How to Use Claude Code for AI-Powered Software Development

Name: How To Learn AI In A Weekend
Uploaded: 2026-04-18T12:01:35+00:00
Duration: 1 h 6 min 46 s
Channel: Ali Abdaal
Description: Summary and key takeaways on How To Learn AI In A Weekend: Summary & Key Takeaways, covering The AI Flywheel The AI Flywheel is a self‑reinforcing loop that

Ali Abdaal

Apr 18, 2026

•

66 min video

•

3 min read

YouTube video ID: qLMYOhaKcZs

Source: YouTube video by Ali Abdaal — Watch original video

PDF

The AI Flywheel is a self‑reinforcing loop that starts by letting an AI interview you about your business or work. It surfaces the most time‑consuming manual tasks, then helps you build automation tools for those tasks. While you build, you also learn how the AI and its underlying technology operate. That new knowledge fuels more ideas for automation, keeping the cycle moving forward.

Prerequisites for AI Development

Begin by downloading the Claude desktop app, which bundles Chat, Co‑work, and Code features. For faster interaction, install dictation software such as Whisper Flow so you can speak to the AI instead of typing. These tools give you the basic environment needed to start building AI‑assisted software.

Identifying Projects to Build

Ask the AI to pinpoint repetitive, manual processes in your workflow. Typical candidates include data collection, report generation, or content aggregation. Frame the problem in plain English—e.g., “I need to pull the latest videos from multiple YouTube channels”—so the AI can suggest a concrete project to automate.

Understanding the Terminal and Claude Code

The terminal is a command‑line interface that lets you control the file system with text commands. Claude Code runs inside this environment and acts like an architect who not only draws blueprints but also carries a toolkit to build, test, and fix code directly on your machine. In contrast, the Claude web app only provides design‑level guidance, leaving you to implement the solution yourself.

Installation and Security

Install Claude Code on macOS with the command:

curl -fsSL https://claude.ai/install.sh

Claude Code operates in a sandbox and requires explicit permission before performing destructive actions such as file deletion. Always review any suggested bash command, especially those containing rm. For extra safety, you can run a second AI instance to audit the first AI’s commands before approval.

Building a Practical Project: YouTube Tracker

Create a project folder – Open the terminal, run mkdir youtube-tracker && cd youtube-tracker.
Obtain an API key – Generate a YouTube Data API v3 key from the Google Cloud Console.
Describe the task – Tell Claude Code, “Pull the 10 latest videos from 50 channels.”
Run a local server – Use python3 -m http.server 8000 to host the project locally for testing.
Deploy – Once the tool works, deploy it to Vercel or GitHub Pages so your team can access it online.

Claude Code will create the necessary files, write the Python or JavaScript code, run it, detect errors, and apply fixes automatically. The YouTube Data API v3 supplies public channel data such as titles, thumbnails, and view counts without manual scraping.

Scaling and Future Iterations

After the initial tracker is functional, iterate by adding features like automated email reports, integration with Stripe for paid access, or deeper analytics. For more complex automation, consider OpenClaw, an advanced, agentic version of Claude Code that can coordinate multiple AI agents via the Model Context Protocol (MCP). Each new capability expands the AI Flywheel, generating fresh automation ideas and further boosting productivity.

Takeaways

The AI Flywheel creates a continuous loop where AI identifies manual tasks, builds automation tools, and teaches you new technical skills.
Claude Code operates directly in the terminal, allowing it to write, run, debug, and fix code on your machine without manual intervention.
Installing Claude Code requires a single curl command, and all destructive actions must be explicitly approved to maintain security.
A practical YouTube Tracker can be built by creating a project folder, obtaining a YouTube Data API key, describing the task to Claude Code, and testing locally before deployment.
Scaling the initial project with tools like OpenClaw and Stripe extends automation capabilities and reinforces the AI Flywheel for ongoing productivity gains.

Frequently Asked Questions

What is the AI Flywheel and how does it improve productivity?

The AI Flywheel is a loop that uses AI to discover time‑consuming tasks, builds automated tools for them, and teaches you how the AI works during construction. This cycle continuously generates new automation ideas, turning each iteration into a productivity boost.

Who is Ali Abdaal on YouTube?

Ali Abdaal is a YouTube channel that publishes videos on a range of topics. Browse more summaries from this channel below.

Does this page include the full transcript of the video?

Yes, the full transcript for this video is available on this page. Click 'Show transcript' in the sidebar to read it.

Helpful resources related to this video

If you want to practice or explore the concepts discussed in the video, these commonly used tools may help.

Ergonomic Mechanical Keyboard For Programmers Recommended

Provides tactile feedback and comfort for long coding sessions in the terminal, reducing strain during intensive development work.

Amazon →

Noise Cancelling Headphones For Deep Work

Helps maintain focus and concentration while navigating complex terminal commands and debugging software projects.

Amazon →

Ultrawide Computer Monitor For Coding

Offers extra screen real estate to view the terminal, code editor, and browser simultaneously, which is essential for the AI-assisted development workflow.

Amazon →

High Quality Usb Microphone For Dictation

Improves the accuracy of speech-to-text software like Whisper Flow, allowing for faster, hands-free communication with AI tools.

Amazon →

Book On Python Programming For Beginners

Provides a foundational understanding of the language used to build the YouTube tracker and other automation tools.

Amazon →

Links may be affiliate links. We only include resources that are genuinely relevant to the topic.

