Voice Mode to see which one is better. So without further ado, let's dive right
into the video. Now, first of all, here is a release on Twitter. This went live just a couple of hours ago, and I finally got access to it, and that's why I wanted to make this review. I'm going
to keep this very simple and to the point without any fluff. So, voice. So, voice is rolling out on desktop and mobile over the next couple of days. I will say I have tried updating my
Grobball app on my phone a couple times now, but I still do not have access to this, which is probably the thing I'm most excited about. I'm very excited to pull up Grockbot on my phone, literally
just speak with my AI agent and have it go and orchestrate different tasks across the different AI agents I've configured inside of Grockbot. Now, if you want to test this out and you don't
already have Grockbot downloaded, there will be a link in the description to sign up and download this for the desktop app, but you could also download this on your phone as well. Now, let me
dive right into Grockbot. And as you can see, I have a bunch of different AI agents configured. These are my demo agents. These aren't my actual ones I use because the ones that I actually use
have confidential information that I can't show and there's no way for me to like hide that information, which is kind of a pain when it comes to filming these videos. But what I'm going to do
is come to my main orchestrating agent. And what this does is this coordinates with all of my other AI agents. Um, and this is the one that I spend the most time talking to. And it's able to go and
delegate different tasks to the other ones. If you're new to Grockbot, this is basically how it works. So, you could have you at the top basically coordinating with your different
executive AI agents. And then these AI agents are coordinating with the operators or I like to call them specialists. They are specialists in a given task. and I have about like eight
to 10 different agents running for me throughout the day. If you want a deeper dive into Grockbot, I have some amazing videos on this channel, so make sure to subscribe as well as I will link them at
the end of the video as well, so you can see exactly how I've set mine up and how I'm running them. All right, so inside of Grockbot, we're now going to see on the bottom right hand side of our screen
this new voice icon. So all I have to do in order to trigger this is to click on this button here. I'm going to set myself to mute and just explain very simply how this works. So, first of all,
we have our main agent right here that we're speaking with. We can speak with each of these agents individually. So, for example, I have my brand partnership demo agent right here. I could always go
into I could always go into this specific agent and have a conversation with this one instead of my main AI agent. So, each of these we can conversate with. Then, we can click on
settings and there's a couple of different things we could change. We can change obviously the voice of these AI agents, which I've only played around with a couple, as well as it shows this
cool little icon for the different AI agents that you have with that voice. This is cool because you could obviously have specific voices to each of these individual agents to differentiate
between them, so they don't all sound the same, which I think that is pretty cool. As well as I think this is really great. We could change the speed of our voice agent as well. So, we could change
this up to 1.5x, which I found very helpful since testing this out because it seems like they're very slow at communicating. I like my voice agents to speak very fast and get to the point.
And when it takes forever to say something, it's just very annoying. So, I speed mine up to 1.25 or 1.5x. And then we could also select the language. So, we have a few different languages
here. We have about like 12 to 15, maybe 20 different languages. Or we could just select auto detect and then it's automatically going to detect which language you are speaking in. So it
could then speak in that language back to you. Now what I'm going to do is I'm going to take this off mute and I'm going to speak with this agent. Hey, what's going on? What's your name?
>> I'm your main agent. The one who keeps you pointed at the stuff that actually moves YouTube AI for non techies. School and Lumi. What's pulling at you right now?
>> So as you can see that is very very slow and annoying. So, I'm going to turn this up to 1.5x and then go from there. Yeah. So, my main thing is I just have a lot of emails I need to reply to
tomorrow. >> That's exactly the kind of thing I root out rather than dig through myself. >> Okay. So, do you go and look at these emails or do we have a specific agent
that goes and does this for us? >> I sent it to the specialist who owns email. I don't dig through inboxes myself. I just make sure it lands with the right person. They'll come back with
a rundown of what's waiting and what's already drafted. So, tomorrow morning you're just reviewing instead of starting from scratch. Want me to have them flag anything time-sensitive
separately so it doesn't get buried? >> Yeah, that'd be very helpful. Um, and then what's the actual name of this agent? Like, give me the specific name. >> I don't have a name for them on my end.
I just route to email and it gets handled. I can ask them to introduce themselves when they report back though. They've already started on the triage. So, you'll have that breakdown soon.
>> All right. So, that's my main agent. Let me go to a separate agent now and test that one out as well. All right. All right. So, I'm now inside of my email agent demo here. I do want to say this
is not real information that you see um because this is confidential. So, I basically told it to change the names of clients, prices, etc. Just so you're aware. But what I'm going to do is now
have a conversation with this specific agent and I'm going to change the voice. Hey, how's it going? Hey, things are moving. Rivermark's contract is still sitting unsigned on
five AI audits worth 21,000. Brightfold's three projects are unsigned at 24. And Stackmir's been stalled on 148 for a while now. want me to nudge any of those today or are you focused
somewhere else? >> Yeah, I think uh if you could nudge the first one, that would be great. >> Drafting a nudge for Rivermark on those five audits.
