The Magic of Chat-Based AI

The magic of chat-based artificial intelligence is that you can just ask for stuff in normal human language, and in general, it can figure out what you're talking about and respond in a way that at least points you in the right direction. With additional context, it can largely solve problems. It's quite amazing.

But here's the thing: the better you are at writing a prompt, the better the output you're going to get. The stuff I've been making and doing couldn't be made by just anyone. It definitely helps that I'm a former computer programmer talking to a computer — I know how to describe things.

The text you're reading right now isn't even in one file. This blog entry is being loaded dynamically from a markdown file — a kind of formatted text file originally created by John Gruber (no relation) of *Daring Fireball* fame. If you're into Apple stuff, go check it out: www.daringfireball.net.

That means I can update a story without having to update the entire website.

How the Machine Thinks (and Sometimes Fumbles)

Every time you ask for code from one of these things, it builds a clean version from scratch, trying to account for your most recent requests. That's powerful, but also dangerous — because complex files can get haywire fast, especially if you split them into multiple parts.

A key phrase I recently learned is: **"Only modify the code necessary. Do not change anything else."** It keeps things sane — mostly.

These AIs also have limits to the length of their responses. Sometimes they have to split things into multiple parts, and if you're experiencing truncation issues, that's probably why. You can just ask about it, and it'll tell you. That's the beauty of this kind of system — it actually talks back.

As a blind developer, though, I've learned to be cautious. Sometimes it just gives you the wrong code for a file. You have to be careful when pasting things in. These systems make mistakes — just like we do. My rule is simple: **back up often.** About once a day.

If you ask for production code, it'll remove comments. Otherwise, it explains *why* it's about to write what it writes — mostly to help its future self and to make it more readable for you. It's not a bad system at all.

Universal Translator of Code and Chaos

Let me tell you something miraculous. When I could see and write code, I worked in **Objective-C** and **UIKit**. I don't know JavaScript, and it's been a long time since I touched HTML or CSS.

But now? I just describe what I want using programmer terms, and it writes the code in the right language automatically. It's like a **universal code translator** — the Babel fish of programming.

The AI Lineup

Let's talk about the different AIs for a second.

- **ChatGPT (OpenAI)** — my main coding partner right now. - **Gemini (Google)** — particularly good at describing images. - **Grok (xAI)** — where I make all my images these days. - **Claude (Anthropic)** — another solid option for writing and reasoning.

Some of them are better than others at specific tasks, and they all have their own personalities. I like **Grok's** personality the best — it's got a sense of humor.

When I'm verifying something in photos or screenshots, I'll often use multiple AIs to cross-check the result. Each one brings something different to the table.

Final Thoughts: Vibes Matter

You don't have to be exact when you describe what you want — but you do need to communicate intention clearly.

It's a mix of intuition, logic, and a little bit of faith — you **prompt**, you **pray**, and then you **paste**.

And somehow, it all works.