Summarize another video

Full Transcript YouTube

There is an incredible piece of software
called Claude Code that I've been using
for the last 3 months and it has
completely and utterly changed my life.
And this video is the video that I'm
sending to my friends, my family, my
team members who have not yet started
using Claude code. Maybe you're like me
a few months ago where you were
intimidated by the thought of typing
stuff into a terminal. Maybe you're
intimidated by the name Claude Code.
You're like, "Ah, but I'm not a coder.
I'm not a developer. I couldn't possibly
use Claude code." Maybe you're still
stuck in that paradigm of talking to
Chad GBT or Claude on the web app and
thinking that that is all AI has to
offer. Or maybe you're like a lot of
people I know and you're like, "Man, I
know I should get into this Claude code
thing or this Claude Co-work thing. I
just don't really have the time."
Hopefully, this video will help you
solve all that. And this video is going
to introduce you to the idea of the AI
flywheel. You basically get AI to
interview you about what you're doing in
your business, your life, your work.
Based on that, you get the AI to suggest
what cool things you can use the AI to
help build that would help save you time
and make you more money. And then in the
process of getting AI to build those
things for you, you are also learning
about how the AI works. The more you
learn about how the AI works, the more
that feeds into giving you ideas. It's
like you're getting a firmware update in
your brain where now you know what's
possible, which means you can then build
even more things to help save you time
and make you more money. Using these
sick AI features, I've built stuff for
my business and for my customers that
literally add value to our customers
every single day where people are raving
about the tools that I've managed to
build for them over the last couple of
months just by learning how to use cloud
code. The final thing to say before we
get into the video is that every single
day the gap between the people that just
use the free version of Chad GPT and
talk to it on the web versus people that
know how to build stuff using flawed
code even if you don't know how to code.
Every single day that gap is getting
wider and wider. And what I hope this
video will do is give you a sort of
beginner's guide on how to narrow that
gap. There are timestamps to everything
down below in the video. So let's get
started. Okay, so there are a couple of
prerequisites that will really help you
if you have these downloaded before we
try doing this stuff. The first thing is
you do not want to be using the web app
for something like Claude. Instead, you
want to download the desktop app. This
is totally free, but when you download
the claw desktop app, you will notice it
says chat, which is the thing that you
are used to. It also says co-work, which
sounds somewhat scary if you haven't
used it before. And it says code, which
sounds even scarier if you haven't used
it before. The second thing you want to
do is you do not want to type. You want
to have some kind of dictation software
installed that does a speech to text.
There are lots of these. In my case, the
one I use is called Whisper Flow. I'll
put an affiliate link or something in
the video description if you are
interested. And the cool thing about
Whisper Flow is I can hit Fn spacear and
then I can say whatever I want and then
it will appear on screen within like
milliseconds. Boom. This is how it
works. This is a far quicker way of
speaking to an AI rather than having to
type to an AI. And those are basically
the only two prerequisites you need. So
step number one is we're going to ask AI
what we should build with AI. I'm trying
to learn how to use AI. My current uh AI
skill set is I just use uh chat GBT and
Claude and use the chat feature. But I
recently watched a video that tells me
about co-work and how claude code is the
best thing ever. So to that end, I want
to learn how to use AI, but I don't want
to just randomly try and learn how to
use AI by following tutorials. I
actually want to build something that's
really useful in my work or in my life.
The problem is I'm not really sure what
to build. So I want you to ask me
questions and interview me and help me
figure out within the context of my work
like I want to build something that will
save me time meaningfully uh or help me
make more money in the context of my
business. Uh and in the process I want
to sort of learn how to use things like
Claude Co-work and Claude code as well.
Okay. So my business actually does I
guess I would say three main things. Uh
firstly we do content. So I have a
YouTube channel and I post stuff on
Instagram and occasionally stuff on
LinkedIn. And the content is trying to
sort of help people build a life they
love and sort of change their lives and
stuff personal development things. Uh
the second thing the business does is we
have um an online business school called
the lifestyle business academy that
helps uh people start and grow sixfigure
lifestyle businesses. And then the third
thing that we do is that we are building
various software products. We've built
an app called voice pal which is a sort
of AI ghostwriter. We're building an app
called Superfocus which is a sort of
productivity thing and we are building
an app called uh a platform called
creator grid which um is a platform to
help creators uh grow and monetize their
audience. Now obviously all of this is
in the context of me owning a business
and doing various different things. Uh
but in the context of you maybe having
your own job then you can ask it to
interview you about what you do in your
job. if you are trying to go for a raise
or just trying to improve your job
performance or anything like that.
Actually, you know, one thing that I
wish I didn't have to do or I wish that
my team didn't have to do is that we do
a lot of data scraping. So, for example,
for YouTube, for Instagram, for
LinkedIn, we spend, you know, my team
spends a lot of time each week looking
at analytics and looking at view counts
and adding up saves and uh impressions
and stuff. And then some of that data we
need to send to sponsors. Some of it we
use to figure out whether our own social
media stuff is going well. We also need
to do this across our competitors. I
mean, I don't really think of them as
competitors. I think of them as
colleagues. But we want to keep an eye
on what our sort of friends in the space
are doing so that if someone is growing
particularly fast or hits on a format
that is particularly good, then we can
experiment with that format ourselves.
And so all of this stuff takes actually
many many hours for my team uh every
single week. So I wonder if there's some
cool stuff we can build that will help
us automate some of that. Do you think
that would be possible? 100% possible.
And honestly, this is a great candidate
because it ticks both boxes. It's
genuinely painful hours per week of
manual work. And it's the kind of thing
where you'll learn a ton by building it.
Now, this is good. This is asking me
really, really useful questions about
how we can automate this like data
collection process. I'm not going to
read all the responses. You can pause
the video whenever you like just to read
the stuff because I know you can read.
Okay, so the current workflow it's like
um yeah, Nicole on my team and Becky and
my team literally goes to YouTube Studio
or to Instagram analytics. Basically,
they go to Instagram insights, they
write down numbers, and they manually uh
write the numbers into a Google sheet
and then use the sum function to sort of
count up like impressions this week or
something. To be honest, I I actually
don't know what's available on Instagram
insights. Maybe they don't have to do
that anymore, but either way, it's a
manual process of copying and pasting
numbers into a Google sheet. Second, how
many of these colleagues are you
tracking? Are we talking five channels
or 50? More like 50. 50 on YouTube, 50
on Instagram, 50 on LinkedIn, that sort
of thing. And what does the output look
like? It sort of goes into a Google
sheet, but then it sort of goes on
random notion pages that are all over
the place. It very rarely goes into
slide decks, but yeah, primarily Google
Sheets and Notion databases. That's
where it goes once it's collected. And
then, you know, we always think, "Oh,
it'd be nice if we looked at it more
often, but then because it's all over
the place, it's kind of hard to look at
more often. In reality, what we often do
is just chuck it into AI and be like,
"Hey, can you give us some interesting
insights?" But I think it's the the
collection of the data itself that like
is kind of annoying. You know what?
Let's not worry about the sponsor thing.
I think let's just focus on the
competitor tracking. And what do I want
to know? I think yeah, it's how many
views did the recent videos get. It is
also what topics are performing well for
them. I mean, if we could get a sense of
like if we're thinking about YouTube, um
I I never want to watch the video itself
cuz I don't want to accidentally copy
it. But I'm very interested in what the
title and the thumbnail was. And if
we're thinking about Instagram reels, I
you know, in a dream world, we'd be able
to transcribe the reel to get the hook
cuz I'm always interested in what's the
first line. But if it's an Instagram
carousel, like what was the format? So
yeah, the challenge with Instagram is
that we've got all these different
formats. So yeah, I just don't really
know how we would go about that. All
right, so this is telling me we're going
to start with YouTube. It's the easiest
win. YouTube has a public API where you
can pull videos, titles, thumbnails,
view counts, publish dates for any
public channel. You give it 50 channel
names and it gives you back everything
you want. That's your week one project.
Sick. Instagram is harder but doable.
Meta's API is more locked down. So we'll
need to get a bit more creative here.
Here's what I imagine. You open a
dashboard, could be a web page, and you
see something like, "Here are the 50
channels I track. Here are the last 10
videos sorted by views. And here's
what's popping off this week." That
sounds great. You can see thumbnails and
titles at a glance without watching
anything. And you could have AI
automatically tag the topics so you can
spot patterns. Like three of my
colleagues posted about morning routines
this week and they did well. Sick. The
beautiful thing is that this is a
perfect project for learning clawed code
because it involves real API calls, real
data processing, and optionally a real
front end, but none of it is so complex
that you'll get stuck for days. Does
that sound like the right thing to focus
on? And do you have a preference for the
output looks like? Would you rather see
a nicel looking dashboard you open in a
browser, or would you rather just
automatically update a update a Google
sheet that your team uses? This is so
sick. Like, man, even me reading this
back right now, I'm like, "Wow, Claude
has such a good understanding of what
I'm trying to do, what I'm trying to
build, it's suggesting things that I
didn't even think of." What a time to be
alive that you can literally ask the AI
to build stuff for you and also teach
you how to build the stuff in the
process. Like, in the past, I paid
thousands of thousands of dollars for
like, you know, automation consultants
to ask us these sorts of questions to
figure out what we can automate in the
business. And none of what they've said
of the $50,000 or so dollars that we've
paid to automation companies and
automation agencies and consulting
whatever people comes close to the level
of clarity that Claude just got through
just a handful of exchanges with me.
This is incredible. Now, once you get
good at using AI to build all of these
different features that are going to
help you save you time and make more
money, what you might need is a web
host. You might need a place to host the
various programs that you're creating
and stick them on a website and make
them accessible via the web, either to
you personally or to you and your team
or even to the public if you want to
like make money from the thing. This is
where Hostinger comes in, who are very
kindly sponsoring this video. Now,
Hostinger has a bunch of different
features that are all super helpful if
you are trying to build a business or
just improve your life in general. The
first one is Hostinger Horizons. This is
basically an all-in-one AI website
builder. You basically just describe
whatever you want and Hostinger Horizons
will basically code up the app entirely
for you. It'll build you a fully working
mobile friendly web app with absolutely
no coding required. And just by talking
to it, you can refine things like the
content and the colors and the
functionality. They've even integrated
features that handle user loginins
behind the scenes in a super easy way
and also billing through something like
Stripes. You can even start charging
people for the various different web
apps that you create. When you're ready
to go live, it's just one click and then
you get a domain, you get your email,
and you get your web hosting all
included as part of the package. In
addition to Horizons, there's also the
