RightTech

The fastest way to turn a cheap AI experiment into an expensive mess...

is to forget two boring things:

usage limits

and security basics.

AI makes building feel easy now.

Literally anyone can now build software.

That’s the good news.

The bad news is it’s also very easy to make mistakes and expose things that shouldn’t be exposed when working with AI.

I’ve seen people spin up a little app in an afternoon, connect it to an AI model, test it a few times... then realise they left the door wide open.

No spend cap.

No access rules.

API keys sitting in the code.

An API key is basically a password that lets software talk to another tool.

If that gets exposed, someone else can use your account and your balance.

Now your “tiny test” is buying tokens all night while you sleep.

And yes... the bill can get stupid fast.

The bottleneck is no longer skill. It really is imagination at this point.

Claude Code and Codeex can basically do anything you can imagine with them.

But imagination without guard rails is how you accidentally fund a robot’s weekend.

So before launch, I check 3 things:

Set a spend cap

Hide keys properly

Decide who can access what

We’re actually entering an age where it’s getting even easier for people.

That makes the basics matter more, not less.

Build with tech, not teams...

but don’t build yourself a surprise invoice.

What’s the worst AI mistake you’ve seen so far?

Live AI builds look a lot more magical from the outside than they feel in the room.

What actually happens is pretty ordinary.

I pick a folder on my computer.

That just means the place where the app files will live.

Then I open the coding tool and switch on plan mode.

Plan mode is exactly what it sounds like.

Before it writes code, it thinks through the steps.

That part matters more than people realise...

I really want people to see how easy this is now.

Hopefully by the end of a session, you can see that literally anyone can now build software.

We’re entering an age where it’s getting even easier for people, and the bottleneck is no longer skill.

It really is imagination at this point.

From there, I give it a plain English prompt.

What the app should do.

What screen I need first.

What data it needs to store.

Then the tool starts building.

Claude Code and Codeex can basically do anything you can imagine with them, but the process still helps when you keep it simple.

Setup.

Planning.

Implementation.

Quick testing in the browser.

That last bit is my favourite.

You open the first version, click around, spot what’s off, and tighten it up straight away.

Nothing mystical.

No big team.

Build with tech, not teams.

You’ll see today exactly how easy it is with Codeex once you break it into small steps.

If you’ve never watched a live build before, that’s the real surprise... it’s calmer than you think.

What part of the process feels confusing from the outside?

If an AI tool can build the app for me... what exactly am I still meant to do?

Quite a lot, actually.

Literally anyone can now build software.

We’re entering an age where it’s getting even easier for people.

Claude Code and Codex can basically do anything you can imagine with them.

But that does not mean you hand over your brain and hope for the best.

The bottleneck is no longer skill.

It really is imagination... and judgment.

What I still do myself is pretty simple.

I define the goal.

I explain the workflow in plain English.

I test whether the thing works in the real world.

And I look for risks.

That last part matters more than people think.

It is very easy to make mistakes and expose things that should not be exposed when working with AI.

So before I trust a build, I check things like:

Who can access this?

What data is it storing?

What happens if a user clicks the wrong thing?

Does this actually solve the problem I asked it to solve?

We’ve used AI to create our own internal tools and social media systems.

This is all vibe coded.

We put our money where our mouth is.

But even then, we do not blindly trust the output.

AI is fast.

It is not accountable.

Your job is to be the one with context, standards, and common sense.

That is the work.

How much of your build process do you still keep in human hands?

A lot of software pricing works like this...

You walk in needing a screwdriver.

They sell you a space station.

That is the bit that finally started annoying me.

Not because the tools are bad.

This is not an attack on SaaS tools... software you pay for monthly online.

A lot of them are excellent.

The problem is simpler than that.

You might be paying $97 a month. Or $197 a month.

And getting a very fully featured product.

But in real life, you are often using 10% of it.

Maybe one report.

One automation.

One awkward little feature buried in tab 7 that your whole process now depends on.

We went through our own stack and found we were spending about $2,000 a month on software licenses.

Not because we needed thousands of dollars of capability.

Because we needed small pieces of a lot of different tools.

That changed how I think about software.

The question stopped being, "What platform do we need?"

It became, "What do we actually use?"

That shift matters because literally anyone can now build software.

We are entering an age where it is getting even easier.

The bottleneck is no longer skill.

It is imagination.

Tools like Claude Code and Codex mean you can build with tech not teams... and you do not necessarily need the overhead of all these giant platforms.

Not to rebuild everything.

Just to create the part you actually need.

That is usually the expensive bit anyway.

Have you got a tool you pay for mainly to use one tiny corner of it?