>> Can you let me know when that nudge is sent? >> I'll flag you the moment it lands. Anything else you want queued up while that one's cooking?
>> Yeah, I want to see if you have the ability to use computer use. Is that possible? >> Yeah, I can get into your computer, browser, files, mail, all of it. That's
how I'm pulling your inbox and drafting those replies right now. What do you need done? >> Yeah, I want you to pull up I want you to pull up the actual computer use
screen and go and try to log into my Instagram. >> I can open the browser and head to Instagram, but I won't type in your login.
>> Perfect. Let's do this. >> Open on Instagram's login page. Nothing typed in. You'll need to handle the sign in yourself. Want me to watch the screen while you do it or move on to something
else? Instagram's up on the logged out wall. Username and password fields. Log in. Forgot password. Login with Facebook. Create new account. No sign in.
>> Thank you. That is all. Okay. So, that's pretty cool. Obviously, I sat here and I told it to go and use computer use to go and log into my Instagram or at least try to sign in. And it went and used the
computer to pull up Instagram. That's pretty cool. Not only can it do that, but it could also flag and basically delegate different tasks to my other specialized AI agents. So, I think that
this is pretty cool, actually. Now, if you've never used Grock before, it's pretty cool how we can connect the different applications that we use. So, for example, I'm having it go and pull
from my email inbox and I'm able to use voice mode in order to go and pull that information. So, it is almost like we have our own Jarvis that can work across the different applications that we use.
We're able to do this with the plugins inside of our Grockbot. So, for example, I have 24 different applications configured, which I could go through and scroll through all these. I have X, I've
got Granola, I have Gmail, Clay, ClickUp, all these different platforms that I've connected to. But let's say that there's an app I want to connect to that I don't see right here. Well, you
could do that with the Zap Your MCP. I personally have this Zap Your MCP server configured. And for example, I use Drip for my email marketing campaigns and there's no plugin inside of Grockbot. So
there's no so there's no way for me to connect Drip inside of Grockbot unless I'm using something like the Zapier MCP. So now what I could do is I could come here and toggle on my voice mode. And
I'm just going to ask it to go and pull information from Drip since it's using Zapier to bridge that gap. Hey, can you go do me a favor and pull up my latest campaign data from Drip via the Zapier
MCP? pulling your latest drip campaign data through the Zapier MCP. Now that that's in motion, I'll have the numbers as soon as they come back. Anything specific you're hoping to see
in there or just the full picture? >> No, I just like to see kind of the main details and main statistics there. >> Main details and stats it is. That's exactly what the poll is set up for.
Should be landing shortly. Now, I will say this is probably going to take a little while for this to pull this information because it does need to do a bunch of tool calls on the back
end. Um, because you can see at the bottom it's saying connecting to Zapier with our little agent icon dancing. Um, man, I just really like Grockbot. It's so simple to use. I love the interface
and I honestly see myself using this more and more every single day instead of using Claude because Claude used to be my main driver, but I'm using Grockbot quite a bit now.
>> 260917 Claude 2.0 O subject Claude just added docs and slides went out yesterday at 3:34 p.m. Lisbon time to subscribers from before September 9th. About 14,831
active subscribers. >> Perfect. Could you actually >> One heads up, open and click rates aren't coming through this drip API. >> Perfect. Can you actually just paste
that information inside of the chat and not give me a full lecture here? >> Latest drip send YT-260917 Claude 2.0. Subject: Claude just added. >> All right. All right. So, it's not very
good at listening to instructions cuz I said I wanted it to quit talking. Um, and I wanted to just send a text overview. But, it did actually pull this information from Drip via the Zapu MCP.
So, that's pretty cool. The fact that we could have it work across the different applications and give us these overviews and also go and perform actions across them as well. I want to compare this
very quickly to the chatt voice mode because I think that this is just so much better in terms of like the voice. I'm able to interrupt it and it sounds like a human more. as well as the fact
that I could trigger this to have codeex go and run specific workflows for me. Hypothetically, it would be amazing if Grockbot voice mode was as good as a chatbt voice mode. But honestly, I just
think OpenAI set the bar in terms of voice agents with these large language models. So, I'm just going to try this out. Hey, what's going on? >> Hey, I'm here ready to help with
whatever you want to tackle. >> Yeah, could you go and take a look at the last task we ran inside of Codeex? >> Sure, I'll pull that up. The last task was run Hakesfield soul
request. It stopped because the local credentials weren't configured. So no image was submitted. The next step is to run the setup in your max terminal and enter your key and secret.
>> Can you actually go and find the last image that we generated? >> Let me check that for you. Found it. It's displayed here for you. >> Okay, so it actually did go and pull up
the last image I generated inside of Codeex. This is pretty funny. This is for a video that I'm actually filming right now, so don't mind this image. But I still do think that chat should be
voice mode is a bit better than Grockbot. And there we have it. That is the new Grockbot voice mode. If you guys want more videos like this, subscribe to the channel. Let me know what you think
of this new update in the comments below. And I look forward to seeing you in the next one. Cheers. >> Free audio postproduction by alphonic.com.