Hostinger VPS service that lets you host
a virtual private server, which if
you're doing particularly fancy stuff
with Claude Code or with OpenClaw with
things like that, you can host it
directly on Hostinger. Domains are
included with all of the Horizon plans
and you also get free hosting included
with the annual plans. Plus, they have a
30-day money back guarantee. So, there's
genuinely no risk in actually just
giving it a go. So, if you want to turn
your idea into a real business, head
over to hostinger.com/aliabdal.
And you can get 10% off if you use the
code Aliabdal at checkout as well. But
thank you again, Hostinger Horizons for
sponsoring this video. And let's get
back to it. Now, at this point, there's
also something super interesting that I
would love to talk about here. So, I
happen to know what an API is, I think.
But if I didn't know what an API was,
it's like my my rule for the AI stuff is
anytime I don't understand something,
what do I do? I ask Claude to explain it
to me. Okay, this sounds really cool.
But um one of the key things I want to
do is I want to understand the
foundations and the philosophy and you
know I'm even interested in like history
uh you know the history of stuff like
this. So when you say API, what does an
API actually mean? Can you help me
understand what an API is? Why it's
useful and how did APIs get developed?
Uh like why why are they a thing in the
first place? So this sort of thing is
how for the last 2 months I have felt
like every single day I'm learning so
much about the world. I'm learning about
the history of computing. I'm learning
about the history of technology. I'm
learning about like random stuff because
when something interesting happens and
claude code or claude coowork or claude
or gbt codeex or whatever is asking me
to build something or it's running a
terminal command that I don't understand
while it's doing the thing I will just
open a separate window and I'll ask it
to teach me about the history of the
thing. For example, when I was
installing openclaw and I was trying to
do some security audits on it and I was
really trying to understand like what
does SSH mean and like what you know how
does a terminal work and what's a secure
shell and stuff and I ended up sort of
spending half an hour going down a
rabbit hole of like understanding the
history of like the telegraph and how
the telegraph morphed into Morse code
and how Morse code then became like
telephone lines and how like the US
military in the cold war was worried
that like you know the telephone
exchanges would get bombed by the the
Soviets and so they had to figure out
this method of information interchange
that involved like packets and so the
packets was invented and that was then
this protocol called TCP. P IP which was
then the precursor for like HTTP which
was the precursor for the worldwide web.
And then when the worldwide web was
invented there was all these people
getting hacked and there was this
student I think in some university in
Poland or something he was pissed off
that his like university got hacked. And
so he like coded up this thing called
SSH which was like a secure way of
transmitting information across packets.
And then this guy ended up trying to
overcommercialize it. And so someone
else made this thing called open SSH
which became like a free platform. And
then this is encryption and public and
private keys and all of this sort of
stuff. And now when I'm typing in SSH
dot dot dot dot into my terminal, I know
what that means and I can appreciate the
history behind it and I can appreciate
why it was such a big deal and how it
was invented. It's just so cool. Ah, and
Claude is just again really good at
explaining stuff. Let's go deep. The
problem before APIs existed. Imagine
it's 2005. You're building a website.
You want to show a map on it. Before
APIs, you had two options. Build your
own mapping system from scratch. Insane.
Or literally take a screenshot of Map
Quest and paste it into your page. I
don't know what Map Quest is. I might I
might ask it about it. This was the
fundamental problem. Software was like a
series of world cities. Each application
has its own data locked inside of it.
And if you wanted to use that data in
your own thing, you had to either
recreate it yourself or do something
gross like write a script that pretends
to be a human clicking buttons and
scraping the screen. People actually did
this and it was as fragile as it sounds.
YouTube changes one pixel on their page
and your whole system breaks. So what
actually is an API? API stands for
application programming interface. But
that name is honestly unhelpful. Here's
a better way to think about it. So good.
Again, this is the flywheel idea. AI is
helping me figure out what to build. In
the process of figuring out how to build
it, I'm also learning about stuff. So
I'm not just blindly writing commands
into a terminal and hoping for the best.
I'm actually increasing my understanding
of it. The reason it's useful to
understand this stuff rather than just
like blindly do it is because when you
start understanding these things, you
get it's like you get a firmware update
in your own brain. Which means even if
you don't really care about like the
history of computing cuz you don't find
it interesting, it's still useful for
you to know that what an API is and what
HTTP requests are because the fact that
you now know that those things exist
means that you might realize, wait a
minute, huh, there's this program that I
use in my work called blah blah blah. I
wonder if they have an API. and you
Google does blah blah blah have an API
and if you weren't familiar with the
word API you wouldn't have known that
you could even ask that question and
then you might find no it doesn't have
an API but it does have an MCP server
now if you are not at all curious and
you're a total dumbass and you don't
then pull on that thread you're like MCP
server oh it sounds technical I don't
care but if you're an intelligent person
you would think huh MCP server I think
I've heard of that but I don't really
know anything about it what's an MCP
server claude and you understand what an
MCP server is and you realize wait a
minute there's all these other ideas for
things that I could build that will help
me save time and make more money that
involve MCP servers. Could I build my
own MCP server? What about my own MCP
client? Okay, this is super interesting.
And you start to learn this stuff
without even trying to learn the stuff.
Like you never had to sit down and watch
a tutorial about like what the hell is
an MCP server and how does it work? But
purely by exercising your curiosity and
pulling on the string of like, huh, APIs
are cool. Does not have an API? It does,
but it also has an MCP server. Oh,
that's cool. Like what's an MCP server?
Oh, sick. It's a way for like AI things
to chat to each other. And it was
invented by Anthropic back in like 2024
or something. Like sick. Now, basically
everything has an MCP server, and you
can even build your own. Sick. That just
gives me so many ideas for like other
things that I can build with my stuff.
Now, we're about to dive into the rabbit
hole of Claude Code, which is the tool
that I've been spending most most of my
time on practically 24/7 for the last
like 2 months. It's absolutely sick. I
was intimidated by Claude Code because
it involved the terminal. I was like,
"Oh my god, terminal. I'm not like some
hacker from the 1990s. Why don't I need
to use the terminal?" And then I
realized I was a total [ __ ] dumbass
for thinking that. And actually, it's
really straightforward, especially if
you ask Claude to explain to you how to
use Claude code. And then once you're in
Claude Code, you ask Claude Code to
explain to you how to use itself, which
is freaking sick. But before we do that,
we have dived down into this thing
around trying to scrape Instagram and
YouTube competitors for social media
management. Unless you are a social
media manager or your job or your
business involves social media in some
way. Uh that specific use case is
probably not relevant to you. Obviously,
because you're an intelligent person,
you know that I'm using that use case as
an example, and you know that you could
obviously just ask Claude to interview
you about your own use cases, and it
will help you figure out what you should
build for your own stuff. But just for
the sake of opening your brain up with
more examples, I want to tell you a
little bit about some of the other tools
that I've basically built for my
business and my life over the last two
months. If you want to skip this bit in
the video because you don't care and you
just want to get to the Claude code
stuff, there'll be a time stamp as is
the case for every other thing in this
video. And in fact, to make this list,
I'm going to ask one of my Open Claw
agents. Open Claw is kind of like Claude
Code on steroids. I've been using
OpenClaw for the last like 2 months as
well. If you're just getting started
with this stuff, I wouldn't worry about
it. I would start off with Claude
Co-work or Claude Code initially and
then you can dabble around with OpenClaw
if you feel like it. But let me tell you
about my various OpenClaw agents. I have
Albus who is my primary open claw agent.
This is what Albus looks like. I have
Hermione who is my curriculum architect
for our lifestyle business academy.
She's the one who helps me basically
think through all the different things
that we are teaching our students and do
research on like what's currently
working in the world of building an
online business in 2026 etc etc. This is
what she looks like. I have Minova who
is the vice principal of the lifestyle
business academy and she helps with all
of the operational things associated
with trying to give our students the
best experience to build their you know
sixf figure lifestyle businesses. And
then Manova along with Claude Code
essentially helps me build automations
that make our lives and our students
lives a lot easier. Next we have Remis
who is my content buddy. So he Remis is
who I talk to when I'm on Telegram to
you know get content ideas and he's the
one who helps build out all of these
like competitor analysis like dashboards
and stuff that we're doing in addition
to Claude code. We have Dobby who is my
general assistant that runs on a very
cheap uh anthropic model called Haiku
whereas all the others run on Sonnet or
Opus and so they're a bit more
expensive. So Dobby's like my cheap
personal assistant. We have Cedric who's
like my relationships coach and so I
talk to Cedric and he does research
around like homeschooling and like you
know gives me ideas for like romantic
date nights with my wife and stuff like
that. So that's cool. And then recently
I added Kaledan who is my health coach.
And so Kaledan tracks my protein intake
every day and he's also my workout buddy
when I'm at the gym. Um he's got access
to all of my DEXA scans, all of my like
workout history which I exported from
the app Strong for like for the last
like 5 years and so he knows exactly
where my lifts are at. He knows that I
had like an I broke my arm recently. I
had a left radial head fracture. He
knows that he therefore has to modify
the exercises based on that. And so when
I'm at the gym, cuz currently my
personal trainer is on holiday. So while
I'm at the gym, Kaledan is my personal
trainer, I'm talking to Kaledan to
figure out what lifts I should do, any
modifications I need to make. If like my
groin hurts while doing squats, Kalin
will tell me to back off and like do
another exercise instead. This sort of
thing. If you're like, whoa, how's any
of this possible? Then don't worry about
it. Just like stop using AI to teach you
what you want to build using AI that
improves your business life or saves you
time or makes you money and then in the
process of learning stuff. And then at
some point you can ask Claude being
like, "Hey, I watched an Ali Dal video
where he talked about having these like
eight different OpenClaw agents that he
talks to on Telegram. How does that
work? What's up with that?" And then
Claude will teach you about the stuff.
You'll realize that there's a bunch of
security risks that you have to be aware
of when you are using something like
OpenClaw more so than when you're using
something like Claude Code. Even though
Claude Code does have some security
concerns, again, you can just ask Claude
about, "Hey, I've heard that Claude Code
has some security concerns. What's up
with that? Like what are the security
concerns? Why is it a big deal? Etc.,
etc. You'll learn about things like
prompt injection and how to protect
yourself against it. And you'll learn
that there are skills that you can
download for Claude, that things like
DCG, which is like the destructive
command guard that like physically stops
it from doing anything that might like
delete the files on your computer or
anything like that. If you're like, "Wo,
holy [ __ ] You're telling me it can
delete the files on my computer?" Yes,
if you do not sensible things with these
powerful tools, it can result in bad
stuff happening. But that's why you sort
of want to do it one bit at a time. You
want to make sure you understand what is
happening rather than just blindly let
the AI agent just like do whatever it
wants. So with that digression, I was
actually going to ask Albus.
