There's a new king of AI coding tools
and its name is Claude code. But 6
months ago, I was using Claude code all
wrong. I was basically treating it like
chatbt with a save button, typing in
vague prompts, and just hoping for the
best. It was like buying a Ferrari, but
never taking it out of first gear. Fast
forward to today, and I'm a Cloud Code
Power user. I can run multiple instances
of Claude Code at the same time and work
on a week's worth of features in just a
few hours. See how each Claude Code
instance is working on different app
feature completely independent from the
other. And yes, Claude Code will merge
all the changes together at the end. In
this video, I'm going to save you 6
months of pain by sharing 36 specific
lessons about how to get the most out of
Claude Code. These lessons are going to
be organized into three clear levels:
beginner, intermediate, and master. In
level one, the beginner section, we're
going to cover Claude code foundations.
We'll cover installation and setup,
essential concepts and commands, and how
to use things like the claw.md file to
put you ahead of 99% of beginners. In
level two, the intermediate section,
we're going to look at workflow
enhancements. We'll cover claiming mode,
using Claude for research and
documentation, and using Claude code to
automatically answer issues and create
PRs in GitHub. And finally, level three,
the master section. We're going to look
at advanced techniques. We'll cover
things like using multiple claude codes
at the same time to work on multiple
different features, custom commands, and
custom sub aents, and using Claude code
with MCP servers. So whether you're a
Claude Code beginner or power user like
me, I can guarantee that you're going to
learn something from this video that's
going to help you cook at a higher level
with Claude Code. But before we dive in,
I'm Afar and I teach people how to use
AI coding tools, even if you're new to
coding. So if you're trying to level up
with AI coding tools, maybe it's for
your job, maybe it's for a new business
you're trying to start, maybe it's for a
side project, please subscribe to the
channel and turn on notifications for
more videos about the best AI coding
tools and how to get the most out of
them. With that out the way, let's get
into level one, the foundations of
Claude Code. All right, so if you're new
to Cloud Code or feel like you might be
missing some basics, this section is for
you. We're going to cover everything you
need to get up and running with Cloud
Code, including things like installation
and setup, basic commands, project
configurations, and debugging. First up,
installation. The way that you install
Claude Code and the way that the
majority of people use it is just on
their local laptops or their local
machines. You go to the Enthropic
website, copy the command for claude
code, and install it on your local
machine using your terminal. It's quick,
it's easy, and you can begin coding
right away. Option two, and this is
great for using Claude Code for back-end
projects, is to install Cloud Code on a
remote server. Just log in to wherever
the server is hosted, for example, AWS,
Digital Ocean, or Herzner, and install
Cloud Code just like you would things
like Python or Node. A bonus of the
remote server approach is that this
allows you to code from anywhere. For
example, you can use the termius app on
iOS to control claude code right from
your phone. Tip number three is that you
can use claude code inside other AI
coding tools like cursor, windf.
This is great if you're already familiar
with those tools or still getting
comfortable with the terminal. To get
started, navigate to the directory of
your choice and type claude in your
terminal. You can also use the resume
flag. That's d-res
any previous session. Tip number four is
to use my favorite feature of Claude
Code and that's to-do list. This is the
thing that made Claude Code stand out
for me among all the other AI coding
tools at the time. It didn't just jump
into writing code straight away. First,
it created a to-do list. Now, this
happens most of the time automatically,
but you can also explicitly prompt plot
code to use to-do list in the prompt
that you give it. The great thing about
to-do list is that as Claude works, it
checks off items so that you can see
what's happening and what's coming next.
And it also enables Claude to work on
big complex tasks without getting stuck
in loops or erasing past work. Next,
let's look at some essential commands.
First up is bash mode. Bash mode enables
you to run bash commands inside Claude
code so you don't have to exit out of
it. Claude can also run bash commands
itself, meaning it can do things like
read and write files, search, and even
handle git commands for you. For
example, you can tell Claude to read all
the files in a certain folder and use it
for context for tasks it will do. Tip
number six is to use Claude for instant
documentation. Documentation doesn't
have to be painful. Just ask Claude to
explore and explain the architecture of
an app and how it works and save its
output in a file called something like architecture.md.
architecture.md.
