With all the fear, uncertainty, and doubt swirling around AI, I want to step back and explain how I actually use these tools—and why they’re not going to put programmers out of their jobs anytime soon.

I use AI every day. I have Claude (both with Claude Code and through Cursor) for coding, and ChatGPT for research, brainstorming, and expanding references. They’re valuable tools, but not magic. The difference lies in how you use them.

Why I use AI (and How)

I don’t “vibe code.” My projects are too large for that; the context breaks down fast. Instead, I give AI detailed prompts—documents that define the architecture, decisions, and examples. With that direction, AI performs far better.

That’s because I don’t ask AI to program. I ask it to code.

This may sound like a meaningless distinction, but it’s not. It’s the most important distinction to understand.

Programming vs Coding

A programmer designs and manages programs: defining architecture, analyzing requirements, breaking work into parts, and guiding execution.

A coder translates directions into code: the raw material that compilers and machines transform into deliverables.

AI can be a competent coder—sometimes better than average. But it’s not a programmer. It doesn’t do analysis, design, or leadership. And when people forget that, they get into trouble

A Lesson from the 1980s

This divide isn’t new. Back in the late 1980s, there was a war between developers who wrote everything in macro-assembler and those moving to C.

Assembler was powerful, efficient, and elegant—but slow to produce. With C, you could build a working system in hours instead of days.

Assembler advocates argued their code was smaller and faster. True, but in practice? Compilers were “good enough,” and C developers could deliver real products faster. The compiler did the assembly for us.

AI today is playing the compiler. Tools like Claude act as compilers or transpilers: they take structured instructions and translate them into code. They’re imperfect, but guided by a skilled programmer, they’re extremely productive.

The Real Risk

Here’s the danger: if you don’t know what good code looks like, AI will happily hand you bad code. You won’t even know it’s wrong until it breaks.

I’ve run experiments giving AI only vague, user-level instructions. The results? Usually poor, sometimes catastrophic. Without proper direction, AI produces junk.

It’s no different than trying to use a CNC machine (a sophisticated programming machine) without knowing mechanical engineering. More powerful tools amplify skill—but they also amplify ignorance.

So yes, people who blindly rely on AI are going to be eaten alive. And people like me, who know how to guide it, will make a good living fixing their messes

How to Use AI Well

• Act like a senior developer. Even if you’re new, you need to define work clearly, divide it up, and review results.
• Treat AI like a junior coder. Direct it precisely, review its output, and correct mistakes.
• Don’t delegate what you don’t understand. If you can’t tell when code is bad, learn more before turning AI loose.
• Remember the rule of thumb: if you need AI, you probably shouldn’t use it. If you don’t need it, you can use it to great benefit.

A Tool, Not a Crutch

AI won’t eliminate programmers. It will eliminate rote coding. The winners will be those who can lead, analyze, and architect—while treating AI as the world’s most tireless (if sometimes clueless) assistant.

The future belongs not to those who let AI code for them blindly, but to those who can guide and command it.

Keep the Light,
Otter
Brian Jones