Yo Albus, I'm actually working on a
YouTube video where I explain how to uh
learn AI in a weekend and you know for
the last 2 months that we've been
building stuff using OpenClaw and using
Claude code. Can you just give me a list
of real business and life use cases that
we have built in our setup that I can
show uh my lovely viewers on YouTube as
an example of what we've built. For each
one, I just want like a quick oneliner.
All right, so here's an example of all
of the stuff that I've built using
Claude Code over the last 2 months.
Firstly, a support ticket pipeline. So
within a lifestyle business academy, we
have 200 support channels for all of our
students cuz currently we have about 200
students and each student has their own
like private slack channel where they
get added to a slack channel with
themselves and like me and their coach
and like Angus and various other team
members where they can post anything
they want and we aim to get back to them
within like 24 hours ideally. Now
previously this was a manual process of
like the coaches having to keep on top
of like which Slack channel is unread
and which one is red. And then if a
student's replying in a thread, you
can't see the notifications, you can't
see the messages. And so what I created
in claude code is this thing that like
automatically creates like support
tickets anytime a student messages us in
Slack. And so that's available in a
dashboard and also pings alerts to a
sort of private Slack channel that we
have in the team which says that hey
John Smith posted this support request.
It's about to be 24 hours someone please
reply to him. You know that kind of
idea. On top of that I created a web
interface that the coaches can log into.
So they can see which Slack channels for
which students need responses and so
they're able to then just click a
button. It takes them directly to Slack
and they can respond to the student
there and then. I created a handful of
studentf facing Slack bots. So the
problem we were having is that
previously we were giving our students a
bunch of custom GPTs. Like we created
custom GPTs for things like niche
generation and first draft offer
creation and like content and like you
know giving them feedback on the sales
calls and things like that. But the
problem we found with custom GPTs is
that we couldn't actually see how the
students were interacting with the
custom GPTs. It was kind of annoying
because then I couldn't audit are the
custom GPTs actually giving the students
useful advice and then the students were
coming up with all these things and I
was like wait a minute where where did
you get that from? and they got it from
the custom GPT and then I can't see like
where things went wrong with the custom
GPT and on top of that we had students
who were like complaining that we have
like too many platforms like you know
some stuff is on Google doc some stuff
is on Slack there's a lot happening on
Slack we've got the course on circle and
now they have to have an open AAI
account and do stuff in a custom GPT and
I said to claw code like is it possible
that we can recreate the functionality
of a custom GPT within a slack bot so
that if students are on Slack anyway
they're just able to message the bot and
get the same kind of stuff but then we
as the coaching team have access to the
conversation so we can see we can sort
of like audit the agents and make sure
they're doing sensible things. And so
over the course of about a weekend, I
coded all of this up using claude code,
using Python servers, making sure that
we have everything in a database that is
sock 2 compliant. So like enterprise
level security, so that we don't like
accidentally leak any conversations the
students are having. And so we created a
bunch of Slack bots that students have
been engaging with and they're having a
great time talking to. For example,
there is Dumbledore, the DM agent. And
so we've had like over a hundred of our
students send thousands and thousands of
messages to Dumbledore the Slackbot. And
Dumbledore has been trained on our like
internal DM methodology for how to sort
of convert a cold prospect into a warm
lead and get them on a sales call so
that they can grow their business. And
so students are like copying and pasting
LinkedIn profiles of their prospects
into Dumbledore and Dumbledore is
helping them craft outreach messages. If
a student is in a DM conversation with
like a prospect and they're like
struggling to turn it into a call,
they'll just like screenshot the
conversation, stick it into Dumbledore,
and Dumbledore will give them advice
based on our connect, understand, invite
framework, for example. So we have
Dumbledore, the DM agent, we have Lupin,
the LinkedIn agent, we have Sprout, the
sales agent. By the time this video
comes out, we we will have released
Flitwick, who's like the delight agent
that helps students like craft a
delightful customer experience for their
coaching or consulting or service
business. And I'm working on a bunch
more, which is super exciting. But the
really cool thing about this is that the
students can give us feedback. I can see
the stats on how much people are
interacting with these agents. I can
literally read the conversations that
students are having with the agents. We
do flag it to them that like I can read
the conversation. So I read the
conversations to see like are the agents
giving students sensible advice. And so
if you use Slack in your business or in
your work, then did you know you can ask
Lord Code to help you figure out what
Slack bots you can create that make your
life easier. We have another Slack bot
that's like, you know, alerts us at the
start of every day to see like, you
know, in terms of like our competitors
on YouTube, like who has released videos
and what's been their like views in the
last 24 hours. So we'll see that, you
know, Homozi recently released a video
called whatever, whatever, whatever. And
that's got 94,000 views in the last 24
hours. And that's like an outlier for
Alex Hozi's channel. So then it prompts
us to think, huh, could we like take
inspiration from the fact that that
video is popping off for Horosi? The
thing that I'm working on now is a sort
of creator HQ, which is a sort of web
dashboard that me and my entire team can
use across YouTube, Instagram, LinkedIn,
and Tik Tok, and Twitter, that
automatically analyzes all of the
content that we've put out over the last
like 9 years on these platforms. Figures
out what are the things that worked,
what are the things that didn't work,
generates content ideas based on
transcripts from stuff that I've said
when I've done workshops for our
students in LBA. Automatically tracks
all of the competitors, and then it
helps us surface ideas for better
content that we can make that will just
perform better. And on top of that, I've
also created my own MCP server, which
then Claude and Chaget have access to,
which means that they can automatically
see all of the projects that I'm working
on within the business, what my goals
are, all of the outstanding to-dos, any
kind of journal items that I've put in
there because I want it to have access
to that. And so now, anytime I talk to
Claude AI, the chat, it now has way more
context about me than I ever did before.
And that context is always up to date
because I'm keeping it up to date when
I'm using Claude Code, when I'm using
OpenClaw, when I'm using Claude. And it
all happens through an MCP server. Now,
that might have been a lot of jargon. I
did not know what any of this stuff
meant two months ago until I decided to,
you know, spend a weekend really just
trying to build something useful using
claude code. And over the last two
months, like every day, as I've done
this learning loop, this sort of AI
flywheel, I've built more and more
things. And there is a danger with all
this like AI stuff that you you end up
in this in this zone where you spend so
long like building agents that talk to
other agents and that like help you
orchestrate other agents that you don't
actually end up doing anything useful.
useful I would define as it adds value
to your customers. It saves you time
personally. It makes you money
personally. And so I was very mindful of
that that like I didn't I didn't want
this to be another one of those like
people spending days making notion
setups to theoretically try and be more
productive but not actually being more
productive. So going back to our data
analytics project for YouTube
competitors. Yes. Okay. I'm now ready to
start writing this thing. Can you help
me understand what the hell clawed code
actually is? Um, I don't really have a
background in coding and so I have been
intimidated by claude code in the past.
You know, when people when I see people
using the terminal with like the black
and the green text, it feels like
there's some hackers out of the matrix
and I feel like I'm not techsavvy enough
to use claude code. Can you help me
figure this out? Okay, let's kill the
intimidation. Now, the terminal is not
what you think it is. That black screen
with the green text, let's demystify it.
Because before we talk about cla code,
you need to understand the terminal
because claude code runs inside it.
Here's the thing. You already know how
to use a computer using a graphical user
interface. A GUI. You see icons, you
click buttons, you drag things around.
Mac OS, Windows, your phone, all
graphical user interfaces. But GIS
weren't always there. Before 1984, when
Apple released the Macintosh, virtually
all computers were operated by typing
commands. You type something like ls to
see a list of files or cd documents to
go into your documents folder. That's
it. That's the terminal. It is a way of
talking to your computer by typing
instead of clicking. The terminal didn't
go away when GUIs arrived. It just got
hidden. On your Mac right now, there's
an app called Terminal. This is what it
looks like. Boom. There we go. I think
on the Mac by default, it's white with
black text rather than black with white
text, but whatever. It's basically the
same thing. You can type whatever you
want. Dot dot dot. If you open it,
you're looking at the exact same
computer, the exact same files, the
exact same everything. You're just
steering with a keyboard instead of a
mouse. Think of it like driving. A GUI
is an automatic car. The computer
handles a lot of the details for you.
The terminal is a manual. Same car, same
road, same destination. You just have
more direct control. And for certain
tasks like the one we're about to build,
the terminal is actually way faster and
simpler than clicking through menus. So
what is Claude code specifically? You
know how when you use Claude in the
chat, you type a message and Claude
responds. Yes, Claude code is the same
Claude, the same brain, but instead of
living in a browser window, it lives in
your terminal. And here's the key
difference. It can actually do things on
your computer. When you chat with Claude
on Claude.ai, it can give you code, but
then you'll have to figure out where to
put it, how to run it, what to install.
You are the middleman. Claude tells you
what to do, and you do it. However,
Claude code removes the middleman. You
say, "Build me a script that pulls
YouTube data for these 50 channels." And
Claude Code will actually create the
files, write the code, run it, see the
errors, fix the errors, and run it
again. All right there on your machine.
You're sitting next to it watching, and
you can steer it. No, not like that. Do
it this way. But you're not doing the
mechanical work of copying and pasting
code between windows. Oh, this is nice.
A concrete analogy. A regular Claude is
like texting an architect. They send you
blueprints and you have to figure out
how to build the house yourself. Claude
code is like having that architect
standing in the room with a toolkit.
They're actually building the thing.
You're watching and you're saying,
"Actually, move that wall over there."
And they just do it. Why? It feels like
hacking, but isn't. The matrix aesthetic
of the terminal is genuinely misleading.
When you use cloud code, most of what
you're doing is just talking to it in
plain English. You type things like,
"Can you create a new project folder for
my YouTube tracker, and it just goes and
does it. It'll ask you questions if it's
unsure. It'll show you what it's about
to do before it does it. You can say yes
or no. It's a conversation just like
this one, except the AI can actually
touch your files. Amazing. Like, you
didn't need to watch a tutorial to teach
you how to do this. All you needed was
ask Claude to explain what Claude code
is. And now you can ask it be like,
"Okay, how do I install it?" Okay, this
sounds really cool. How do I install it?
You can download it using the desktop
app if the terminal feels scary. Or you
can type in one command in the terminal.
Okay, so on a Mac, open the terminal
app, hit command space, type terminal,
hit enter, then paste the single command
curl-fs
claw.install sh. So this is now where it
starts to feel like coding. It feels
like hacking. It feels like, oh my god,
this is this is intimidating. But super
simple thing. I'm going to copy this.
Explain to me what the hell is going on
here before I do it, please. dot dot
dot. You don't have to do this with
every single terminal command, but like
