Subscribe
Share
Share
Embed
A compact Codex masterclass for creators, operators, and non-coders who want to turn Codex into an AI employee instead of using it like a chat box.
$32,700 in savings. 10 months of runway. non-technical founder building a real SaaS to $10K/month using only AI tools - or going broke trying.
this playlist is my full journey from $1,940 MRR to $10K MRR, filmed in real time with real numbers. the wins. the fuck-ups and the dollars. no code experience. no safety net. just AI tools and a deadline.
new videos every week. subscribe to see how it ends.
Stop asking what codex can do. Start telling the job you want done.
Most codex videos accidentally make this thing look like it's only for developers.
Terminals, repositories, benchmarks, brave faces. That is not how I use it.
Over the past three months, I've used codex for more than 300 hours as a creator operator.
And over the last month alone, I've used up over 16 billion tokens,
which is the equivalent of $8,791.
I'm using codex for everything. Emails, comments, scripts, thumbnails,
landing pages, lead magnets, launch plans.
And today, I'm going to show you the normal person version.
Teach an AI employee to turn a real audience signal into something you can share.
Why bother even using something like codex?
Well, we need to understand a few terms here.
So a chatbot like chatGPT answers what you ask.
So you ask a question, it will give you an answer simple.
Then an AI agent is codex.
And that is something that does things for you on your behalf.
Finally, a skill remembers the way you like it done.
What I'm saying here is that the flow that I've just mentioned to there,
chatbot, agent, and skill, it's not random.
What we're going to do in this video is set up the employee,
teach it how to think, give it tools, and then make it work for us.
The work is email and comments.
Then we're going to turn that into an audience signal
that is validating the content we create,
which will then be a carousel.
Then we're going to turn all of that into a landing page.
Then we're going to create a launch video.
We can do all of this stuff within codex.
And I'll be giving you the cheat sheets to do every single one of them.
It will be called inbox to launch,
because that is the sort of outcome that you're going to be getting from today.
Build once, launch frequently.
You are not bad at codex.
You're just typing like the prop costs you money.
Before we hit the ground running,
here's a really important meta skill that will help you use codex 10 times faster
and will get 10 times more use out of it in all the best ways possible.
This is dictation.
It sounds small, but it really does change how codex feels to use.
Here is why typing is friction.
Friction makes you write short prompts.
Short prompts give codex no context.
No context gives you mush.
Then you decide the tool is bad.
This is how you end up thinking that AI is shit.
That is the wrong lesson.
You didn't fail.
You've just underbreed the agent.
When I dictate, as you'll see throughout today,
I give codex the outcome I want,
the context or the surrounding background information,
the constraints and the weird preferences that I would forget
if I were typing because it's not just a stream of consciousness flow from my voice.
Codex is much better when you give it the shape of the job,
not just the task.
This is important because a normal person
doesn't really give a fuck about what an MCP server is,
a context window or prompt files.
They just want to know one thing.
What can my AI employee do for me today?
So every time we add a new capability throughout this video,
Fred will level up.
First, Fred listens, then Fred gets plugins,
then Fred gets skills, then Fred reads emails,
then Fred scans comments, then Fred builds assets,
and then Fred launches.
That is what your AI employee will be able to do
by the end of today.
Basically, Fred gets a job,
and then you get your evenings back.
If you ask codex for one giant vague thing,
you get mush.
Here is how you set it up, so you don't.
First off, plugins.
These are how codex actually touches your other tools,
like Gmail, Drive, Notion, Browser, Computer Use.
Then we've got skills,
and these are reusable ways of doing a job.
It's not a single prompt that you paste once.
It's the whole way you want a task done.
So it has things like tone, inputs,
output shape, what to ask before starting,
what never to do without approval.
Then finally, artifacts.
These are visible outputs codex creates while you work.
Think about a script, a brief, a table, a file, you can inspect.
It's not just a chat bubble that you have to scroll back
to find in a thread, it's an actual document.
Beginners ask codex for one giant vague thing in one chat,
and that is how you get mush.
