I Built the Ultimate Team of AI Agents in 20 Minutes (Grok Bot)
607 segments
In this video, I'm going to be showing
you how to build the ultimate army of AI
employees for your business using the
latest Grok bot. If you didn't know,
Grok bot is the latest AI product by
SpaceX or and xAI, and they are
basically marketed as AI teammates that
can do work and not just talk to it like
Claude or ChatGPT, but they can actually
sign into your tools like email,
calendar, CRMs, and use the tools just
like you do. And they can even learn how
you do tasks. For example, you can
screen share and teach Grok bot how to
do a certain task, and it will learn
from that screen sharing just like a
real human would and perform that task
for you. So, in this video, I'll be
building six different AI agents
together to form an AI army of
employees. And if you don't know who I
am, my name is Adam Chan. I've been
running my own AI agency Legacy AI over
the past 2 years. And if you're a
business owner you want to work with me
and my team, you can use the link in the
description to book a free discovery
call with us. First, let me first show
you what we are actually building today.
So, we are building an an AI team of
employees. First of all, we have our
first agent, which is the chief of staff
agent, and it connects to nothing. Its
whole purpose is just to route work and
never does the work. So, they would be
basically leading a team of different
bots and telling them, "Okay, for this
specific task, which bot would be most
suitable for that task?" And for the
actual employee, the AI employee, there
are five different bots. The first one
is an email or inbox bot. All it does is
basically connect to Gmail or Outlook
and check for the emails and reply if
necessary or ignore. The second bot is a
calendar bot, which is used for
appointment bookings. The third one is
prospect bot. This one actually does
research over prospect over the
internet. I'll be using screen share,
and that bot is actually connected to my
browser or computer use, so that it
would actually do the research on its
own. The fourth one is the content bot.
And the content bot is used to draft
well any content for me. And we have the
ops bot, which is actually connected to
my CRM, and then it can check what the
deal status is and what's missing, etc.
So, the first step that you want to do
is actually go to xai.bot. So, you can
see meet Grok Bot. So, what you want to
do is download for macOS. So, I'm just
going to click download. And after you
download, you should have an app that
look calls Grok Bot. Once you open it,
you'll be asked you to sign in to your
account, and you can either do the xAI
subscription or a Cursor account. If you
don't know what Cursor is, Cursor is
basically an IDE, so basically where we
can actually code just like Visual
Studio Code, right? It's a development
environment. And one of their plan is
that with Grok Bot, you will need a
Cursor account. And you can either get
the one-week trial, or you can get the
ultra which is $200 a month. So, for
now, we're just going to be using trial,
and I'll just click continue. And I'll
just say sign in to Grok Bot. Just click
sign in. Open Grok Bot. I'll just click
sign in. Sign in. Open Grok Bot. And
now, we have the Grok Bot set up
complete. I'll just click next. I'll
just click next. Next. What do you use
every day? So, I'm just going to click
on Google Workspace, Slack, Notion,
LinkedIn. Look, I use GitHub, not really
the rest. I use Calendly. I do use
Calendly, Loom. Yeah, that's it for now.
Again, feel free to adjust it as you
need it for your business. I click next.
And now, we have the entire interface
for Grok Bot. So, the first thing you
want to do is what we want to create a
bot. So, I'm going to be creating the
first bot which is kind of the chief of
staff, right? So, I'm just going to
click on chief of staff. And I'll click
start it. And then it says, "Hey, I'm
your chief of staff. Here to take the
operation load so you can stay on what
they actually need to." Right. So, now,
I'm going to give it a prompt so that it
will actually know what it's actually
doing. So, I would say, "Your days are
already running through Slack, Notion,
GitHub, and Google." I can plug into
this now so I'm not saying blind. I want
you to connect this now. I'd rather just
tell me the first job. So, I'll just say
here, I'll just kind of copy the prompt.