it's good habit to understand the code
that the machine is trying to run on
your machine. Good instinct. You should
never paste something into a terminal
without understanding what it does. Let
me break down every single piece.
Reading left or right, curl. This is a
program that's been on basically every
computer since 1998. The name stands for
client URL. All it does is fetch stuff
from the internet. You give it a web
address. It downloads whatever's there.
It's like a browser, but instead of
displaying a pretty web page, it just
grabs the raw content. FSSL. These are
four flags squished together. Flags are
like settings that modify the command
behaves. Fail cleanly, silent, still
show errors, follow redirects. Oh, okay.
So, FSSL together means download this
quietly, follow any redirects, and only
talk to me if something goes wrong.
Claw.ai/install.sh
is a web address. Points to a file on
anthropic servers. If you're curious,
you can actually open that URL in your
browser and read the script before
running it. Huh. Okay, let's do that.
Oh, nice. It's just like some text
stuff. And if we wanted to, we could try
and like copy and paste this text into
Claude and get it to help us explain
every single line. To me, that's a
little bit overkill. In general, there
does come a point where you do in fact
have to trust the thing that you're
interacting with. Now, in my case, I
trust Anthropic. I trust Claude. I trust
that they're not going to do dumb things
on my computer because they're like a
$400 billion company and it would just
be dumb for them to like put malicious
software on my own computer. You might
think differently. If you have major
distrust of like big corporations, then
you are very welcome to go through this
line by line and make sure that every
single command that's running on your
computer, you understand what's going
on. You audit it, validate it, make sure
it's legit. But like, you know, we all
trust big corporations. You probably
have a Google account. You probably
trust Google with your emails, with your
maps, with whatever. You probably have
an Apple account. You probably trust
Apple with your private photos of like
your friends and your family and
yourself. Yes, there are security
concerns with a lot of this stuff. And
if you're downloading a random plug-in
from some random developer that you've
never heard of, then you want to have a
higher threshold for like being
skeptical or being like security
conscious about that. But if you're
downloading like the official Claude
code app from Anthropics website
directly where you can verify that it is
in fact on Anthropic's website,
calibrate it to your level of risk. The
nice thing is you can be as security
conscious or as cavalier as you like. In
my case, I started out very very very
security conscious about every little
thing. Make sure I understand every
single command. And then I sort of
realized, okay, this is like overly
cautious, especially when I'm
interacting with local files where it
doesn't involve talking to like, you
know, customer data or anything like
that. I sort of lower my bar for like
how much I think about the security side
of things. But as soon as it involves
interaction with customer data, that's
when I'm like, all right, let's like
lock down the security walls around this
and make sure that like what I'm doing
is legit. But you can decide how
security conscious to be uh or not. Ah,
you were right to ask. This pattern of
download a script and immediately run it
is super common for installing developer
tools, but it does require trust. You're
saying, "I trust that whatever Anthropic
has put at that URL is safe to run on my
computer." In this case, it's Anthropic,
a major AI company, so that trust is
reasonable. But as a general rule, never
run a command like this from a source
you don't trust. Exactly. Great. So then
once you've installed Claude, you
literally just type in Claude into your
terminal. And look what happens.
Accessing workspace uses Alidal. Is this
a project you created or one you trust,
like your own code or work from your
team? Um, so yes, I trust this folder.
Great. Now you'll see this doesn't look
scary anymore. This doesn't feel like
you're trying to hack hack away in a
black and green terminal. Welcome back,
Ollie. There's a little icon. Opus 4.5
Claude Team Sparkle Studios. Hello. What
the hell is going on here?
Coalescing. Oh god. Please run login.
Okay, I need to log into my Claude
account. Let me do that. Great. I would
like to connect to Parker Studios, which
is the name of my business. Yes, I'm
happy to for it to do stuff. Great.
Login successful. Press enter to
continue. Wonderful. Hello. What the
hell is going on here? Boom. Hey, you're
in Claude Code. Anthropic command line
interface for Claude. Your login was
successful. I can help you with software
engineering tasks like writing,
debugging, or explaining code,
navigating and understanding code bases,
running commands, and managing files,
answering technical questions. What can
I help you with? We basically have
Claude running inside the terminal that
is able to then access files and do
stuff on our computer with our
permission. Crucially, you might be
thinking, "Oh my god, there's all sorts
of security concerns here. Ah, I'm so
scared about it deleting all my files."
If that's the case, actually, you know
what? Let's talk to Claude Code. Hey, so
this is the first time that I've
actually installed Clo and I am a little
scared. I'm worried about the security
implications of having an AI that can
touch my files and potentially delete
everything on my computer. Can you help
me understand what is going on? Totally
fair concern. Giving an AI access to
your file system is a big deal. Here's
what you should know. You're in control.
I ask for permission before running
commands or editing files. You see
exactly what I'm about to do and can
approve or deny it. You can configure
how strict the permission system is.
Commands run in a sandbox by default.
This limits what shell commands can do.
If you don't know what that means, just
ask Claude, "What is a sandbox? What are
shell commands? I can't access files
outside your working directory without
you navigating there." Destructive Git
commands like git reset hard require
explicit approval. If you're like, "What
the [ __ ] is a git command?" You can just
ask Claude, "What is a git command? How
does it work?" Then you're going to go
in a rabbit hole to understand that git
is this like thing that was created by
Linus Tovalt, the inventor of Linux back
in the day, because he wanted a better
versioning control system compared to
having like document 1, document 2,
document 3, v5, final, v6 final, etc.,
etc. So, Git is like this open source
thing that's on that you can install on
basically any computer that basically
creates a snapshot of files so that if
anything goes wrong, you can just revert
to the previous version of the file. You
might not have known that, but you do
now because you're not a dumbass and you
know that if you cla involving the
security on your computer that you do
not understand, you just do a little bit
of research by asking it to explain it
to you and it will do a really good job
of explaining it to you in a very
engaging way. What I can't do without
you seeing delete files, run arbitrary
shell commands. What's a shell command?
I don't know. Let me ask it. Push code
to remote repos. What the [ __ ] is a
repo? I don't know. Let me ask Claude.
Install packages. What's a package? I
mean, let's ask Claude. Practical tips.
Start in a test project or git repo so
you can revert changes. What's a git
repo? I don't know. Let's ask Claude.
Watch the permission prompts. They'll
show you exactly what will happen. Type
help to see the available commands and
settings. The honest truth. Very AI. I'm
a powerful tool and with great power
comes the need for you to pay attention
to what you're approving. But the system
is designed so you always have
visibility and veto power. Want me to
walk through a simple example so you can
see how the permission flow works? Oo,
yes, please.
It's so helpful. I love it. Watch what
happens when I attempt this. Create file
cla test.txt. Hello. This is a test file
created by claude code. You approved
this action which is why it exists. Feel
free to delete me. Do you want to create
claude code test.txt? I can say yes.
Yes. Allow all edits during the session.
I.e. I'm just giving it blanket
permission to create files and edit
files. Might want to not do that
initially if you're new to this sort of
stuff. Let's say yes. Terminal would
like to access files in documents
folder. Sure that's fine. So even the
Mac is giving me an additional like
security prompt before doing this. So
here we go. Write claude code test.txt.
Wrote four lines to claude code
test.txt. You should have seen a
permissions prompt before that went
through showing you the exact file I
wanted to create and the full contents
of what I wanted to write. You approved
it. So now the file exists. Let me read
it back to you to show that it works.
Read one file. Notice that reading
didn't require a permission prompt. It's
a non-destructive action. So it just
happens. Now let's clean up. Watch the
permission prompt again. Okay. So bash
rm uses our example code test with txt
delete the test file we just created.
Now you might be thinking what the hell
does bash rm mean. And so this is where
especially when it starts to say delete
stuff. This is where I would recommend
copying the command pasting it into
another clawed window or even a chatgpt
window if you're like wait a minute I'm
using claude to tell me about the
security settings in claude. What if the
AI goes [ __ ] rogue and like misleads
me? If you want, you can use a different
AI to do the security audit on the AI
that you're using. But again, I broadly
trust anthropic. Let's go into claude.
Be like, okay, I'm using claude code for
the first time, and it's trying to run
this command. Uh, what does this
actually mean? What is bash? What is RM?
Two things to unpack here. What bash is
and what RM does? And the second one is
worth paying attention to. Bash is the
language your terminal speaks. The name
stands for born again shell, which is a
nerdy joke, 1979, blah blah blah blah
blah. RM, this is the one you should
care about. RM stands for remove. It
deletes files. This is the one terminal
command where you should always read
what comes after it before hitting yes.
Unlike dragging something to the trash
on your Mac, RM does not have a
recycling bin. When it's gone, it's
gone. So RM some temporary file.txt,
fine, it's cleaning up a single
temporary file. For example, RM users
areal claude code test.txt. I know that
I now know because I took the time to
understand what is bash and what is RM
that oh, all it's asking me to do is
it's asking for permission to just
delete that one file. RMR some folder.
It's deleting a whole folder and
everything inside it. The -ash r means
recursive. Go into every subfolder and
delete everything. Still fine if it's a
folder that claude code just created.
And then we have the famous command, the
one that everyone worries about. rmrf/
This would attempt to delete your entire
computer. Nobody should ever run this.
Claude code won't suggest it, but I'm
mentioning it so you understand the
spectrum. The key habit to build.
Whenever cla code asks permission to run
a bash command, read it. Most of the
time it'll be harmless things like
creating folders, running code, or
installing packages. But RM is the one
where you should always glance at what
it's removing before you approve it.
What does the full command say after RM?
I can tell you whether it's safe to
approve. I know it's safe to approve,
but just for the sake of the argument,
I'm going to copy and paste. This is the
command. And so then it's going to tell
me totally safe. It's deleting a single
test file called claude code test in
your home folder. That's almost
certainly a file that Claude Code itself
created a moment ago to test that it
could write files on your machine and
now it's cleaning up after itself. Think
of it like someone checking does the pen
work by scribbling on a scrap of paper
and then throwing the scrap away. Go
ahead and approve it. Okay. Oh my god,
it's scary. Done. Here's what you just
experienced. Action. Write a new file.
Yes, you saw the content first. Read a
file? No. Permission needed? No. It's
non-destructive. Delete a file via bash.
Permission needed. Yes, you saw the
command first. The key takeaway, you are
always in the loop for anything that
modifies your system. If you ever see
something remotely sketchy in a
permission prompt, just deny it. Feel
more comfortable now? Any questions
about how this works? Mate, our AI
overlords are very good at helping us
feel good about the commands and the
permissions that we are allowing them to
have to run stuff on our machines. This
is pretty cool. So now what you
basically have is yes, you can chat to
Claude via claude.ai or Claude chat, but
once you have it on your computer on
your terminal, you can also literally
talk to Claude code via your terminal.
It's just that the terminal version
gives you the option with your
permission to access your files and add
files and remove files. That one's you
got to be careful about and be able to