The better way is give your AI employee real tools,
teach the AI employee reusable jobs,
and make the AI employee put work somewhere
where you can review it.
In short, plugins are what you use,
skills are what you build, artifacts are what you ship.
In chat GPT, you ask, can you write this for me?
In codex, you ask, can you help me build and run this thing?
Projects are what make codex more powerful than chat GPT,
because these projects mean that codex has access
to all of these files and only read access to them,
but it can also write access,
which means it can create new files
and also edit the existing ones.
Without a project, codex is just another chat window
with extra buttons.
With a project, codex has somewhere to stand.
It knows what you are working on,
what you've already tried,
and what you don't want it to do without asking.
And if you want to understand that,
codex stops feeling like just a more complicated chat GPT,
and it starts feeling like a place where real work happens.
A chat is somewhere to ask,
a project is somewhere to work.
The most important file in your codex project is not code.
It is the file that tells codex how you actually work.
Now it's time we give FWED and your AI employee rules.
This file basically just contains a bunch of rules
and information about you,
and it's the most important thing that you can do in codex period,
because codex will go and read this agents.md
at the start of every new conversation,
and it gives it information about how to operate.
For example, if you were a creator,
it would be a good thing to have in the agents.md
that joy is a creator working on project X.
In plain English, the agents.md is the operating manual.
It tells codex who you are, what matters,
how you like work done,
and what it must not do without approval.
And this is where beginners get a massive advantage.
You don't need to know how to write a good operating manual.
You just need to know how you work.
We're also going to use outcome-based prompting.
This is the best sort of prompting tip I can give you,
and you don't need to be a prompt engineer to do it.
It's as simple as this.
What is the thing?
The end result you want the AI agent to do,
or to achieve,
and then tell it to do that.
You just need to know what you want.
So now FWED and your AI employee has an operating manual
and knows the outcome that we're building towards,
which means now is time to give your AI employee its first skill.
This is where FWED and your AI employee will learn inbox
or email triage.
And this is not glamorous,
and that is honestly why it's useful.
If codex can help you sort email,
draft replies, and find follow-ups,
you start trusting it with actual operator work.
So now FWED and your AI employee can turn inbox mess
into a controlled review system.
Where it's smart, source to emails,
it saves you time,
and it allows you to focus on what matters.
And it's doing this because it's determining
what is important to keep,
what can archive, what needs to reply,
and what needs to follow up.
And finally, great system start with smart triage.
But here is the line.
Automation does not mean your AI employee starts doing whatever it wants.
For beginners, the safest automation is a scheduled summary.
But don't begin with maximum autonomy.
Begin with maximum visibility.
Now we use the same idea on comments.
And this is really where the video
that you're watching right now starts to compound.
The difference is this.
Email told us what people ask privately.
Comments tell us what they ask publicly.
And together, they become the audience signal.
Instead of building the skill first,
we'll actually run the workflow once,
prove it works,
then ask Codex to turn the working process into a skill.
Sometimes you know the recipe before you cook.
Sometimes you cook once, then write the recipe.
Here's an important point.
So I ran the prompt,
and it was having trouble using the computer
like I wanted it to.
And this is something I wanted to demonstrate to you.
So I got Codex to look into it.
And it just wasn't able to fix the computer use bug.
This is the reality of using AI.
Shit will break,
and shit will not always work as you want it to.
But don't let that be an excuse for you to fall behind
and not learn AI.
Because nothing is going to be perfect the first time.
I'm not here to come and show you the entire
polished process from start to finish
without breaking a few eggs,
because that's not the reality.
We got to deal with this shit and move on.
That took a long time.
While that was working,
I had a nap,
meditated, had lunch,
fucked around on my phone,
and then it just completed.
But the insights are super useful.
I interviewed a guy on thread
who got 150 million views in 90 days
and asked him what his method for creating content was.
He said, literally,
he will go into the comments,
look at what people are asking for in terms of content,
create content about that,
and then repeat.