I'll just say, "You are my chief of
staff. You manage a team of five
specialist bots, and you're my single
point of contact. I talk to you not to
them. So, my team will be have the inbox
spot, which is the email, calendar bot
for scheduling, prospect bot for
research on companies and people, the
content bot for turning raw material
into posting drafts, and our sales the
sales spotlight. Um so, read what I
asked on this side which specialist are
needed, and then if you need some more
than one, work out the order, research
before writing, facts before I reach,
and never ask two bots for the same
thing.
And then third one is delegate with
specific self-contained instruction.
Give the specialist everything it needs.
It can only see our conversations. And
then wait for them. If one comes back
saying they're wrong, send it back with
what is missing. Do not accept a fake
answer and pass it to me.
And then the fourth one is come back to
me with one answer, not a transcript,
not six reports.
And then we'll just have a kind of work
session. So, read what I asked um and
then just try to do the research before
writing and in delegation. Basically, a
lot of what I repeated. I'll just say I
run um
for context, I'll just say I run Legacy
AI.
Just click send. And as you can see, it
automatically create the team of the
bots that I needed needed to create,
right? I run them and you get one
answer, right? I'm the single point of
contact building the five specialists
now. Teams up, inbox, calendar,
prospect, content, and ops. You talk to
me, I run them, you get one answer. Send
the first job, right? So, let me just
check
um let's see the inbox bot description
in here. So, you say, "You are inbox
bot, the email specialist on Legacy AI
chief of staff team. You handle the
email triage, drafting, and follow-ups.
You report only to chief of staff, never
to the user directly. When given a task,
do it fully and return one clear
result." What you found, what you
drafted, the sands, etc. So, by just
messaging the chief of staff itself
configured the um inbox bot, right? So,
you can either like By the way, each of
these bots is have its own virtual
computer. So, it's kind of working Think
of it like a human, right? So, if you
hired a administrative um instructor or
an admin, they would have their own
computer. Similarly, each of these bots
would have their own virtual computer
they can work on. So, if I open the
screen
every second now. So, and see that this
is the screen that they're using. So,
they have their own like Google Chrome
to to search for for example, I'll just
choose weather today.
And it will say like the weather today,
right? This fine. They have their own
terminal
and they have their own file manager to
be using, right?
So, I can just let me just close this.
Close this. So, for the inbox box, I
actually want to connect it to Gmail.
So, we're saying installing Gmail now.
I'm just going to authorize.
You can close this tab. All right. So,
now let me just go to Grok bots.
Start I'll just say it's started.
So, it will just say you can see like
Gmail is connected pulling the inbox now
so I can send Chief of Staff a first try
it, right? So, it's connected to the
user Gmail.
So, it said messaged Chief of Staff,
right? So, it messaged the Chief of
Staff. Gmail is live etc. You can see
like this is
some emails that we have. Fairly charged
here. This is I forgot to pay my quota
subscription. Sorry, guys. But, anyway,
you can see from the Chief of Staff bot,
it will say got it. I'm the single point
of contact during the first five such
that teams operate. And then you get the
message from inbox bot. So, now the
Chief of Staff got the message from the
inbox bot and is now telling me what to
do. So, this is the advantage of having
like the the the the first layer which
is kind of the Chief of Staff bot. The
reason why is because
I if I have like a hundred of these
bots, I only need to talk to one bot,
not a hundred of these bots, right?
So, it will say okay, this is the email
that I'm getting.
So, it will say want me to open an
Arabic building so you can update that
card. So, I'll just say not now because
I've already done it. But, you can just
like to it will automatically
turn on turn on tasks
exactly that you need to do and it will
inform you. So, now I can just ignore
it. And now let me configure it the rest
of the bots. So, let's turn on calendar
bots. So, it will say hey, I'll take
scheduling, availability, meeting prep
from Chief of Staff and gets them done.
I'm checking whether the is successfully
set up so I can start, right? So, I'll
just click on connect to Google
Calendar. So, I'll just click on
authorize.
I'll just select all and click continue.