write code, be able to run code. Whereas
the clawed chat, it can still create
websites and PowerPoint presentations,
but it does it in a web browser. So you
have to download the file or it wouldn't
be able to actually run a web
application for you because you'd have
to run it on your computer. So it's sort
of like claw chat is like, hey, I've got
all these I've got the ability to do all
these things, but like I can't actually
touch your files. Everything just
happens in a web browser within
claude.ai. But claude code is like, hey,
I've got all those powers as well, but I
can also make files directly on your
computer. Okay, so now I've installed
claude code. Where do we go from here?
Like, how do I build this like uh
YouTube competitor analysis dashboard
situation that we were talking about?
Okay, here's where it gets fun. Three
steps before you start talking to cl
code. Create a project folder in your
terminal. You want to create a dedicated
folder for this project. Think of it
like creating a new folder in your
desktop or via the terminal. So I could
do mech, which means make a directory.
It creates a folder. CD means change
directory, then start cla code inside
that folder. That sounds interesting.
Let's just do that. So, I'm going to go
in my terminal. I'm going to get rid of
claude code. I happen to know that you
use C to like exit out of cla code. If I
didn't know that, I would ask cla code.
So, now I've exited claude code and I'm
now back within the terminal interacting
with my computer directly. So, what I
could do is I could go on my desktop, I
can rightclick, I can do new folder and
I could call it
YouTube tracker. Then I can open the
folder YouTube tracker. And you know how
this works. You know how to create a
folder on a desktop because you know how
to use a computer obviously. So that
would be the manual GUI graphical user
interface way of creating that
particular folder. Let's delete that
now. But I can do the same thing from
the terminal itself. I can use mcdur. So
let's go in the terminal and type in
mcdder YouTube tracker. I hit enter. And
it looks like nothing has happened. But
if I look inside my home folder,
turns out there is now a new folder
called YouTube tracker. I was able to
run a command from the terminal. The
terminal understood what I'm trying to
do and it created the folder for me.
Sick. So now it's like cd YouTube
tracker. Okay, let's copy and paste
that. Let me make sure I understand what
it means. CD means change directory.
Okay. Yeah, that seems fairly harmless.
Boom. And you'll notice that it's gone
from Ali Ali's Micro Pro 3 uh with this
little wavy symbol. What's the What's
this wavy symbol called? Something else
called It is called Tilda. Okay, fine.
So, previously we were running commands
in my home folder. aliadal with this
which is shorthanded via this tilda
command as I've just found out from
claude and now we are inside the folder
YouTube tracker. So then it's telling me
now start cla code inside that folder by
just typing in claude. Okay, I just
typed in claude. Oh, nice. Accessing
workspace YouTube tracker. Uh yes, I
trust this folder. It's got nothing in
it cuz I just created it. And now I am
running claude code within the context
of this YouTube tracker folder. Next
step, get a YouTube API key. Oh god,
this is going to be complicated.
Remember how to set API, supply, the API
keys, access bridge. You just need to
know who's making the request. People
use it. The key is free. Google just
wants you to register. So go to
consolecloud or google.com, create a new
project, search for the data API, go to
credentials, and create an API key. It
takes about 3 minutes. And then step
three, tell Claude Code what to build.
This is the magic part. Once you're
inside Claude Code in your project
folder, you literally just describe what
you want in plain English. This is like
super cool. So I could literally say, I
want to build a YouTube competitor
tracker. I have a list of about 50
YouTube channel names. I want a script
that uses the YouTube data API to pull
the 10 latest videos from each channel,
including title, thumbnail, URL, view
count, and publish date. Store the
results in a local JSON file for now.
Here's my API key. Let's start with just
three channels to test. And then cloud
code will start building. It'll create
files. It'll write the code and probably
run it to test. You'll see it asking for
permissions along the way. Can I create
this file? Can I run this command? And
you just approve each step. Caveat, you
want to understand what it's doing
before you approve the step. And then
over time, you'll realize that you'll
you'll start to get a feel of what you
need to worry about versus what you
don't. My suggestion, start tiny. Don't
try and build the whole dashboard in one
go. Start with just three channels. Get
the data pulling working. Look at the
output. Then say, "Okay, now can you
make this into a simple web page where I
can see the thumbnails and titles in a
grid and layer it up from there." The
beauty of Claude Code is that you can
iterate in conversation. Actually, can
you sort by views? Can you add a column
for how old the video is? Can you
highlight anything over a million views?
You are steering. Claude code is
building. That's the workflow. Again, a
digression. It's not really it. It is
relevant. One of the things that really
holds people back from running this AI
flywheel is encountering friction and
being unwilling to figure it out.
Developers don't have this problem. When
you're a software developer, you
encounter friction all the time. You
encounter a bug and the bug shows that
something's not working. You run try and
connect to an API and you realize that
like the API documentation is out of
date. Back in the day, the skill of
being a good developer was knowing what
to Google cuz it's like knowing what
error you Google and what what you copy
and paste into Google to then go on a
website called Stack Overflow, which is
like a forum where developers are
discussing like how to fix things and
then figuring out what the fix is that
someone else has done and then trying to
run it on your machine and seeing if the
fix is still valid. And if you're a very
good developer and you're good to the
developer community, you would then
update the Stack Overflow forum post
with what the latest version of the fix
is. These days, you don't actually need
to do the Google Stack Overflow thing
because you can just copy and paste an
error message into Claude and it will
tell you exactly what to do and then you
know all this sort of stuff. So
developers are naturally good at pushing
through friction. Most people who are
not developers in the context of their
work have probably not encountered like
technical friction unless you're in a
particularly technical role. You might
have encountered human friction like
what do you do when your boss refuses
you like annual leave request and then
you have to figure it out. But when it
comes to doing stuff like this, you will
encounter some technical friction. And
the thing that will separate you from
your grandma is that you are going to be
willing to push through the technical
friction and figure it out. So getting
API keys from platforms is an example of
technical friction that in particular
with like Google projects is a little
annoying and it's a little it can be a
little hard to get your head around. But
I trust you, you're intelligent, you
will be able to figure this out. And if
you can't figure it out again, you have
Claude, you have Chantyt that you can
just like copy and paste stuff into to
figure it out for you. So let's get a
YouTube API key. So go to console.cloud.
cloud.google.com. Huh. Okay. Yep. Open
link. Create a new project. Call it
YouTube tracker or whatever. How the
freak do I create a new project? Ah. Ah.
Ah. Friction. Friction. Google Cloud.
Okay. Maybe if I click here. New
project. Ah, [ __ ] How do I create a new
project? I don't know. Ah, complicated.
Oh my god. Complicated. Okay,
we're going to calm down. We're going to
relax and I'm going to ask Claude. Yo,
Claude. Um, I'm on the Google Cloud
Console thing, but I can't see a button
to create a new project. Uh, what the
freak do I do? Huh? Google Cloud Console
is not the most intuitive interface in
the world. Let me look up the current
layout so I don't send you on a wild
goose chase. Thank you, Claude. Very
kind. Yeah, Google Cloud Console is
notoriously confusing. Here's exactly
where to look. At medium, the top of the
page. What the [ __ ] does that mean?
Click the button directly.
Okay. Ah, okay. So, if I click on that,
I can then type in new project. YouTube
tracker test. Okay. Yep, that's fine.
Notifications. Create project YouTube
tracker test. Okay, cool. Great. Let me
click select project. Ah, okay. I just
saw YouTube tracker test. You're now
incurring charges in your billing
account. Uh-oh. It's probably fine. Set
up Google Google Cloud for your
organization. Ah, I don't know what any
of this stuff means. What are we going
to do? We're going to ask Claude. So,
this is what I like to do whenever I'm
on a page where I have no freaking clue
what's going on. I command a command C.
Copy. Stick it into Claude.
Help me understand what the [ __ ] is
going on on this page. Dot dot dot. I
literally just copied and pasted the
entire contents of the website, chucked
it into Claude and now Claude's going to
explain it to me. Hopefully Claude is
going to say, "You don't need to worry
about it, bro. Like, let's just build a
YouTube tracker." But I actually did
this many a time. I was setting up
Cloudflare. I was setting up like a
database thing, making sure it's secure.
And you know, sometimes the instruction
Claude gives are sort of out of date
because like Cloudflare, which is like
the security provider, might have
changed the way their interface looks.
And so I would just end up copying and
pasting the contents of the page or
screenshotting it, chucking it into
Claude, and then asking Claude, "What do
I click on?" Again, we need to be good
at pushing through technical friction if
we find it and then the friction will
disappear over time and it will just
become a part of your usual skill set
that like if something doesn't work
first time, you figure out why it didn't
work and you do something about it. Deep
breath. You can ignore almost everything
on this page. Thank you, Claude. What's
happening? Google Cloud Console is
designed for enterprise companies
running massive infrastructure. Think
Netflix deploying thousands of servers
or banks processing millions of
transactions. So when you create a
project, Google assumes you might want
to set up Terraform infrastructure
automation because that billing alerts,
security policies, identity management,
and all this other stuff that huge
companies need. You need literally none
of that. You're here to get one API key
so you can pull some YouTube video data.
It's like walking into a car dealership
to buy a bicycle, and they're trying to
sell you fleet insurance. That's quite
funny. The good news, now you can see it
says now viewing project YouTube tracker
test at the bottom, which means your
project is already created. You're done
with that step. Okay, now do this. Look
at the left sidebar. Hamburger menu in
the top left. Click on APIs and
services. Ah, sick. I clicked on APIs
and services. Click library. Okay. Ah,
love it. I love it when the instructions
that the AI gives me actually matches
what I see on screen. Doesn't always
happen. This is the friction that
sometimes you have to go through.
Library. Okay. In the search bar, type
YouTube data API v3. YouTube data API
V3. Oh, sick.
Nice. The YouTube data VPI is an API
that provides access to YouTube data
such as videos, playlists, and channels.
Sick. Click the result, then hit enable.
Nah, it can't be that easy, can it?
Enable. Surely not. That would be too
easy for like dealing with like Google
Cloud stuff. Okay, what's happening now?
Okay,
stuff is happening. Ah, okay. Chill out.
Let's go back to Claude. Once it's
enabled, go back to APIs and services
credentials. Uh, okay. Well, APIs and
services credentials. Okay. Click create
credentials API key. Where the [ __ ] is
create? Okay. Create credentials API
key. Okay. Copy the key it gives you.
Ignore Terraform. Ignore billing setup.
Ignore migration center security command
center. I am blah blah blah blah blah
blah. That's the enterprise stuff.
You're just grabbing a key and leaving.
Think of it like checking into a hotel.
They're offering you the spa package,
the restaurant reservation, the concage
tour, and the loyalty program. You just
want the room key. Okay. Back in the
day, you'd see a page like this. You'd
be like, name. Does the name matterh?
And you spend ages googling on Stack
Overflow being like, does it matter what
I name my YouTube API key? APIs that can
be accessed using the key. You'd be
like, oh [ __ ] now I've got this
enormous, oh god, what's all this list?
Analytics, BigQuery. What the freak is
BigQuery? Oh my god. CL [ __ ] Like
you're like do I need to use a service
account or not? Like what the hell is a
service account etc etc like ah
application restrictions what should I
do website like ah this was my life 10
years ago when I was trying when I was
building my first like coding projects
back in like 2015 I built like this sort
