How to Build with AI (Without Being Technical)

The AI coding space is exploding. Everyone's offering courses, tutorials, "best practices", or launching new tools altogether.

Use Claude Code to build an app in 5 minutes. Or Cursor? But what about Codex? No wait, Google just dropped something new – maybe that's the thing you should use instead.

Here's my anti-tutorial: skip all of that.

Because having 8 tools and 32 tutorials bookmarked will get you exactly 0% closer to actually shipping something.

I built a simple app that lets people apply my frameworks – the 10K Threshold, the Opportunity Scan, the Concept Design – without staring at a blank page. It will be part of the AI 10K Hunt. A client of mine, a coach with zero technical experience built a "mini-me" version of himself in a week that is now making him $5K/month.

No special "AI coding tool" required.

This isn't theory. You can build real, useful software without being technical.

Let me show you how.

You Don't Need a Tool Zoo

When it comes to building with AI, your options are overwhelming.

Claude Code. Cursor. Lovable. Bolt. Replit Agent. GitHub Copilot. All of them promise to make building software effortless.

But in my experience, all of them also add complexity before you've even built anything. How are you even supposed to use these tools if you don't know what you're doing yet?

They come with settings, configurations, project structures, and workflows designed for developers – not for someone who just wants to build a simple tool that solves a problem.

They're great once you've started, but not for getting started.

The point is, you need to start.

I used Claude to build my Profitable AI app and deployed it myself. My client used Gemini and hosted everything on Replit.

The tool doesn't matter.

What matters is getting moving on a path that doesn't lead to a dead-end.

What Actually Works

Forget the tool zoo. Pick ONE of these three:

1. Claude Opus 4.5 (via web app or Claude Code)

2. ChatGPT Codex

3. Google AI Studio with Gemini 3

That's it. These are general-purpose AI tools that can help you build without learning a new platform first.

Pick the one you’re already most familiar with.

Here's the workflow I use:

Step 1: Describe what you want to build. Don't overthink this. Just explain the problem and what you want the tool to do.

Step 2: Don't let it dive in directly. Before it starts writing code, ask it to make a plan. Say something like: "Before you build this, walk me through how you'd approach it. Don't make assumptions. Ask me clarifying questions first."

This saves you from getting code that solves the wrong problem.

Step 3: State what matters to you. Be honest about your constraints. I often say things like:

• "I want this to run offline."

• "I want to get this working in a day."

• "I don't want to spend more than 2 hours on this."

This changes how the AI approaches your project. It'll suggest simpler solutions instead of over-engineering.

Step 4: Host it somewhere. Run it locally, or use a service like Replit to deploy it. No servers, no domains, no infrastructure. The AI will walk you through it.

That's the whole workflow.

Describe → Plan → Build → Host. Repeat.

The One-File Rule

Here's a tip that saved me hours of frustration when starting out:

Limit your application to one file only.

A.k.a. a single-page HTML application. One file. Everything in one place. No build process, no dependencies, no folder structure to manage.

Why does this matter?

Because one file fits entirely into the AI's context window. It can see everything. It can understand the whole system. It can actually help you instead of guessing about code it can't see.

When you start with multiple files, frameworks, and complex project structures, you're creating problems you don't need yet.

This constraint also forces you to build a tool that does one job.

Don't try to reinvent the next enterprise SaaS on day 1.

Build something for yourself first. Something that solves a real problem you have.

You know your own problems better than anyone. You'll immediately know if the tool works or not. You won't need to imagine a user – you are the user.

Solve your own problem. Then worry about whether anyone else wants it.

There Is No "Best" Setup

I'll be honest with you.

I don't like built-in AI agents like the one on Replit – even though it's super well integrated and runs Claude Opus. I prefer to work in Claude (web app or Claude Code) directly, commit to GitHub, and pull from there into Replit. For smaller changes, I just copy/paste directly.

Would I recommend this workflow to others? Probably not.

But it makes me feel like I have more control.

That's my setup. I found it by doing the work, not by contemplating what's "optimal". I trust that the process will get better over time.

And that's the point: there is no "best" setup to start with.

You'll find your best setup along the way. By building things. By hitting walls. By figuring out what annoys you and what doesn't.

You already have everything you need to start: a simple chat window where you can ask any question.

Just make sure you have thinking mode and the best available model enabled.

The Bottom Line

The work might be less than you think.

You don't need to learn a new IDE. You don't need to master version control. You don't need to understand what a "build process" even means.

You need one AI tool you're comfortable talking to. One simple project. One file.

Start there. Get something working. Make progress.

That app I mentioned — the one that helps people apply my frameworks? Who knows, maybe at some point it can even generate the solutions for you.

We'll find out.

Now close this article and start building.

See you next Saturday,
Tobias