HomeVideos

Grok Bot: 5 Must-Try Use Cases for Work and Life (Full Tutorial)

Now Playing

Grok Bot: 5 Must-Try Use Cases for Work and Life (Full Tutorial)

Transcript

571 segments

0:00

Hey everyone, I think Grockbot is the

0:03

future of personal agents and I'm going

0:06

to show you how to build five of my

0:08

favorite bots. In this video, we're

0:11

gonna build an advisor to create and

0:13

manage your other bots. A YouTube

0:15

researcher to find out videos in your

0:18

interest. An ex Scout to surface the

0:21

most insightful and funny tweets from

0:23

your timeline. A digital Marie Condo to

0:26

clean up your email and find paid

0:28

subscriptions to cancel. A personal

0:30

concurge to help you get good prices on

0:33

travel and vacations. And as a bonus,

0:36

we're going to build a gamer to see if

0:38

Grockbot can play classic games like

0:41

Doom and Red Alert. We'll finish with a

0:44

discussion about privacy and whether

0:46

Grockbot should be your new AI daily

0:49

driver. All right, but first, let me

0:52

show you what makes Grockbot different.

0:55

A few weeks ago, I tweeted that chat GBT

0:58

was one feature away from building the

1:00

AI product that I really want, which is

1:02

a dedicated personal computer in the

1:04

cloud for running AI agents. Lee from

1:07

the cursor team DM'd me shortly after to

1:10

test Grockbot.

1:12

What exactly is a dedicated cloud

1:14

computer? You know the meme of people

1:16

walking around with their laptop half

1:18

open to keep their agents running? You

1:21

don't have to do that anymore if you're

1:22

using Grockbot. Basically, Grockbot

1:26

lives in a computer sitting somewhere in

1:28

SpaceX AI servers instead of in your

1:31

home. It has its own browser and

1:33

operating system that you can use and

1:36

logging to your favorite apps. Now,

1:38

let's do a quick comparison of Hermes,

1:41

Chat GBT, and Grockbot. Hermes I run on

1:45

my Mac Mini at home. It's on 24/7 and it

1:49

gives it a persistent computer, but you

1:51

have to go buy the machine and set

1:54

everything up for yourself. You can

1:56

technically also run Hermes on a virtual

1:58

private server, but that again requires

2:01

more setup work. Now, chat GPT work uses

2:04

plugins and a cloud browser, but the

2:07

problem is that browser at is right now

2:09

cannot stay signed in to your favorite

2:11

apps. The UX for chat GBD work also

2:15

feels scattered across chat work and

2:18

codecs. Let's take a quick look here.

2:21

It's just kind of like a mess of chat

2:23

threads here. There's chat GPT codeex.

2:26

There's like another tab up here for

2:28

work and chat. And it's just all really

2:30

confusing even to someone like me. In

2:32

contrast, Scrotbot gives you a

2:34

persistent cloud computer right out of

2:36

the box. And I think it's the easiest of

2:39

the three to set up and get going. and

2:41

his interface, as you can see here, also

2:44

feel a lot more focused and delightful.

2:47

I love the little bots and the

2:49

animations that they make when they

2:50

start working. Right, let's say look up

2:54

the weather. And you can see here that

2:56

there's some pretty cool subtle

2:58

animations. Each bot has a different

3:00

personality and style. And this is the

3:03

work of Kurser's great design team,

3:05

including Jenny Wen and other folks on

3:07

her team that I've interviewed in the

3:08

past. Overall, I think this UX feels

3:12

more like talking to a coworker than

3:14

getting lost in a 100 chat threads. So,

3:18

can Grockbot actually replace Chat GBT

3:21

and Codeex, which is my current daily

3:23

driver? Let's run through some of my

3:25

favorite bots right now. So, the first

3:27

bot that you should create after you

3:29

install Grockbot is an advisor. Here I

3:33

have my advisor ped to the very top of

3:36

my Grockbot list. With your adviser, you

3:38

want to start by telling it about your

3:40

work and life and then actually ask it

3:43

to give you some ideas for boss to

3:45

create. Here's what I told my adviser.