of uh online question bank for medical
students. This was the first business
that I built that you know did
reasonably well I say reasonably well.
It got to like six figures in revenue
and then I sold it a little bit later.
But now you just copy and paste all of
it into claude and it helps you
understand it directly. You know what?
I'm just going to copy and paste all of
this. It's giving me all these options.
What do I type in where dot dot dot
classic Google over complicating things
again. Here's what to do. Name. Type
something like YouTube tracker. This is
just a label for your reference so you
remember the keys for. Okay. YouTube
tracker. Thank you. APIs that can be
accessed using the key. Where it says
API restrictions, click restrict key and
then find YouTube data v3. Okay, cool.
Okay, that's fine. Everything else,
ignore it. Thank you. Don't touch
application restrictions. Don't trust
the service account thing. Don't worry
about our websites, IP address, Android
apps, iOS, n. None of this applies. Just
give it a name. Restricted to data, API
v3. Hit create, etc. blah blah blah.
It'll show you a long strings of
characters. Copy that immediately and
paste it somewhere safe. Okay? Don't
share this key publicly. Don't paste it
in a YouTube video. Don't commit it to a
public GitHub repo. It's like a
password. It's tied to your Google
account. If someone else gets it, they
could use up your free API quotota. For
the video you're filming, you want to
blur it or cut away when it's on screen.
Note to editors, please, can you do
that? Okay. So, I hit create. Great.
Now, it's given me an API key. I'm going
to copy it. Once you got the key copied,
come back and we'll fire up Claw Code
and start building. I've got it copied.
What do I do next? If you're still with
me in the video, like this is actually
what the process looks like. It's just
how it works. And it might seem
intimidating right now, but like
honestly, just give it a day. Give it a
weekend and play around with it. Like
push through the friction. I promise
your life will freaking change. So now
we open our terminal, run these three
commands. Make directory, YouTube
tracker, CD, YouTube tracker, and
claude. Okay, so I've already done that
actually. Great. So I'm in YouTube
tracker. I'm running Claude. Okay, once
cloud code is running, you'll see the
prompt paste something like this. I want
to build a YouTube meta tracker, etc.,
etc., let's just copy and paste it.
You'll see pasted text lines 1 to 9. My
API key is, and I'm going to use the
clipboard thing here. Then just hit
enter and watch it work. Two tips for
your first time. Don't try to be perfect
with your prompt. You can always say,
actually, can you change X afterwards?
That's the whole point. It's a
conversation, not a oneshot. And read
what it's doing, even if you don't fully
understand the code. I agree. You'll
start picking up patterns naturally. And
if you see something you're curious
about, just ask it. Hey, what does this
line do? right there in cloud code.
Okay, so we can see like this is what
cloud code is now doing. And I'm like
fingers crossed what's going to happen.
Huh? Let's see. Let's see. Listing one
directory. Okay. Yep. Empty directory.
Let me build a cracker a Python script
to fetch data in an HTML patient
displayer. If I didn't know what Python
was or what HTML was again, I would just
be like copy into paste. What does this
mean?
Oh [ __ ] Now there's all this code. Oh
my god. Intimidating. It's fine. It's
fine. It's fine. We don't need to worry
about it. Okay, so it's creating a file.
Fetch videos py fetch latest videos from
YouTube channels using the data API v3.
Okay, it's doing all of this stuff. It's
writing all this code. API key base URL
channels alto Frankella. Cool. Um, okay.
At this point, this will feel like
friction and so people will stop doing
it because they'll be intimidated by the
fact that it's writing code. The trick
is to not get intimidated by the fact
that it's writing code. It's only code.
If you want, you can look through it
line by line and try and understand what
the code does. Would I recommend that?
Well, I mean, you can do it. It
generally is good practice to understand
what broadly the code is doing that
you're writing on your computer. To be
honest, often the way that Claude Code
and Codeex and these other AI tools like
write code, it actually is somewhat
human readable. So, let's see what's
going on here. You'll see that it's it's
even explaining fetch latest videos from
YouTube channels using the data API v3.
Import all of this stuff. It's huh.
Okay. Okay, whatever that means. It's
setting an API key. Base URL channels.
Yeah, videos per channel 10. Yeah, I
understand what that means. Def API get
point. Ah, so and then I'm like, huh,
this is a comment. Make a get request to
the YouTube data API. Okay, so it's
doing that. Then find a channel name.
Search for a channel name by ID and
return its ID and sual title. Okay,
yeah, that seems reasonable. Like often
as you're browsing code, you don't need
to understand what every single line of
the code does, but the AI will usually
do a good job, and this is what good
developers do as well, of commenting the
code as they go along. So, it'll help
someone reading the code further down
the line, understand the gist of what
the code is trying to do. Okay, cool.
Search for a channel. Get the uploads
playlist for each of the channels. Yep,
that makes sense. Get the latest video
ideas from a playlist. Okay, yeah, that
makes sense. Get the full details for a
list of video ideas. Okay, okay, okay.
Yep. All of this stuff. sort the videos
by view count descending. Okay, so do I
want to create this file? Well, I'm just
creating a file. I'm not actually
running it yet. So, I'm just hit yes. If
I wanted to, I could copy and paste it
into Claude and be like, "Hey, can you
explain every single line of the code?"
The nice thing about Claude Code is that
it'll give you little tips for using
Claude code as you go along. Like use
/btwide
question without interrupting Claude's
current work. I use that a lot. Uh or I
mean you can just command tab into
Claude chat and just ask the questions
there. That's usually what I do to be
honest. I get Claude code to build the
stuff and I use claw chat to help me
understand things as it's building
stuff. Okay, now we're running a bash
command. This command requires approval.
So, Python 3 fresh videos fetch
videos.pi. This command requires
approval. This is where I'm like, I'm
going to paste this command. Okay, it
seems like it's trying to run this as a
batch command. Is this something I
should be concerned about? I'm making a
big deal of the security things like uh
because I see so many comments from
people on YouTube being like, "Oh my
god, security." And it's like, not at
all. This is exactly what's supposed to
happen. And it's a great example of how
Cloud works. Fold code has just finished
writing a Python script. Blah blah blah
blah blah blah. It's asking for your
approval. It's just running the code it
wrote, etc., etc. Great. Let's allow it
to run the command. Okay, so it's
running the command. Fetch videos py.
Let's see what happens. Oh, fetching
Ali. Fetching Thomas Frank
plus five lines. Control O to expand.
Huh. What happens if I do control O? It
doesn't matter. All 30 videos fetched.
Okay, now it's going to ask me for more
permission. Bash command preview top
five videos by view count. By view
count. Uh, at this point I'm like, okay,
I'm just going to proceed because that
seems reasonable. Let me verify the data
looks good. Working perfectly. Here's
what was built. Fetch videos. py hits
the YouTube data API. Pulls the latest
10 videos from each channels. Searches
for the channel name. Get uploads. Blah,
blah, blah. Index. HTML. Dark themed
grid display that read videos.json.
Shows thumbnail title view count bad
channel name time ago, etc., etc. Oh,
interesting. To view the HTML page, open
it via a local server needed for the
fetch call to load JSON. Python 3 M HTTP
server 8000. Then open local host in
your browser. What would you like to add
next? Now I'm like, okay, this is
confusing. What the hell is going on
here? So, I'm going to You know what? I
don't even need to copy paste it. Wait,
help me understand what this command
does. What's a local server? The first
time you try creating an HTTP server and
running things on local host, you might
be like confused as to what the hell's
going on. But if you just ask it to
explain it to you, you'll understand,
oh, this is what a local server is, and
then you won't need to worry about it
again. Good question. Let me break it
down simply. The problem when you double
click index.html html to open in a
browser, the page tries to load
videos.json using JavaScript fetch.
Browsers block this for security reasons
when opening files directly from your
computer. You know what? Let's test
this. If I go and find her, ah, when I
look in my YouTube tracker folder, I see
three files. I see fetch videos. py, I
see index.html, and I see videos.json.
Turns out I can actually open all these
files and I can verify what's going on.
So, if I open this with, I don't know,
like text edit or something. Oh, sick.
This was the code that Claude wrote for
me. Okay, nice. If I look at index.html,
I can open it in a code reader and I can
read like, okay, there's all this HTML
code, whatever. And then I can look at
videos.json. I can open it with text
edit. Hopefully, that'll load. I can
see, oh, it's basically just data. Video
ID, title, thumbnail. Sick. Okay.
Channel name from Mattyella. Nice.
Channel name Thomas Frank. Channel name
Ali Dal.
Nice. Okay. It's kind of interesting.
Um, so now what happens if I just try
and open index.html? Could not load
videos. Run python 3 fetch videos. py
first. Okay, so it's not working. Uh,
but Claude already knows that it's not
going to work because it knows that this
is the problem. The solution, Python 3M
HTTP server, starts a tiny web server on
your own computer. It makes your files
available at localhost 8,000, which is a
URL the browser trusts. So that fetch
works. Local host is your own machine.
Nothing goes to the internet, so no one
else can access this. 8,000 is the port
number like a door number. Arbitrary,
just needs to be unused. And it only
runs while the terminal command is
active. Hit control C to stop it. Huh.
Nobody else can see it. It's purely
local. Want me to start it up for you?
Sure. So now Claude is asking me to run
this bash command. Python m http server
8000. Okay, sure. Server is running.
Open this in your browser. Local host
8000. Okay, let's see if it works. Going
to copy and paste this URL. Open it in
my browser. Localhost 8000. Whoa,
that was really cool. Look, I've
literally just opened this in localhost
8000 and I can see that we've
automatically scraped data from Thomas
Frank, Matt Daveella, and me. Ooh, I see
that we've scraped some short YouTube
shorts for me. That was not my
intention. Uh, but like whatever. And
then I can filter it. Wow, look at that.
We have just created an online dashboard
that's running purely on our local
machine. It does. No one on the internet
has access to this other than me because
it's on local host. But if I wanted to
make this accessible via the web, what
do I do? Again, I ask Claude, "This is
really cool. You know what? Let me just
talk to it." Um, but this is on local
host. Um, how do I make this accessible
on a website so that my team can have
access to it? because this is really
nice. And then Claude will teach me all
about what deploying stuff to the web
means and then I can ask you more
questions about it and then I can deploy
it to the web and now I have an actual
website and then I might say to it but
like I don't want anyone in the world to
be able to access the website like how
can I you know how do I prevent people
from accessing it and then it will teach
me about authentication and say that oh
you should just add in a username and
password. I'll be like how do we do
that? It'll tell me you can do it
manually or you can use Google's oorthth
builtin or you can use versell
authentication. I'll be like what the
hell is Google oorthth? I can understand
about that. And now before you know it,
I have coded up a fullyfledged web app
that only me and trusted people have
access to. Or if I want, I can allow
anyone to access it and make an account.
I could even get them to pay for it if I
wanted to make it paid by just asking
Claude, "Hey, I'd like to make this app
available to paid subscribers. I want
them to be I want to charge them $19 a
month. How do we make this happen?" And
Claude will tell me, "Ah, great. Great
way of thinking about it. What you want
is a billing API. Make a Stripe account.
Go on Stripe.com. Make an account. Give
me the API key, and I'll set it all up
for you. Now, anyone in the world can
sign up to your app, but they can only
use it if they pay you $19 a month. How
amazing. You've now built a web app, a
SAS, a software as a service that could
potentially make you some money. You
have not needed to know anything about
how to code in order to do this because
all you've needed to know is how to talk
to Claude, how to be curious, how to ask
Claude to explain stuff to you that you
don't know the answers to, and the
little skill of being able to push
