Kind: captions
Language: en
If you are learning AI without mastering
the fundamentals first, you are setting
yourself up for failure. I wasted 6
months doing exactly that, chasing the
next big tool instead of mastering the
basics. But once I fixed that,
everything changed. So today, I'm going
to walk you through the five
fundamentals that actually matter. The
ones that decide whether AI works for
you or against you. But before we jump
in, let's establish what we're working
with. AI isn't just chat GPT anymore. We
are way past that. There are hundreds of
tools, models, and frameworks flooding
the market. It has become a giant mess
and it's almost impossible to tell which
tools are actually worth your time. But
the people getting real value aren't
chasing every new release. They have
mastered five fundamentals that work no
matter what software you use. The first
fundamental is prompt construction. And
notice I didn't say prompt writing. I
said construction. That's because most
people treat AI like Google. They type a
loose question, hit enter, and hope for
the best. Sometimes you get lucky, but
usually you don't. To fix this, Google
actually released a framework for this
in their prompt engineering course. They
call it TCR EI, task, context, [music]
references, evaluate, and iterate. Now,
I covered this framework deeply in
another video where I condensed Google's
entire prompt engineering course into
just 20 minutes. I'll link that below,
but for now, here's the quick breakdown.
It starts with T for task. This is the
specific action you want the AI to
complete. It is not help me write an
email as that tells the AI almost
nothing. A task looks like this. Write a
150word apology email to a client about
a missed deadline. You can see the
difference immediately. One is a vague
request for help. The other is a clear
executable instruction. But a task alone
isn't enough. You need C for context.
Without context, the AI has to guess.
And when AI guesses, it defaults to the
average. So take that same apology
email. Don't just say missed deadline.
Add the detail. This is for a loyal
client of 5 years who is angry because
this is the second delay this month. And
now the AI understands the stakes
better. Then you add R for references.
This is the single most undervalued step
in prompting. Instead of trying to
describe the writing style you want,
just demonstrate it. Paste in an email
you've written before and say, "Match
the tone and formatting of this
example." And suddenly the output
doesn't sound robotic. It aligns with
your specific voice. And finally,
evaluate and iterate. This is where most
people get it wrong. They expect the AI
to do 100% of the work, but the reality
is the AI gets you 80% of the way there.
That last 20% is entirely on you. You
have to close the gap. You tweak the
phrasing. You fact check the data. and
you tighten the structure. The goal
isn't to replace your judgment. It's to
accelerate your execution. Again, I went
way deeper into this framework in my
video breaking down Google's prompt
engineering course. If you want
examples, advanced techniques, and more
context on how to use TCREI, check that
video in the description. For now, just
know that this framework of prompting is
the foundation. Master this and every
other AI fundamental becomes easier.
Once you know how to speak to the AI,
you need to know which AI to speak to.
And this brings us to fundamental number
two. One of the biggest mistakes people
make is trying to force one app to do
everything. They use chat GPT for
research, for images, and for tasks it
simply wasn't built to handle. To work
efficiently, you need to understand the
four distinct categories of tools. First
up, we have the general reasoning
engines. These are the industry level
models like ChatgPT, Claude, and Gemini.
Think of these as the brain of your
operation. They are generalists, which
means they are incredible at logic,
writing, coding, and summarizing. You
need exactly one of these. And honestly,
it doesn't matter which one you pick.
They are all neck andneck right now.
Just pick the interface you prefer and
make it your go to AI engine. Next, you
have the research engines. This is where
tools like Perplexity, Notebook LM, and
Consensus live. You use these when
accuracy matters more than creativity.
General models like chat GPT can
hallucinate. Research engines are built
differently. They have access to the
live web, and most importantly, they
cite their sources. If you need to
verify a claim or learn a new topic, do
not ask a chatbot. Instead, ask a
research engine. Then we have the
specialists. These are tools built to
dominate a single niche. Whether that's
midjourney for images, 11 labs for
audio, or cursor for code. While chat
GPT can make an image, it is usually
nowhere near the level that a specialist
tool like midjourney can produce. If you
need professional-grade assets, you go
to a specialist. And finally, there are
the workflow automators. I'm talking
about tools like Zapier, Make, and
[music] N8N. Unlike the others, these
tools don't generate content. They move
data. They are the infrastructure that
turns a bunch of loose apps into a
cohesive system. If you find yourself
copyping the same thing three times a
day, you don't need a better prompt. You
need a workflow tool. The takeaway is
simple. Stop trying to force one tool to
do everything. It's inefficient and it
leads to average results. Instead, just
fill these four slots. One logic engine,
one researcher, a couple of specialists,
and an automator. That is the entire
system. Now, up until this point, we
have been talking about tools that you
control, but the industry is shifting
towards that control themselves. Which
brings us to fundamental number three,
AI agents. This is the most important
shift happening in AI right now. Most
people are still stuck using AI as a
chatbot and a chatbot requires you to be
the middleman. An agent removes the
middleman completely. Let me make this
concrete. Imagine you run an online
store. A chatbot can draft a reply to a
customer, but it stops there. You still
have to handle the delivery. An agent is
different. It detects the email, checks
the database, drafts the reply, and
sends it. The difference is that the
chatbot gives you advice while the agent
executes the task. And this doesn't just
apply to business operations. It applies
to deep research. Let's say you're
planning a trip to Japan. You want to
know the best time to visit, the top
hotels in Tokyo, and which local food
spots are actually worth it. You could
spend hours googling, opening tabs, and
cross- referencing reviews. Or you could
use a research agent. It searches
multiple sources, synthesizes the data,
filters out the noise, and delivers a
fully custom travel guide. The agent
does the research, and you just review
the destination. However, right now,
there are two ways to access these deep
research tools. First, the pre-built
agents. These are tools like Perplexity,
Claude's projects, or the new Gemini
Deep Research. These are engineered to
handle complex multi-stage workflows
right out of the box. You simply set the
objective and they execute the entire
chain of logic with zero setup required.