3:47

I'm a creator and founder of Behind the

3:49

Craft. I have two daughters and I live

3:50

in the Bay Area. I publish newsletter

3:53

posts and YouTube videos focus on

3:55

practical AI tutorials and podcast

3:57

interviews. And I also have a membership

3:58

portal over here. So based on all this,

4:01

suggest five bots that can help me save

4:03

me time or money. And for each explain

4:06

what it would do in a few sentences. As

4:10

you can see here, it created five bot

4:12

ideas. A daily AI news scout, YouTube

4:16

comment miner, membership concurge,

4:18

social media poster, and podcast

4:20

prepper. Right now, what many people

4:24

don't realize is that you can actually

4:26

get one bot, in this case, our advisor

4:29

bot, to help us create the rest of our

4:31

bots. Here I asked my adviser to build a

4:34

single bot that both researches other

4:37

channels and mine's comments and let's

4:40

call it YouTube researcher. Right? So

4:43

now let's go ahead and check out the

4:44

YouTube researcher use case. You can see

4:47

that the advisor kicked off the YouTube

4:49

researcher bot with this instruction to

4:52

create a morning brief to send me

4:54

YouTube intel every morning. But I want

4:56

to give it more specific instructions.

4:58

Right? First, I asked it to monitor

5:00

specific channels in my niche. And let's

5:04

scroll down here. And it's starting to

5:05

monitor channels. And here is its first

5:08

attempt at a brief. You can see here

5:10

that the brief is like just like very

5:12

verbose and hard to follow. It's like a

5:15

lot of stuff here. It did end up pulling

5:17

my YouTube comments and finding themes

5:19

there. So, that's great. But, this is

5:21

just like too much to process, right?

5:24

So, I gave it some more instructions. I

5:25

told it to give me a report in the

5:27

specific format which is first list your

5:30

top three content ideas. Then give me

5:33

the top five outliers for other channels

5:35

and then give me the top performing

5:37

overall. Limit your research to the last

5:39

14 days because YouTube with AI stuff is

5:42

very topical and also add the top common

5:44

themes back to the report. And by the

5:48

way, this is how you should be working

5:49

with AI to refine its output instead of

5:52

trying to oneshot something. Now you see

5:55

here that the report is much more

5:56

concise and easier to follow. Let's take

5:58

a quick look. Top three content ideas.

6:01

It wants me to make a video on my spec

6:03

scale that I plan to do soon. It tells

6:06

me to make a video on Grockbot that

6:08

we're making right now. And here are the

6:11

top outliers from the other channels.

6:13

Outlier meaning that it beat that

6:15

channel's performance for the last 14

6:16

days. Yep, pretty interesting topics

6:19

here that we can follow up on. and top

6:22

performing overall and also the top

6:24

common themes and it did more here and

6:26

then I told you to send a full report

6:27

again just to make sure it looks good

6:29

and now what I can do is basically

6:33

there's a daily job that sends me this

6:35

report every morning right just this

6:37

morning for example it sent me a new

6:39

report with the top angles the watch

6:42

list and comments and I basically have a

6:45

YouTube researcher that proactively

6:47

finds interesting topics for me to make

6:49

videos on and gives me feedback back

6:51

based on my user's comments. Okay, so

6:54

this is the process that we're going to

6:55

follow for pretty much the rest of the

6:56

boss that we're going to make, which is

6:58

we're going to kick it off with an

7:00

initial prompt. We're going to iterate

7:01

back and forth with it to make his

7:03

output good, and then we're going to

7:05

schedule a routine or job that runs

7:07

daily, weekly, or monthly to have it

7:09

proactively do work for us. Now, let's

7:12

actually see how we can follow the same

7:14

process with our X Scoutbot.

7:18

So, Grockbot is part of Space X AI,

7:22

right? So, it should have firstparty

7:24

access to X and Twitter data. And even

7:27

though I have an unhealthy addiction to

7:29

X, I still miss plenty of great content

7:32

or bookmark tweets that I never look at

7:34

again. Here's the prompt that I gave my

7:37

exbot. Find the most viral ex posts from

7:40

the last seven days in my niche and

7:42

create a weekly report with the top 10

7:44

tweets from people I follow, engaged

7:46

with, or bookmarked grouped into a few

7:48

categories. Include a full tweet copy

7:51

link and your analysis and end with

7:53

three content ideas for me to tweet

7:55

about. Next, you can see here that it

7:58

found my ex account and it automatically

8:01

created a routine to do this every

8:03

morning. And here is the initial output,

8:06

right?

