How I Build Web Tools with AI as a Complete Beginner
What I've found is that AI works best when you treat it as a tool for shaping your own ideas — not a replacement for having them. When I can't quite find the words for something I want to build, talking it through often helps me get there. As long as you stay clear on your vision, AI can be a genuinely good creative partner. Here is how I navigate this relationship.
I wrote about why I built Chizmo Tools in my last post. This time I want to go deeper — into the how. How does someone with almost zero programming experience actually ship web tools? Here's the specific approach I use.
Keep Each AI Instruction to One Feature
The most common mistake beginners make is giving AI a massive instruction: "build the whole game" or "finish this website." The result is code that's too complex to understand, and when something breaks, you can't fix it yourself.
My approach now is simple: one instruction, one feature. "Make this button copy the text when clicked." "Show an error message in Japanese." At this level of scope, even if I can't read every line of the generated code, I understand what it's doing. Keeping scope small is the single most important thing I've learned about building with AI.
Use CLAUDE.md to Compensate for Missing Memory
AI has no memory across conversations. Every time you open a new chat, you have to re-explain the project spec, the tech stack, and every decision you've made so far.
My solution is a file called CLAUDE.md in the root of each project. It contains the stack, hosting configuration, design principles, and a list of things not to do. When I start a new session, the AI reads this file first and immediately has the context it needs. Without it, I'd be repeating myself constantly.
When an Error Appears, Paste the Full Context
Early on, when I hit an error, I'd just ask "what does this error mean?" The answers were often unhelpful.
Now I paste the error message, the relevant code, and what I was trying to do — all together. When the context is complete, the quality of the AI's answer improves dramatically. In my experience, when an error can't be resolved, it's almost always because the information was incomplete or the scope was too large.
What I Delegate to AI, and What I Don't
I delegate the actual code writing to AI. But what to build, how to structure it, whether a feature is even necessary — those decisions stay with me.
When I was building a Roblox game, I once handed off a data persistence feature entirely to AI. The code worked, but the structure was too complex for me to understand. When a bug appeared, I couldn't fix it myself. I ended up abandoning that project.
That experience taught me that "code AI can write" and "code I can understand" are two different things. Depending on code I don't understand leads to dead ends. Now I make a point of reading through generated code — roughly — and understanding what it does before moving on.
Anyone Can Build Now — That's Exactly Why "What" Matters
AI has made it possible for people without programming experience to ship web tools. But AI won't tell you what to build. The problems you've personally run into, the things you genuinely wish existed — that's the only place original products come from.
Every tool in Chizmo Tools grew out of a place where I got stuck — in Roblox development or in everyday life. AI is just how I turn those ideas into something real. Deciding what to build is still a human job.