It's the best way to create content
because it's pre-validated.
It's already proven to have demand
because people have literally asked for it.
The table says the audience is clustering around
one big meta-demand.
They do not want more content tips.
They want proof-backed operating systems
that explain why growth happened,
how to repeat it,
and how to turn attention into leads.
You are creating content by listening
to what your audience and people outside of it
are commenting on your posts
and building what they need
because it's already validated.
In one sentence, your audience is your strategy,
which means,
Fred and your AI employee
can now convert comments
into a demand map.
Now we turn that workflow into a skill.
This is one of the biggest ideas
in the whole video, so pay attention.
You don't need to write the perfect prompt up front.
You can run the job once,
fix the messy part,
get a good result,
and then say,
turn what we just did
into a reusable skill.
That is how we train
and your AI employee
without pretending that you know everything at the start.
Congrats.
Fred and your AI employee
can now scan audience signal repeatedly
instead of making you start from scratch.
Your AI employee has gained a 200p
and you've found the signal.
Now it's actually time to build what they want.
Now we turn signal into our first asset
and this is where the viral carousel maker comes in.
Now, the important part is not
that codex can make a carousel.
Plenty of tools can make a pretty square.
The important part is that the carousel
is downstream from the real signal,
from the audience signal scan up
that we just completed.
This means that we're not guessing the angle
we just found the painful language.
Now we just need to turn that into a visual argument.
Stop building landing pages from vibes.
Build them from the comments
people already wrote you.
Small but important point,
the carousel that we've just created
is not the asset.
The carousel is the thing that earns attention.
But if that thing works,
if the carousel works,
people need some way to go.
So now we need a simple lead magnet.
It's nothing fancy.
It's just a useful little pack
that gives people the next step.
So the carousel creates demand.
The lead magnet captures demand
and the landing page gives that lead magnet a simple home.
That auto matters.
Signal first, carousel second,
lead magnet third, landing page fourth.
This means that the page we're going to create in a moment.
It's not guessing and it's not random.
It's built from the comment language.
The carousel argument,
the lead magnet promise and the actual outcome.
The AI content forensics checklist.
Find why one post or video works,
separate real signal from weak correlation
and turn the finding into your next evidence-backed content move.
And this lead magnet is literally that checklist.
So you see how it's all coming together now.
Inbox, signal, carousel,
lead magnet.
Your content is not missing ideas.
It is missing proof.
Well, at this point,
your AI employee can now turn proof
into a page people can act on.
They can go to the landing page,
put their email address in,
get the lead magnet in return.
And at this point, you're getting leads.
Creating our viral launch video involves five steps.
The first step is where most people go wrong
and they don't even think about doing it.
It is to create a storyboard where we plan the story,
we map the hooks.
Quick recap of the inbox to launch process,
we just went through together.
We discussed plugins, skills and artifacts.
Plugins are tools.
Skills are like standard operating procedures
or instructions and artifacts are outputs.
Like the landing page,
like the lead magnet,
like the viral launch video, etc.
Plugins are what you use,
skills are what you build,
artifacts are what you ship.
Throughout this entire video,
we didn't have to do any coding.
We didn't have to open up a terminal.
We didn't have to write any code.
Our AI employee did all of that stuff for us.
The key thing to remember there was outcome-based prompting.
Always put in your prompt.
The outcome I want for this is this.
Because that is the end result that you want.
We gave our AI employee the ability to use the computer
to manually go ahead and scrape all of those comments.
And the prompt that we use got our AI employee
to turn these comments into an actual signal table
that consisted of pains, questions, objections and quotes.
Then we ran this new sort of workflow
of asking our AI employee to perform a task
and once it's completed that task properly,
we then get it to look back at everything it's just done
and turn that into a skill.
All of that stuff is free inside
of my free school community full circle.
Link to that will also be in the description below.
This entire masterclass took me three whole days to film,
so if you got any value from it whatsoever,
why can subscribe?
It really helps me out.
And until the next one,
don't be a dropkick.
Peace.