I'm going to click close this app. So,
you get two messages with Chief of
Staff. So, you can see the Chief of
Staff is communicating with the calendar
bot, right? So, uh let me just see the
Chief of Staff.
Um
Calendar bot's connect us. You know,
they report to me
with the Google Calendar when there's an
actual scheduling job. Nice. Uh we're
waiting for scheduling work from the
Chief of Staff, right? Nice.
So, now we have the calendar bot
connected. Um let's do the Let's do the
prospect bot bot first, because the
prospect bot I have to actually teach
them how to research a lead, right? I'll
just say, "I'm going to be sharing my
screen to train you how to research a
prospect properly, okay?" I'll walk me
through it. I'll watch you how you
actually research a prospect and lock
that in, right? So, I'm going to be
clicking on the prospect bot screen. I'm
going to open it, and I'm going to click
on teach a task. And now, this whole
screen is actually recorded for the
purposes of training that prospect bot
to do exactly what I teach you to do.
So, first of all, I'll just click on
Google Chrome. Let's say I have to
research a prospect. Uh let's say I'm
researching plumbing company right to
work with. And I'll just say, I'll just
type in plumbing company. Let's say we
just want the phone numbers and we're
doing cold calling um for these
companies, right? So, then you would
just go to like Rain Plumbing Services.
You go there, and then you will try to
find their phone numbers. You've got the
phone number, and we're doing cold
calling for these clients, right? And
maybe I just need to know like the name
of the businesses. So, if I go to the
website, maybe I want to I want the the
agent to research the rainplumbing.com,
the website, the phone number. And here
you go, there's an email,
service@rainplumbing.com.
So, this is where I can outreach to
them, right? So, now I'm just teaching
the prospect bot how to look up the
information and save this information,
right? So, this is now we have the the
email, right? So, this is actually how
we train it. And now I just click finish
recording and there's just
like learn from demonstration. So, turn
on screen going to reusable skill. So,
this is actually managed by cursor. So,
this is actually a function. Think of it
like a template. So, learn from
demonstration its name the description
is is is basically a tool. Turn on the
screen recorded demonstration in
computer reusable skill. And then you
would basically learn from demonstration
you're giving a screen recording of user
demonstration and tasks on your computer
and then etc. So, this is basically
instructions by cursor that you don't
have to build yourself. And from this
screen recording the prospect will be
trying to learn what is it. So, he's
saying watch the recording now I'll put
the steps out and turn into skill.
You're on LinkedIn looking at a person
looks like Daniel at services firm and
the company page for people watch for
correction. You are on Crown Plumbing
sites. Then Google searching plumbing
company emails. Right, exactly what I'm
trying to do. Knockout Plumbing shows up
in the knowledge panel watching the walk
full walk through now, right? Still
watching the full clip. Apparently I'm
pulling the tabs and search here to see
the results from that session so the
URLs etc. By the way, while it's think
while it's talking you can also create a
routine task within GrokBot so that
these are recurring tasks this agent
runs on schedule. For example, I can
create a routine where we say research
prospect, right? So, I'll just say
research prospect find the emails and
phone numbers and the websites and the
company names of different plumbing
companies, 20 plumbing companies of 20
plumbing companies, right? And then when
to run so I can adjust on the schedule.
So, I'll just say every day at let's say
9:00 a.m. Right, let's say 9:00 a.m. So,
it will just say a every day at 9:00
a.m. it will try to find the emails and
phone numbers and websites and the
company name
um of 20 plumbing companies, right?
Exactly. And then I can just turn it
active and inactive, right? etc. So, now
it's seeing the pattern which is Google
a local trade open GPP websites. Uh hit
contact check maps the phone address and
booking link and then hunt emails for a
hunter rather than the bulk list. So, I
can just message it to correct it,
right? Based on his training. So,
actually what I want you to do as you
can see from the screen recording is go
to Chrome and and then look up uh some
plumbing companies and then from Google
Maps you can see its phone number and
the website name. And then from the
website name I want to go into there to
see if you can find an email for it.