This is great when you're getting up to
speed with a new project or touching a
project that you haven't worked on in a
couple weeks or just documenting key
things for you, your teammates, or even
Claude to reference later. Tip number
eight is auto accept mode. Claude will
usually prompt you for permissions to
make changes. If you want to let Claude
take the wheel, you can turn on auto
accept mode by pressing shift and tab
together. This enables Claude to make
changes without asking each time. Tip
number nine is about model switching.
Now, Claude code allows you to choose
from a variety of anthropic models to
power its AI coding sessions. The /model
command allows you to choose which model
to use. You can use Opus, which is the
most powerful model when you need deeper
analysis or want the most powerful
coding capabilities for your toughest
challenges. And you can use Sonnet for
routine tasks. So if you just want to
save tokens or get faster responses. I
usually keep it at the default mode,
which is using Opus until you've hit 50%
of your usage limits for that month. But
my new favorite is actually the Opus
plan mode, which is where it uses Opus,
the more powerful model to plan, and
Sonnet, the more costefficient model, to
actually write the code and do the
execution. It's something I recommend
you play around with and see which model
works best for your use case. Next up,
don't be afraid to interrupt Claude. The
escape key is your friend. You can press
the escape key once to interrupt Claude
and twice to go back to a previous
prompt in your session. Don't hesitate
to interrupt Claude if you see it
heading in the wrong direction. It's
much better to redirect early than to
waste precious tokens. Next up in the
beginner section, let's talk about
fixing your mistakes or debugging with
Claude code. It's actually much simpler
than you think. First up is a screenshot
method. Now, claude code can take images
as input. So, if you run into any UI
issues or any bugs in your application,
you can actually take screenshots and
give them to Claude Code as inputs for
your prompts. This allows Claude to see
all the context and it usually can tell
exactly what went wrong or it allows it
to pinpoint exactly where in the UI you
want to make a particular fix. Now you
can also use the image input for other
things like giving it UI designs and
diagrams as well, but I think the most
common usage is for debugging. The next
level of debugging is letting Claude
write your tests for you. This is as
simple as just asking Claude to write
tests for a particular feature or write
tests for the new onboarding flow. Just
make sure to focus on general end toend
tests rather than implementation
specific ones. And we can take this
further with tip number 13 which is
using cloud code for testdriven
development or TDD. Now test-driven
development is a really popular software
engineering philosophy. But when it
comes to Claude code, essentially what
this means is asking Claude to first
write tests for a given feature and then
to implement it. This helps Claude catch
issues early and it gives it a bit of
structure to work with, which can lead
to better outputs. Next up is project
configuration. And that brings us to
what I think is the most important thing
in this video, which is the claude.md
file. You can think of the claude.md
file as your project's memory. It's a
file that gets added to context every
single time Claude Code does any task
and it's all the information that you
want to give Claude Code to work on your
project your way. Here are some snippets
of my claude.md file for an iOS app that
I'm working on. It consists of things
like Git workflows and best practices,
project and architecture overviews,
build commands and best practices for
things like testing and debugging, and
analytics and documentation how-tos.
Now, the great thing about putting these
in the claude.md file is that Claude
will follow these rules automatically.
It'll create proper branches, commit
regularly, and never accidentally push
to main. It'll follow my preferences for
documenting its work and always add
tests and analytics when building out
new features. This ability to follow
complex rules and tailor both workflows
and output has made Claude my favorite
AI coding tool, outshining the likes of
cursor and replet. Now, before you worry
that you have to manually write your
claw.md file, you can actually just ask
Claude to write it for you. That's going
to be a recurring theme in this video.