8:08

So it found themes around playbooks that

8:10

people save not just like from Greg. 23

8:13

ways I use AI agents to grow my startup.

8:16

Chinese developer from loop to graph

8:18

engineering and more tweets from Greg.

8:22

Another theme that I found is just about

8:23

Grockbot codeex and chat GPT work

8:26

compared and actually used right a tweet

8:28

from Riley and from Ben who works on the

8:30

cursor team on the most loved internal

8:33

Grockbot use cases and more. And the

8:36

third theme is around creative workflows

8:38

that I actually am interested in tasing.

8:41

And they ended this report with the

8:44

three things to tweet about next. Some

8:46

ideas here. And just for fun, I asked

8:49

you to also give me the top five

8:51

funniest tweets from my timeline. I

8:53

think digging through it. And the number

8:55

one funniest tweet is as usual, OpenAI

8:58

and Anthropic kind of sniping at each

9:00

other and kind of joking,

9:03

right? And then there's some more funny

9:04

tweets down here. all kind of in my AI

9:07

knee niche. You can see here that it set

9:09

up a routine automatically and it sent

9:12

me a new report this morning. Now, let's

9:14

actually tell it to do something else.

9:16

I'm going to turn on voice here. Can you

9:18

send me this report to

9:20

peterbehindthecraft.com

9:24

and also make sure you include a full

9:26

tweet copy in your report. Send to my

9:29

email right now. All right. So, let's

9:31

see if Ashi is able to send this report

9:33

over email because I don't want to have

9:34

to open Grockbot each time to see it. I

9:36

want it just in my inbox every morning.

9:39

Awesome. Here's a report from Grothbot.

9:42

You can see here that it has this

9:43

themes. It's included the full tweet

9:45

copy. You can scroll down here and yeah,

9:50

I I think this looks pretty good, right?

9:52

And then it has down here, I'm sure, the

9:54

top three things to tweet about next.

9:56

Again, the process here is to give

10:00

Grockbot initial prompt, have it pull up

10:03

the information, iterate with it to get

10:04

the output and the report right, and

10:06

then either tell it to send you a daily

10:09

job through Grockbot itself or through

10:11

your email. In this case, I prefer email

10:13

because I like to wake up with the top

10:14

tweets to consume directly in my inbox

10:17

instead of having to open a separate

10:18

app. And by the way, it's able to send

10:21

me an email because I connected to a

10:23

bunch of plugins, including the Gmail

10:25

plug-in. Now, let's move on to a fun bot

10:30

that I call Marie Condo that cleans up

10:33

your digital clutter. There are three

10:35

places where clutter piles up, right?

10:37

Your email, Google Drive, and your pay

10:41

subscriptions. Now, I've connected both

10:44

Gmail and Drive in plugins to Grubbot.

10:48

And then I gave it this prompt. Audit my

10:51

Gmail, Google Drive, and recurring email

10:53

receipts and create a cleanup plan. Find

10:55

newsletters I already open. Find large

10:58

or abandoned drive files. Try to

11:01

identify pay subscriptions from email

11:03

receipts. And also you can hook up

11:05

Mercury MCP to get this data. Mercury is

11:07

the bank that I use for my business. And

11:10

then group everything into these

11:11

different categories. Use a number list.

11:14

And crucially, I told it to not move,

11:17

delete, or unsubscribe or cancel

11:19

anything without my approval. Right? The

11:22

last line is really important for a bot

11:24

that cleans up your files. You always

11:26

want to review what it plans to delete

11:29

before letting it remove anything. And

11:32

you can see here that first Grockbot

11:34

confirmed that it's connected to Google

11:36

Drive and Gmail. And then it actually

11:38

connected to Mercury MCP. I had to sign

11:41

in on the remote cloud computer to make

11:43

this work. And then it gave me this

11:46

initial report. This is a massive list

11:49

of stuff to clean up in my email

11:52

subscriptions to review and other things

11:55

that it should not touch. Right?