Essentially, you are a uh lead getter
bot so that every morning you would run
the research prospect routine in order
to get me leads for plumbing companies
for my AI agencies skipping bulk
mailing. Now, this is pretty This is um
pretty good conclusion, guys. Yep, so
now proceed. This is maps for phone and
website then the site for email uh
plumbing unless you name in the city.
Grabbing a first batch now so you have
something tonight. Right, so it's
actually go ahead going ahead and
actually you can see what it's doing.
It's going to the plumbing companies for
San Jose, CA. Save the method as
research for first leads landing
shortly. Right, so now it's actually
scrolling through this list and it's
actually using its own virtual computer
and actually performing tasks that I
trained it to do, right? Pretty cool.
And now it actually went to the website
on its own, right? And doing this task.
So, now while it's doing this task,
while it's actually just like a human
employee, we'll look through the website
and try to get the phone number and the
leads information. And let me now go do
something else, content bot. So, yeah,
this is basically just drafting um
content, right? Um this is fine. And
then now I have the ops bot which is the
the operation bot. I'm on the pipeline
and deal hygiene from here checking what
real due dates I already have so I can
get a real status together.
So, I'll just say, "Can you connect to
Slack, please?" Because all of our
clients live in the Slack space. Slack
is the right place if that's where the
clients live. Checking the connector and
getting it connected, right? Now, I'll
just click on authorize.
Uh I'll just click on my workspace and
then just click allow and close this tab
and go back to your GrokBot. It's like
it's connected, right? And it will look
through the workspace next and start
mapping where the clients actually live,
right? So, now as you can see, the
prospect job finished researching
everything and it got eight San Jose
plumbers from the Google Maps and three
of them have public emails and they save
all of them in a nice CSV, right? So,
let me just go down here.
CSV, let me just save it and let me open
Google Sheets just to show you guys uh
um what it looks like. Well, I'll just
uh
open. So, you can see it got all the
name, it got all the phone number, it
got the website, the one that they
found, and the one that has emails, and
the one that has address, and the Google
Map links, and the notes. Like, this is
pretty good. Like, as a first run, this
is pretty cool. Like, it's very very
useful, right? As a lead getter,
essentially. So, you can just train it
one time and it can perform the task
that it needs to, right? And let's go
back to GrokBot. So, we say pipeline 10.
Um so, yep, booked today. I'm going to
be chased. So, these are basically all
the deal Well, all of my existing
client. And then, the operations bot,
yep, this is from the op So, this is
actually from the chief of staff bot.
So, the ops bot actually tell the chief
of staff bot like what's happening. So,
just say uh pipeline 10 etc. So, the ops
will actually actually summarize all the
channel status and actually put together
a markdown file of all the status of all
of our clients, right? So, now let me
just talk to chief of staff and ask it
to run through a diagnosis. I'll just
say uh "Hey, can you look through just
for today just to see what appointments
I have and just to see is there anything
urgent for my team to do right now? And
can you also ask like prospect bot to
also start gathering leads for me for,
let's say, a uh London-based uh
restaurant, right? For my AI agency
clients."
And please also help me draft together a
LinkedIn post to do today to do with AI
voice agent. And just check the emails
to make sure that I didn't miss anything
important today. This is a big task. So,
let's see if the chief of staff can
actually delegate a task correctly to
the appropriate sub agent to be doing
the things correctly. Today's calendar
that was urgent tonight today on the
email sweep, restaurant leads and the
voice agent LinkedIn draft, right? So,
he messages all these team members, the
the off spot and the calendar board and
the prospect bots are all kind of
getting their ping and they are then
they're already trying to figure it out
essentially. Okay, so it's messaged five
agents. So, all five agents has been
messaged essentially. Offset nothing's
on fire tonight. Still waiting on
today's calendar. So, offset you said
nothing. See, so you see the the off
spot didn't reply to me. The off spot
replied to the chief of staff, right?