Using Claude code to help you use Claude
code better. What this means in this
case is asking Claude to update the
claw.md file as you add specific
features to your project or as you
discover preferences that you want to
make Claude respect. Tip number 15 is
the message Q. Now the message Q is a
real gamecher because it means that you
don't have to wait for Claude to finish
its current task before you give it
input. You can just type a message while
Claude is working and it gets added to
the message queue and Claude will act on
it once it's finished with the current
task. Just add tasks as you think of
them and Claude will work through them
all in order. And that brings us to tip
number 16, which is the final tip in our
beginner section and that is for long
prompts, you can actually put your
prompt in a markdown file and feed it to
Claude by referencing that file in the
prompt using the at symbol. This is much
cleaner than trying to type everything
in the terminal and it's great if you
have long prompts if you want to think
things through before you give it to
Claude. Now, that brings us to the end
of the foundation section that gives you
everything you need in order to get
started with Claude code. Let's take it
up a notch to the intermediate section
where we're going to look at Claude Code
workflow enhancements. Now that you're
comfortable with the basics, it's time
for level two or the intermediate
section and that's about working
smarter, not harder. I'm going to share
with you the enhanced workflows that I
use in Claude Code to plan better,
research faster, and integrate with my
existing GitHub workflows. First up is
planning and thinking features that are
going to help you level up and enable
Claude code to tackle your toughest
problems. Tip number 17 is to use
planning mode. Now, most of the time,
you don't just want to jump straight
into writing code. Often times, you want
to plan and strategize before you start
implementing. To do this, use the
planning mode on Claude Code and ask it
to plan before it starts working on a
new feature. Just hit tab and shift
until you see the plan only mode and
that will ensure that Claude gives you a
plan to review before any code is
written. I use the planning mode all the
time for architecture decisions or to
force Claude to think through how to fix
nollie bugs before it takes action. When
it comes to planning, a hidden setting
to make use of is the opus plan mode.
This means that you can use opus for
planning and sonnet for implementation
just by selecting the /model command.
This is a good default to ensure that
you get the most powerful model for
planning but the cheaper model for
implementation. Another underrated
feature when it comes to planning is to
use sub agents to create multiple plans
for how to do let's say implementing a
given feature or to fix a certain bug
and let you or Claude pick the best one.
Look out for more on planning and
parallel sub aents in the advanced
section of the video. Next up, did you
know that you can control how much
Claude thinks when it's working? Meet
the think keywords. These allow you to
tell Claude how much time to spend
thinking when you're working on tough
problems. There's actually three think
keywords. There's the basic think,
there's think hard for deeper analysis,
and for maximum thinking power, use the
keyword ultra think. Just include the
appropriate thinking keyword in your
prompt and it will tell Claude to use
more of its thinking budget. Now imagine
you're prompting Claude to add a new
feature and you're using the think hard
keyword in your prompt. But you're also
using planning mode and you're saving
that output into a markdown file for
future reference. That's just one
example of combining tips that we've
seen in this video into a powerful
workflow for vibe coding. The next
unlock for me was that I could use
Claude code for more than just code. My
favorite example of this is research.
Claude has a web search and web fetch
tool that enables it to search the web
and read website content and take action
on it. This means that you can use
Claude code to research small things
like how to use a Stripe API or ask it
to do larger research tasks like put
together a report on the best O tool to
use in your application that also has
pre-built login components. It's that
powerful. This saves you a ton of time
and allows you to take into account more
of your app's context as you add tools
and features. Another hidden feature
when it comes to research is that Claude
Code can read PDFs. This means that it
can combine information that you have in
PDF files with information that it finds
in its web searches. My favorite example
of how to use this feature is feeding
Claude PDFs of Chad GBT deep research
reports so that it can reference those
in its research phase. Beyond research,
Claude code is great for document
generation. This is important because
I've noticed that you need to think like
a product manager and give Claude code
all the context that it needs in order
to do its job well. The good news is
that you don't have to write all of
these product specs by yourself. You can
actually leverage Claude to generate
artifacts that both you and Claude
reference later. I use Claude to
generate things like product requirement
docs or PRDS for features, user
experience guides, API documentation,
and technical design docs. I find these
docs to be more useful than the ones
made in chatbt because they take into
account more of the actual structure of
my application and leverage all the
context of my project for better
results. Tip number 26 is something that
I use to stay organized in the projects
that I build and that's using claude
code for automatic change tracking. This
is about leveraging clawed code for
things like creating change logs for it
to reference later, documenting the most
up-to-date features of your app for use
on your website, or making things like
decision docs, which outlines why you
made a particular decision. I highly
recommend leveraging Claude Code for
keeping track of how your project
evolves over time. Your future self and
your future Claude Code agents will
thank you. Next up is a Claude Code
GitHub integration. Claude Code is
fantastic for teams thanks to its deep
integration with GitHub actions. Now,
you may remember how you can use Claude
Code to handle Git commands for you as
you develop features. But this Claude
Code and GitHub actions integration
takes it to the next level. First, run
the /install GH actions from the Cloud
Code menu. And once you've set that up,
you can tag Claude Code in issues and
pull requests and ask it to fix it. and
Claude code will run using GitHub
actions and submit a PR or a fix without
needing to run on your local or your
remote machine. Claude can even review
PRs automatically as they come in.