11:57

Honestly, this list is like pretty

11:58

overwhelming. It's incredibly long. And

12:00

I gave it some feedback of like, hey,

12:02

list it properly, categorize it

12:04

properly. It's still an incredibly long

12:06

list. Right? Again, this stuff is not

12:07

going to get it right in one shot. And

12:10

eventually I told it to just show me max

12:12

10 items in each list and get rid of all

12:14

the random labels that it has. So now

12:17

this is much more digestible. It has a

12:19

bunch of emails here, Google Drive and

12:22

pay subscriptions. And I gave it one

12:25

final piece of feedback to only show me

12:28

emails to unsubscribe to, Google Drive

12:30

files to delete and pay subscriptions

12:32

that I want to cancel. And then here's

12:34

the update list that I came up with. Now

12:36

it has this list as a number list and my

12:40

feedback is to actually get it to take

12:42

action which is email unsubscribe to

12:45

three and four 5 6 7 and 8 Google Drive

12:49

delete the extra tax return delete some

12:51

of these large files and pay

12:53

subscriptions cancel 11 and 15 right and

12:56

just as a general tip it's always good

12:58

to ask Grothbot or AI to give you its

13:01

response in a number list like this to

13:04

make it super easy for you to just tell

13:06

it to do things by just referring to the

13:08

number in the list instead of have to

13:09

type everything over again. I use this

13:11

pattern all the time. All right, so now

13:13

Crockbot is actually taking action,

13:16

right? This is where the magic happens.

13:17

You can see here is already unsubscribed

13:19

to a bunch of senders. It's trashed

13:22

three Google Drive files that we asked

13:24

it to and now it's cancelling Lovable

13:26

and Equipped Foods, which is a protein

13:29

company. Now to cancel lovable I first

13:32

have to sign to the cloud computer with

13:34

my lovable credentials which I did

13:36

manually and we found out that actually

13:38

lovable is already scheduled to be

13:40

cancelled. So lovable is done. Then it

13:42

asked me to sign to equip foods and I

13:44

did that and then it cancelled uh the

13:46

protein subscription for me. By the way,

13:48

Equip Foods is a great protein company.

13:50

I'm only canceling because I have too

13:51

many protein powders at home already.

13:56

Uh so it basically did all this work,

13:57

right? trashed a bunch of stuff. It's

14:00

unsubscribed to a bunch of emails and

14:01

it's canceled a bunch of subscriptions

14:03

to save me money. And you can see that

14:06

it did all this in around five minutes

14:08

when it would have taken me probably 30

14:10

minutes to an hour to do all this

14:11

manually. Marie Condo is very very

14:14

useful. Of course, and there's something

14:16

missing with this bot. I think we should

14:18

ask it to can you talk like Marie Condo

14:22

from now on? Give me an example because

14:25

right now it sounds too much like a

14:27

robot and not so much like Marie Condo

14:29

and let's see what it comes up with. All

14:31

right. So now uh Marie Condo is actually

14:34

going to talk like Marie Condo. You can

14:36

see Hermis and the singing have

14:38

completed their work. We thank them and

14:40

place them in the trash where they may

14:41

rest. Equip foods no longer sparks joy

14:44

and release prime protein subscription

14:46

with gratitude. [laughter]

14:47

Yeah. So now we can set up Marie Condo

14:50

to maybe talk to us every week or every

14:52

month to clean up our digital files and

14:55

spark joy, right? So this is definitely

14:58

a a bot that I recommend you setting up.

14:59

Just remember to ask it to give it

15:02

output in a number list instead of just

15:04

doing the cleanup for you so that you

15:05

can review it first, right? You don't

15:07

want it to accidentally delete some

15:08

important file. Now let's cover the

15:11

personal concurge bot. This is a bot

15:14

that I want to use for all my vacation

15:16

and travel plans. It has access to my

15:19

vacation document where I've listed my

15:21

December trip to Japan. Here's a quick

15:23

preview of the document. It has the full

15:25

iterity for Japan that I created with AI

15:28

as well. Let's go back to Grockbot. And

15:31

what I want Grockbot to do is I haven't

15:34

quite booked my flights to Japan yet.

