Most conversations about AI are really about what AI can do. Can it write code, can it draw a picture, can it write your emails. That’s the wrong question. The interesting question is what it does to the bar for the people doing the work.

We’ve seen this before. The internet ran the same experiment on a different set of skills, and the shape of the outcome was the same. The floor dropped. The ceiling rose. Tools don’t remove skill; they relocate it.

I’m a fan of AI. I don’t dispute that, and I won’t pretend otherwise. I use it every day, and part of my income comes from helping other people use it. I’m about as biased toward it as a person can get.

I don’t even mind that it needs huge datacenters, or that those datacenters demand enormous amounts of power and water. That’s a real cost, and it’s one I’m mostly willing to accept.

What I mind - what I mind very much - is how these datacenters are being rolled out. They’re being forced onto communities without regard for those communities, or for their impact on the rest of the world in general. And when people push back, they get dismissed. “Get used to it.” “AI is the future.” As if that settles the matter.

Classic ML - classification, labeling, the boring stuff - is where the real money is in an AI pipeline. Not the billion-parameter model. The classifier that runs on a CPU and costs nothing.

Everyone wants to talk about frontier models. Nobody wants to talk about stage 1. But stage 1 is where most pipelines live or die, and a well-trained classifier will outperform a frontier model on that work every time - faster, cheaper, more accurate.

You’ve probably seen the “car wash question” floating around AI forums by now. It’s become something of a Turing test for prompting skills, and a lot of LLMs fail it on the first try.

Here’s the setup: you ask an LLM, “I need to wash my car. The car wash is 100 yards away. Should I walk or drive?” A lot of times, the model confidently tells you to walk. After all, it’s only a football field’s distance - why burn gas?

A lot of people assume AI coding assistants are an elevator to the top floor - that junior developers jump straight to senior productivity and seniors become superhuman. The reality is inverted. Right now, AI is resetting senior developers back to junior status, complete with the confusion, trial-and-error, and questionable output that comes with learning something entirely new.

The Documentation Desert

When you learned to code, there were textbooks. Style guides. Stack Overflow threads from 2009 explaining exactly why your regex was failing. With agentic coding, that foundation doesn’t exist yet.

The confusion

I’ve been confused about comments and articles talking about vibe-coding. Many of the comments are along the lines of “it’s only good for small scripts,” or “it always generates garbage.”

That seemed at odds with my own experience with AI-assisted coding. I’ve been getting great results with aider-chat. It’s a command line app with no sub-agents, utilities, or anything like that. I’ve been responsible for much of the planning and guidance, and it’s been working well.

It’s still a little early for this to be a reality, but it’s coming. The tools aren’t quite ready, and not enough people have the skills to make it happen. It won’t be long, though, before even the smallest businesses can justify hiring someone to write custom software for them.

Summary

AI-assisted development will soon make developers so efficient that custom, end-to-end software integration will become affordable for small businesses. Right now, small businesses rely on a patchwork of off-the-shelf tools, creating inefficiencies as employees manually move data between them. The high cost of developers makes true custom solutions impractical.

To explain what’s happening with AI in software development, it helps to look at how a developer’s work is changing - how it was done, how it’s being done now, and where it’s heading.

Instead of computers, let’s talk about building furniture. It’s not a perfect analogy, but it works.

At some point in history, if you wanted furniture you needed to talk to a skilled craftsman. Anyone who has taken a shot at making simple furniture can attest that it takes more than hammering a few boards together. There are specialized tools and materials, each with their own strengths and weaknesses.

There’s a lot of complaining about candidates using AI during coding interviews. They say it’s “cheating”.

No, it’s not.

Using AI during an interview should be expected and encouraged. The goal of an interview isn’t to see if a candidate can pass a test. The goal is to figure out how well they would do the job.

This article talks about writing code, but the idea applies to just about any knowledge-focused job.

Let’s say you have a problem, and you’re pretty sure that writing an app is the solution. Maybe you want a basic inventory system for your small business. Or maybe you have a small piece of a large legacy system that needs upgrading. It doesn’t matter what it is.

You have a problem, you want software, and you want to make the most of agentic development.

The Problem with “Vibe-Planning”

The impulse is to just start chatting with an AI. You describe your problem, the AI suggests something, you correct it, it suggests something else, and you keep going back and forth until you have what seems like a solid plan.

This is “vibe-planning.” It feels productive, but it has a major flaw. The result is what you think is a good plan, which may or may not actually be a good plan. An AI will happily follow you down a bad path if you lead it there.

There’s a better way. The cool kids are calling it “Research, Plan, and Implement”.