Calendar is clear for today. Nothing's
booked tonight. That's correct. Inbox is
clean of important mail today. Cloud
building is the only one I've really
like paid for it. That's fine. LinkedIn
draft is in waiting on the restaurant
leads and you'll get the full pack,
right? So, the content bot actually
finishes drafting a LinkedIn post and
then go back to the chief of staff. By
the way, if you want to like make it
refine it further because as a content
bot itself, like the model is not very
good at like writing fully fully human
like post. So, what you can potentially
do is you can literally go on the screen
and then you can write a LinkedIn post
yourself or looked at viral LinkedIn
post. And from there the content bot can
then learn from what a good LinkedIn
post would look like and therefore write
better LinkedIn post, right? So, if you
want the bot to be performing at better
things, train it like how you train an
intern. Right? Just do it on inside the
computer screen. That's the beauty of
using Rock Bot. There's very little
prompting as you can see, but there's a
lot of there is some prompting, but
there's a lot of more just showing them
how to do it. And to be honest, the the
the core idea of to of the entire Rock
Bot, the beauty of it is just how you
organize your team, right? And the
prospect bot is working, right? It's
looking through the the internet, right?
Okay, so you got like the LinkedIn
draft. Okay, obviously this is not the
best post you've ever seen. The But to
be fair, it this is just from one
sentence, create a LinkedIn post. It
doesn't know what a good LinkedIn post
is. So, exactly, if you want to improve
upon it, do the method I just showed
you. Or better, if you literally just
upload like files of like past viral
LinkedIn posts, if you want to. Now,
this basically said 10 restaurant leads
to average strongest flat iron dish room
full list with resources on my computer
at prospects London restaurants or MD.
So, basically, it's saved a markdown
file of different prospects in the
prospect book, right? And then it got
more independent just landed on the same
list. Strongest free at still file.
Okay, cool. Okay, can you I'll just say,
"Okay, can you send me the full CSV
file, please?"
Because technically, you can go to the
computer and then you can go to like the
file manager and you can see the
prospects London restaurants markdown.
I'll just say I'll just say open. Okay,
so it
looked at the markdown file is say the
client,
the area, website, the sources, right?
All the different notes they research
over its virtual computer and then just
going to close this task. Close this.
Close this.
Going to go back.
And now go to the chief of staff making
the CSV. So, now you've created a CSV as
well that I can download. So, you can
see this is in an Excel format, the
name, restaurant name, area, website,
why public contact, sources, page fit,
etc. Really nice. Yeah, so this is
pretty good from Joseph Bonnici example.
So, this is actually how you would be
able to build
an AI employee for your business. And if
you enjoy this kind of content, remember
to subscribe. And based on my
background, this is something that I
like to do at the end of every video.
So, if you're new to this channel, you
might not notice, but at the end of the
video, I would just I would just love to
pray with
people who are watching this video. So,
if you're comfortable with that, stay to
the end of the video. If not, thank you
so much for watching and I hopefully
I'll catch you in the next one.
So, let us pray. In the name of the
Father, and of the Son, and of the Holy
Spirit. Amen.
Heavenly Father, thank you very much
again for giving me a platform to
educate the public and everyone else.
And I hope everyone who's watching the
video have a blessed day ahead of them.
Through Christ our Lord. Amen. In the
name of the Father, and of the Son, and
of the Holy Spirit. Amen.
Thank you and I'll see you in the next
one.
Ask follow-up questions or revisit key timestamps.
This video provides a comprehensive guide on building a team of AI employees using the xAI Grok Bot. The creator, Adam Chan, demonstrates how to establish a hierarchy consisting of a 'Chief of Staff' agent that delegates tasks to five specialized agents (Inbox, Calendar, Prospect, Content, and Ops). He explains how to set up each bot, connect them to relevant business tools like Gmail, Google Calendar, and Slack, and train them through screen sharing and demonstration. The video showcases how these agents interact to perform real-world tasks such as prospect research, email triage, and content generation, emphasizing the efficiency gained by having a centralized AI management system.
Videos recently processed by our community