15:36

So, I wanted to monitor for price

15:38

changes and alert me on what routes are

15:41

the best deal. Here's a prompt that I

15:42

give it. Read my vacation document and

15:45

monitor the exact flight legs for my

15:47

family trip. Get the dates, the best

15:49

options for each flight leg. Check

15:51

regularly and let me know when the price

15:53

improves. You see here that I found my

15:56

flight legs and is using Google flights

15:58

to check the prices. Now, you may be

16:00

wondering what's the advantage of using

16:03

Grockbot for this instead of just

16:04

setting up Google flight price alerts.

16:07

And the value here is that Grockbot can

16:09

understand my whole trip based on my

16:11

document and decide what's a better

16:13

option for my family. If you scroll down

16:16

here, the big finding is that a Tokyo

16:20

round trip is about $2,700 cheaper than

16:23

the open jaw flights that my dog

16:26

prefers. Right? My dog prefers it to fly

16:29

from SFO to Tokyo and then from Tokyo to

16:32

Fukoka, which we're going to, and then

16:34

from Fukuoka back to SFO. But actually,

16:37

Grockbot found a better itinerary. It's

16:38

about $2,700 cheaper to just fly round

16:41

trip to Tokyo instead of doing this

16:43

itinerary that I laid out, right? A

16:46

simple Google price alert would not have

16:48

found this. And now I ask it to check

16:51

every morning at 9:00 a.m. to see if the

16:52

price improves. And you can see here

16:54

that the price is still around the same.

16:57

The Tokyo round trip is still much

16:59

cheaper. Eventually, I can ask Rockbot

17:01

to also just go ahead and book the

17:02

flight for me or check into the flight

17:04

when the time comes.

17:06

And generally speaking, I think it's

17:08

always a good idea to have a travel

17:11

thread or travel bot to both help you

17:13

plan travel ahead of time and also when

17:15

you're at a location to help you book

17:18

amusement parks, to help you figure out

17:20

what to do every single day. So, I

17:22

imagine I'm going to be using this

17:23

personal travel concurge a lot. All

17:26

right, let's test one more use case with

17:28

Grockbot. Because Grockbot gives me a

17:31

dedicated cloud computer, I thought it

17:33

would be fun to ask it to install and

17:35

let me play some retro games. I asked it

17:38

to install Red Alert, Doom, and

17:41

Commander King. And you can see here

17:43

that it found the files and installed

17:45

it. So, why don't we just ask it now?

17:47

Let's say open Doom for us to play. And

17:51

let's see how things work out. All

17:53

right, looks like Doom is up. Let's open

17:55

it in our virtual cloud computer. And

17:58

here it is. And here we can start a new

18:01

game. Let's pick this episode.

18:05

And yeah, let's say Hurt Me Plant Penty.

18:08

Here's Doom. And here's kind of where

18:10

Grockbot falls apart a little bit.

18:12

Unfortunately, because it's on a virtual

18:14

cloud computer, the mouse isn't quite

18:16

configured right to actually play Doom.

18:19

For some reason, it's looking at the

18:20

floor all the time, and I can't seem to

18:22

adjust the mouse to actually look up.

18:25

All right. So, Doom doesn't quite work

18:26

here right now.

18:28

Now, let's actually try playing some

18:30

other game. Let's try asking to play

18:31

Commander King. I'm not sure if you guys

18:33

know this, but Commander King is an

18:36

awesome platform game that I played in

18:37

my youth. And now it's loading Commander

18:40

King. So, let's see what it comes up

18:42

with. Awesome. This is Commander King.

18:44

Let's see if it plays well or not. Let's

18:47

start a new game. One player normal

18:50

difficulty. And here we go. Here we go.

18:54

It's a very basic platformer, but I

18:56

remember really enjoy it in my youth.