Finally, in this intermediate section, I
want to talk about the mindset that
helps me get the most out of working
with Claude code. For best results with
working with Claude code, you need to
think like a product manager. And this
means two specific things. Firstly, it
means giving Claude lots of clear,
relevant context and constraints when it
comes to achieving your goals. Secondly,
and probably most importantly, it means
letting go of reviewing the code
yourself or needing to understand
everything that Claude has implemented
and working to verify Claude's outputs
at higher levels of abstraction. What
this means is that you don't need to
understand every line of code outputed
by Claude. Can you verify that the app
experience is what you want? Can you
verify that the tests pass? Can you
verify that the app works as intended?
That's what matters. This approach
literally changed how I built with AI
coding tools. And it's a mindset shift
that I had to make, similar to what I've
experienced in real life when I had to
work and manage smart technical people
on my team at previous jobs that I've
held. Now, this brings us to the end of
level two, the end of the intermediate
section of Claude Code. I'm sure you're
seeing just how powerful Claude code can
be and are probably super excited to use
it in your own projects. But hold up
because we have more to learn in level
three and that's the master section
where we're going to go through advanced
techniques. Level three is where you
become a Claude code master by
leveraging advanced techniques that
allow you to feel the full power of
Claude code. We'll cover things like
using multiple instances of claude code
to develop multiple features at the same
time, custom commands, custom sub aents,
and using claude code with MCP servers.
First up, let's talk about working in
parallel or using multiple Claude codes
at the same time. For those of you
keeping count, this is tip 29 and it's
using parallel sub aents for planning
and exploration. This allows Claude to
explore multiple solutions
simultaneously and then you pick the
best one. It's like having a team
brainstorming session. All you need to
do is ask Claude to plan and use
parallel sub aents to explore solutions.
Parallel sub aents are great if you want
to explore how to add complex features
to your app or if you want to fix gnarly
bugs or problems. Now remember at the
start of this video where I used three
claude code instances to work on three
features at the same time. I'm going to
show you how to do that using the power
of cloud code and get work trees. Get to
work trees give you isolation that
enables multiple claude agents to work
on your project without conflicts. Just
create a hidden folder in your project.
I called mine trees and use the git add
commands to create a work tree for each
different feature or thread that you
want to work on. Once the work trees are
created, you can spin up a terminal
session for each work tree and then run
claude code in each. Here's me doing
that in cursor, but you can also do it
in your terminal. This gives you a
beautiful clean git history that's easy
to understand and review. Using cloud
code and git work trees allows you to
work on multiple features at the same
time on the same codebase without
running into issues down the line. When
you're done, just ask Claude Code to
merge each of these work trees in your
folder together and fix any conflicts.
And boom, thanks to Cloud Code and using
multiple Cloud Code instances and Git
work trees, you can have a week's worth
of development done in just a few
minutes. Now, that's how to use Cloud
Code in parallel. Next up is the
advanced customization section. And the
next tip is using custom commands.
They're basically like having prompting
shortcuts for Claude Code to act on
repetitive tasks versus typing out the
same long prompts over and over again.
Now, the easiest way to make
custom/comands is just to ask Claude to
do it for you. But if you want to do it
manually, here's how to set up custom
commands yourself. Create a commands
folder in the Claude folder in your
working directory. Then add a markdown
file with the name of your command. For
example, changelog.md.