through the small frictions that you
will encounter. Okay, so anyway, uh how
do I make it accessible on the website?
Versell is the easiest thing. One
command deploys it to your URL like
YouTube tracker.l.app. Free tier is
plenty. Your team just visits the link.
Sick. It's even free. I don't need to
pay for it. GitHub pages. Push to GitHub
repo. Enable page and settings. Free.
Okay. Nice. Net lifi. Submit a versel.
Nice. All three host static files for
free. The one thing to think about right
now the data is a snapshot to keep it
fresh. We'd set up a GitHub action which
runs a script on a schedule and deploys
it automatically. Sick. Which option
sounds best to you? I'd recommend
Versell. It's the fastest to get
running. Nice. And I'd be like what is
versell and how does it work etc etc you
get the idea right we're building the
thing and in the process of building the
thing we are learning about how websites
work how code work how deployments work
etc etc and before you know it you will
be so far ahead of your previous self
you'll be so far ahead in terms of
understanding the [ __ ] compared to your
friends who didn't give it the time that
like you will feel like a different
person I feel like a different person
now than I did 2 months ago when I first
started discovering the stuff and I know
how to code I've been coding [ __ ] since
I was like 13 and like build software as
a service myself wasn't particularly
huge but like you know me and my brother
had built stuff when we were at
university that was making like $150,000
a year in revenue which is you know it's
not bad it's not like zillions but it's
like it's not bad for like university
kids I knew how to code and still my
mind has been freaking blown by the
power of claude code the power of just
being able to speak to the AI to get it
to interview you about what to build
that saves you time and makes you money
helps you follow your curiosity to
understand things like web servers and
versell and SSH and all this funky stuff
push through the friction associated
with dealing with API keys and stuff
like that And before you know it, you've
built some really cool stuff that
genuinely adds value to your life, your
customers lives, your boss's life if you
want to try and get a raise at work. You
can build things that track your
workouts. You can build like, you know,
I've got my Kaladins openclaw agent that
helps me track my protein and helps me
give me give me like workout
recommendations. And every single day,
me and my team, at least my team members
who are now into this AI stuff, and I
hope this video will make all of them
into the AI stuff. Actually, I'm going
to send this video to literally everyone
on my team. You then start realizing all
of the cool [ __ ] you can build to
automate aspects of your life and save
you time. It's so cool. Oh, what a time
to be alive. Anyway, before we do this
thingy, I'm like, okay, I look at this
page and I'm like, huh, this is fine,
but I don't like the fact that YouTube
shorts are showing. Okay, um, let's
forget about the web deployment for now.
Um, I don't like the fact that it's
showing YouTube shorts. I don't really
care about YouTube shorts. I only want
to show long- form YouTube videos. And
let's see what Claude says. I'll filter
out shorts by checking video duration.
Shorts are 60 or under. Just need to
pull content details from the API and
skip short videos. Sick. Do you want to
make this edit to fetch videos py? Okay.
Well, okay. Now here like green means
it's adding it and red means it's
removing the line. So here it's like
fetch per channel 30 fetch more than 10
to account for sure it's being filtered
out. Yeah. Okay. Sure. Do you want to
make this edit? Okay. So it's removing
some stuff. It's adding some stuff.
Okay. Yeah. Seems reasonable. Yes.
There's a good meme of like what it
feels like to be using forward code
these days. Here it is.
Okay. Rerun the fetch. Yep. That's fine.
Lolly gagging. That's why I like Claude.
Claude is has personality. Ah, YouTube
recently expanded shorts to be 3 minutes
long. So, let me bump the filter to
exclude anything less than 3 minutes. It
figured it out. It took a few attempts
to figure it out, but it figured it out.
Now, as you're doing this sort of stuff,
one thing you'll realize is that, you
know, while it's doing things, it's a
real challenge to figure out what to do
while waiting for it to do stuff. So,
what you do is you either chill out or
think about life or like do some
meditation and mindfulness or you do
what all the pros do and you create a
second terminal window. So while the
first one is doing working on one
feature or one thing, you've got a
second claude code instance that's
working on another thing and then you
realize that you still have time on your
hands. So then what you do is you have
four terminal windows open, each of
which are running a different instance
of claw code and you're working on a
different feature for each one. It is
annoying if you're trying to work on
four terminals for the same feature
because then they start like not
realizing that each other are working on
the same files. There's ways around it,
but like to keep things simple, you
start off with one. You realize, wait a
minute, I can just be twice as
productive if I just had another window
open. So then you make a new terminal
window. You like stick it on the side of
your screen like this. Boom. Uh you do
this one like this. Boom. You then zoom
out and now you got one terminal window
there. You zoom in here because we don't
like that. We don't like to read typing
claw over here. And now we have another
claw code window here. And then you're
doing stuff here. You're like doing
stuff here. You're working on a feature
here, working on something there. And
then you're doing stuff and you're like
still got too much time. And then
eventually you get into this like
four-way setup where you do windows like
this. This kind of thing where you have
a terminal window here, a terminal
window there, a terminal window there, a
terminal window there. You got like
everything everything with a terminal
window all over the place. I literally
just bought a 52in Dell Ultrashar
monitor for the sake of being able to
have multiple terminal windows open cuz
my Mac Studio display broke, which is
why I'm using my laptop. Anyway, um,
okay. Open local host V3 in your browser
or incognito window. Let's see what
happens. Oh, it worked. Nice. So now the
now the shorts have been gotten rid of.
Amazing. That's cool. Sick. GG. I mean,
this doesn't do anything. It just wastes
wastess token tokens, but I like to
compliment Claude when it's done a good
job. Glad it's working. Let me know when
you want to build on it. Things like
adding more channels or to refreshing
the data or deploying it for your team.
Now, this is a very basic use case, but
you've just seen without knowing
anything about how to code, literally by
asking Claude to interview me about what
I do that I find annoying. Downloading
Claude code, trying to understand what
commands it's getting me to do, I have
just built this dashboard. And this is
so so so freaking basic. This is like
the most basic thing you could possibly
build and it's still really cool. And
the great thing about this is that, you
know, I'm really not sure what I want to
build on top of this. can you ask me
some questions and help me figure out
like what what we should build next. The
key thing is when you're building AI
stuff, you don't want to build stuff for
the sake of it. You want to build stuff
that actually helps you in your work or
in your life. So, for example, you want
to build things that add value to your
customers or that save you time or that
make you money. And generally, Claude is
very smart at helping you figure out
what that stuff actually is. Here we go.
A few questions. What's the goal? Are
you tracking competitors to learn from
their content strategy or is it more
about benchmarking your performance
against theirs? Great question. What
decisions does this help you make? When
you look at this dashboard, what do you
wish you could immediately answer? For
example, what topics are trending in my
niche right now? Or which of my
competitors is growing the fastest? Who
on your team should use this? How many
channels do you want to track? How often
will you check this? Daily, weekly,
before a weekly meeting. That affects
whether we need auto refresh
notifications, etc. Okay. How is Claude
code doing the session? Good. Very good.
Thank you, Claude. I love you. Start
with whichever question feels most
interesting. We don't need to answer all
of them. And again, this is where you
just open up Whisper Flow and you just
talk. You just spiel out whatever you
want to do. And then Claude will do a
good job of helping you figure out what
feature to build next and even then
build the feature for you. And now
before you know it, you've, you know, in
my team, like Becky, my YouTube
producer, has automated like 2 days a
week of her job by just building cool
[ __ ] in Cloud Code. Uh Nicole, our
social media manager, is in the process
of learning how to use cold code and
called co-work to automate again hours
and hours and hours of manual time that
she's spending every week doing this
sort of stuff, which means she's then
able to focus on more interesting things
that are higher leverage, more strategic
stuff, coming up with new content ideas
and creative things like that rather
than the manual nuts and bolts of
pulling stuff from like the Instagram
API and trying to like copy and paste
numbers into a Google sheet. In every
person's job and in every person's
business, there is annoying tedious work
that needs to be done which is so so so
easy to automate these days. Super super
back in the day if you wanted to
automate anything you had to hire a
software developer to do this code and
then no code automation tools like
Zapier came about. Later on, at first it
was Zapier, then it was make.com and
then it was NA10. And all of these tools
were basically like drag and drop
interfaces for saying, "Hey, connect up
the data from my notion, chuck it into
Zapier, and then connect it to Google
Sheets so that I can automate that
tedious part of my job." But now, you
don't even need to use tools like
Zapier.com and NA10. Like NA10 is so
2025. Now you can literally just get
Claude Code to be your full-time genius
level software developer. It can create
all of this stuff automatically. You can
run these things called chron jobs on
your own server. You can run them as
GitHub actions. You can get it so that
every minute it pulls in data from
YouTube. Every day it like gives you a
analysis of your top 50 competitors and
like what they're doing. You can get it
so if you're running like a marketing
agency and you're trying to close a new
client, it automatically scrapes
everything on the internet it knows
about your clients. It automatically
looks up the meta ads library to see
what competitors are out there. It
automatically creates for you a proposal
that you can record a loom over and send
to the client to try and close them.
There's so much stuff you can do with
this. It's incredibly powerful. And the
way you access this power is that you
spend a few hours, you spend a weekend
just choosing to go through the friction
of like sitting down with a terminal
window talking to Claude and asking
Claude about how it can help you make
more progress in your job or in your
life, in your business, help you save
time, help you make more money. It is an
incredible time to be alive. Okay, if
you got to the end of this video, I
would love to see a comment down below.
Did you find this useful? I mean,
hopefully you did if you got to the end
of this video. Um, but I would really
love to know what more would you like to
see in our little AI education series.
I'm diving deep into all of this stuff.
I love doing like tutorials about, you
know, this this stuff because this is
the future of productivity. Back in the
day, it was like notion. Notion sort of
changed the game for people's like
productivity setups and stuff like that.
Before then, it used to be like how to
type really fast, but now it's about
like how to use AI tools to save you
time and make you money. It's just
absolutely incredible stuff. And I would
love to make more videos about this, but
I would love to know what you guys would
find interesting or useful on this
front. And if you enjoyed this video and
you want to see another video about how
specifically I use AI and a whole suite
of AI tools to improve my own
productivity within the business
context, check out this video over here.
Thank you so much for watching and I
will see you in the next one. Bye.

Watch Read Summary

What DeleteMe and Incogni aren't telling you

Reject Convenience

Heimler's History

May 01, 2026

Watch Read Summary

PDF

Prerequisites for AI Development

Identifying Projects to Build

Understanding the Terminal and Claude Code

Installation and Security

Building a Practical Project: YouTube Tracker

Scaling and Future Iterations

Takeaways

Frequently Asked Questions

What is the AI Flywheel and how does it improve productivity?

Who is Ali Abdaal on YouTube?

Does this page include the full transcript of the video?

Helpful resources related to this video

Share This Summary

Embed This Summary