18:59

How do I jump? I forgot how to jump. Oh,

19:01

so control is jump. And yeah, it plays

19:05

better than Doom for sure, but there is

19:07

some lag in the keyboard and mouse that

19:09

makes it difficult to actually control

19:11

the character. Okay, so um yeah, that's

19:15

just for fun. But basically, Grockbot is

19:18

not replacing your gaming PC or GeForce

19:22

now yet. But the fact that the agent can

19:24

install and launch these games on its

19:26

own computer gives you a sense of how

19:28

open-ended this could become and how

19:30

much potential there is here. Right? So

19:32

maybe Space X AI can use all the data

19:35

centers and GPUs in space to actually

19:39

deliver AAA games through this virtual

19:42

cloud computer. That would be a dream,

19:44

right? And while we're at it, I open

19:47

this file manager here. And here's all

19:49

the stuff that we installed into our

19:51

virtual cloud computer. It kind of looks

19:53

like Windows 3.1 a little bit. I'm not

19:55

sure what kind of thing it is. There's

19:57

Japan flights. There's other stuff here,

20:00

right? And then there's games that we've

20:02

installed. And also we have Chrome and a

20:05

terminal. So that's kind of our virtual

20:06

cloud computer. All right. Now,

20:18

All right. Now before we wrap, I want to

20:20

talk about probably the biggest hurdle

20:22

for Grobbox adoption, which is trust.

20:25

When I see a Google sign screen like

20:27

this on my laptop, I don't really think

20:29

twice before signing in. But because

20:31

this appeared on the virtual cloud

20:33

computer, I hesitated a bit because how

20:36

do I know that nobody else is seeing

20:37

this screen on the virtual cloud

20:39

computer? Now, if you go to the Grogbot

20:42

website, if you scroll all the way down

20:43

here, there is a note about privacy

20:45

here, which is how does Grockbot handle

20:47

my privacy? Grabbot uses the same cursor

20:50

SSO off and privacy mode you already

20:52

trust. Your cloud computers encrypted in

20:54

transit and at rest and there's no AI

20:57

training on top of it. Right? So, you

21:00

know, I'm willing to give Grockbot and

21:03

the virtual cloud computer access to all

21:04

this stuff because I'm an early AI

21:07

adopter, but I can see normal people

21:09

struggling to understand what this cloud

21:11

computer thing even is and kind of

21:15

hesitate to sign into their favorite

21:17

apps on this device. So, I think the AI

21:20

Asian platform that figures out trust

21:23

will be the first to get mass adoption.

21:27

But overall, I think Grockbot is a clear

21:29

sign of the future. We're moving away

21:32

from manually using our keyboard and

21:34

mouse to do work on our laptops to using

21:38

our voice to orchestrate a bunch of

21:40

agents that live in a dedicated cloud

21:43

computer. And Grockbot is the first

21:46

product to actually enable this. Now,

21:48

it's not quite my daily driver yet

21:50

because I think Chat GPT still offers

21:52

more for $20 a month while Grockbot

21:55

requires paying $200 a month to use on a

21:58

regular basis. But I think Grockbot has

22:01

much cleaner UI than chat right now and

22:04

is also very very capable. And overall,

22:07

it's just great to be in a world where

22:09

cursor and SpaceX AI are just as viable

22:12

a competitor as OpenAI and Anthropic. I

22:15

think cursor may even have the edge if

22:17

it can continue to support multiple

22:19

models from all providers.

22:22

So, Grockbot is available for free and I

22:25

definitely recommend downloading and

22:27

trying it to see a glimpse of the

22:29

future. I'll include some of the prompts

22:32

from this video in the pin comment

22:33

below. And I also have an exclusive

22:36

interview with the Cursor team on how

22:38

they built Grockbot coming up in the

22:41

next few weeks. Overall, I'm really

22:43

impressed by Grockbot. I think the team

22:45

really cooked here and I can't wait to

22:47

hear the story behind how they built

22:48

this. So, please like and subscribe if

22:51

you enjoy this video and I'll see you

22:52

next time.

Interactive Summary

This video explores Grockbot, a cloud-based AI agent platform that provides a persistent virtual computer environment. The creator demonstrates how to build and utilize five specialized bots: an advisor for managing tasks, a YouTube researcher for content strategy, an X (Twitter) scout for social media monitoring, a digital organizer ('Marie Kondo') for email and subscription management, and a travel concierge. The video also touches on the potential for retro gaming on these virtual instances, while candidly discussing the platform's current limitations, privacy considerations, and the trade-offs between its innovative UX and cost compared to competitors like ChatGPT.

Suggested questions

4 ready-made prompts