Your command file should have a
description of what the command does,
the list of allowed tools, usually
things like bash, read, and write, and
the actual command prompts, which is
what you want the command to do. Command
prompts can go beyond just plain text
and include things like arguments for
dynamic input, bash commands, and things
like file references. A pro tip here is
that custom commands can be project
specific or they can apply to all your
Claude Co projects and this is called a
personal command. One more thing to note
here is that custom commands don't
automatically get added to context. So
if you want something to always be added
to context, for example, information
about your application, put these in the
claw.md file, that always gets added to
context. Custom commands don't. This
brings us to tip 34, which is using
specialized sub aents. Just like custom
commands, you can create custom sub
aents. These are specialized AI systems
with custom prompts and tool permissions
that can handle specific tasks
autonomously. And just like custom
commands, they can be project specific
or they can be personal and apply to all
projects. For example, you may want to
create custom or specialized sub aents
for things like user experience design,
API design, security review, running
tests or database administration and
other domain specific tasks. Each sub
agent has specific tools that it can
access and a domain specific purpose
reflected in its system prompt. The
easiest way to create custom sub aents
is to use the / aagents command in
claude code. This is another example of
using Claude Code to help you use Claude
Code better. There's a wizard that walks
you through setting up a new agent. And
this is much easier than setting it up
yourself because Claude does the heavy
lifting for you. In this example, I'm
setting up an agent to do mix panel
analytics. And here's what's going on
under the hood. Each agent has a
markdown file which specifies its agent
type, usage, accessible tools, and
system prompts which gives you an
overview of what the agent does. You can
view all the agents that you've created
with the / agents command. Once you've
created your sub agents, Claude will
automatically know that they exist.
It'll understand what it's used for
based on what's in the usage field
description in your agent markdown file
and it'll automatically delegate tasks
to your sub agent when relevant. Claude
will also use these sub agents if you
explicitly ask for them in your prompt.
So for example, you could say, hey, use
the testr runner sub aent in order to
run these tests or use the database
administration agent to make sure that
the queries are correct for this new
feature. Now the final thing in this
advanced section is using cloud code
with MCP servers. Now what happens when
you want to extend Claude codes
capabilities to take actions in third
party tools like databases or tools like
Figma or Canva? This is where model
context protocol or MCP comes in. Here
are some ways that I like to use MCP
servers with cloud code. First up is
database MCPS. These are great for doing
direct database work. I've used the
MongoDB MCP, the Postgress MCP server,
as well as a Superbase MCP server for
various database related things in the
apps that I've built. Next up is a
Playrite MCP for browser automation.
This is great because it allows Claude
code to actually visually see your
application and your web UI, and it's
great for things like testing and
finding and fixing bugs. Finally,
there's a Figma MCP, which is great for
going from design to code. Now, there's
many other MCP servers out there. I'm
actually curious if there's a particular
MCP server that you found most useful.
Let me know in the comments. But those
are just three that I like. These
capabilities are incredibly powerful and
I think they'll become even more
powerful as the MCP server ecosystem
matures. So, that brings us to the end
of level three. That's the master
section of Claude Code. We've gone from
beginner where we've looked at the
foundations to intermediate where we
looked at enhanced workflows to the
master section where we've covered
advanced techniques for getting the most
out of claude code. Congrats on making
it this far in the video. But before we
end off, let's talk about a very
important thing with AI coding tools and
especially with Claude Code, and that's
money. How much does Claude Code cost
and is it worth the money? Now,
unfortunately, there's no free version
of Claude Code, but there is two ways to
access it. The first way is via one of
the Claude paid plans, and there's two
of them that you can choose from. The
first one is a $20 a month pro plan. You
can use this to get started or to play
around with. And when you're serious
about building, I'd recommend the Max
plan, which goes for either $100 or $200
a month. The Max 5X plan gets you 5x
higher rate limits than the Pro plan for
$100 a month. And the Max 20X plan gets
you 20x higher rate limits at $200 a
month. This is the plan that I
personally use and I'd recommend it if
you're serious about AI coding tools,
but you know, pick whatever plan fits
your budget. The second way to use cloud
code is through the anthropic API, but
this is just ridiculously expensive and
unless you're part of a large company
that will pay for the anthropic API
usage for you, I'd recommend picking a
claude pro or max plan that fits your
budget and not looking back. Quick note,
Anthropic announced that at the end of
August, they're going to be imposing
additional rate limits, and that's
weekly rate limits on Claude Max users.
I'm not quite sure how this is going to
impact Claude users. I don't think the
impact is going to be that big. I still
think Claude is good value for the
money, but we'll have to see what these
rate limits do come the end of August.
That brings us to the end of this video.
I hope it helped demystify Claude Code
and made you super excited to build and
unleash your creativity with Claude Code
by your side. If you learned something
from this video, don't forget to like
and subscribe. And if you thought there
was a feature that was really cool or
maybe something that I didn't cover in
this video, it could be about claude
code, it could be about AI coding in
general, let me know in the comments and