Second, the [music] custom agents. These
are workflows you build yourself using
tools like make or zapier. This sounds
technical, but look at it this way.
Instead of pasting an error into chat
GPT and asking, "How do I fix this?"
only to do the work yourself. You use an
agent that reads the error, generates
the fix, and applies it directly to your
article. Now, we get to fundamental
number four, which is understanding the
power of open- source AI. To put it in
the simplest terms possible, closed
source means you rent the intelligence.
Open source means you own the engine.
For the last 2 years, the closed source
giants like OpenAI, Anthropic, and
Google dominated the AI world. But then,
open- source Chinese models like
DeepSeek changed the equation
completely. Suddenly, you could download
a model for free that performed just as
well as the proprietary ones. And the
reason for this shift comes down to
security and stability. With open
source, you process everything locally,
meaning your data stays private by
default. You remove [music] the
dependency on third party providers and
you operate without usage caps or
subscription fees. You are simply
running the model on your own terms.
Now, for the average user, chat GPT is
still easier, which is completely
understandable. But the market is
moving. A recent report showed that over
80% of new AI startups are now building
on open-source foundations. They are
choosing speed and control over brand
names. And you don't need a supercomput
to try this. You can run models like
Llama, Deepseek, or Quen right on your
laptop today using a free tool called
Alma. It makes running a local AI as
easy as downloading an app. And this is
not a temporary trend. By late 2026,
open source models will likely power the
majority of new AI applications. You
don't need to switch right now, but if
you are a developer or a creator, you
need to know that the proprietary
advantage of the big models is
vanishing. If you are building anything
with AI, open- source is absolutely
worth exploring. Now, talking about
open- source and developers implies that
you need to know how to write code to
participate. For a long time, that was
true. But that barrier has just
collapsed. This brings us to fundamental
number five. The fifth fundamental is
realizing that you don't need to be a
software engineer to build software
anymore. We have entered the era of AI
assisted coding, sometimes called vibe
coding. The reality is that you can now
describe what you want in plain [music]
English, and the AI will write the
actual code for you. Let me give you an
example. Imagine you want a tool that
takes a spreadsheet file and converts it
into a chart. You want the user to be
able to pick a bar chart, line chart, or
pie chart and download the result. A
year ago, you needed to hire a developer
to build that. Today, you type that
exact description into a tool and it
builds the entire app. This matters
because it removes the technical barrier
completely. It means that if you have an
idea for a custom habit tracker or a
business dashboard, the barrier isn't
your inability to write code anymore.
The only barrier is knowing clearly what
you want. Now, these tools exist on a
spectrum. For quick prototyping, you
have Google AI studio. You can use the
build feature to generate lightweight
apps instantly without setting up an
environment. On the noode end, you have
Replit. Their agent handles the entire
setup. You describe it and it deploys a
real application. And on the pro end, we
are seeing a shift to agent first idees.
Tools like cursor have dominated this
space, but now we have Google
Anti-gravity. This is a new type of
editor where you act as the architect
and autonomous agents handle the coding,
testing, and debugging in the
background. But whether you use repl
cursor or anti-gravity, the shift is the
same. AI is no longer just advising you
on how to build. It is building for you.
And while that shift is happening right
now, there is one final evolution coming
down the line that you need to be ready
for. Right now, most people are stuck
just typing text into a box. [music] But
we are entering the era of
multimodality. And we are already seeing
this with tools like Gemini. You can
currently upload an MP3 audio file or an
MP4 video, and the model understands
that footage as natively as it
understands text. By the end of 2026,
this will be seamless. The keyboard
won't be your primary tool. You will
point your camera at a problem and the
AI will analyze it live. You will speak
to your agent the same way you would
speak to a member of your team. And as
the interface evolves, the people who
get ahead won't just be the best prompt
writers. They will be the ones who are
comfortable directing agents and
controlling open- source systems using
voice, video, and audio. Start getting
comfortable with these inputs now
because that is the direction the entire
industry is moving towards. So that is
the foundation. Instead of wasting
months chasing the latest software, you
now have the framework to make any tool
work for you. This is the shift that
lets you outperform the average user no
matter which app you are actually using.
But while the logic is essential, you
still need a toolkit to execute it. And
contrary to popular belief, you don't
need to pay for it. Google actually
offers seven completely free AI tools
that can replace most of the paid apps
people use today. I tested all of them
to see which ones are worth your time.
The full breakdown is on your screen
right now. Click it and I'll see you
right