On AI

The waves I have seen

I got onto the web in 1993, when a page was a text file and a form was an event. I was there for the first browsers, the first content management systems, the first mobile sites, the cloud and the app stores. Each wave came with the same two crowds: the ones who said it would change everything, and the ones who said it was a toy.

Both were usually wrong in the same way. The tools changed. What good work looked like did not. This wave is the first one where I think the stakes on both sides are real.

The threat and the potential

AI is a very divisive topic, and I think both sides have a point. It can truly be a source of so much good: earlier health diagnoses, better climate modelling, software for people who could never have paid to have it built. It also comes at a cost. Sustaining it takes a lot of power and a lot of cooling, and so much of it is being driven by profit.

Profit, left to itself, leads to the frivolous. Giving people a fun, easy way to make memes and silly songs so they stay engaged and captive to advertising is a waste of those resources. I believe AI is both an existential threat to humanity and a path to humanity reaching its full potential. The part I find pretty cool is that we get to pick the threat or the potential, simply by choosing how we use it and how we interact with it.

What AI changes

The cost of building a careful piece of software has fallen through the floor. Not the cost of deciding what to build, or of knowing whether it is right. The cost of typing it out, trying it, fixing it and trying again.

That matters because most useful software never got built. It was not worth an agency quote or a venture round. A small choir, a regional events board, a municipality with one person on the website: these were never going to be anyone's product.

What it does not change

Judgment. Taste. Accountability. An AI coding agent will confidently write the wrong thing, and it will write it beautifully. Someone still has to know what good looks like, read what came back, test it and sign their name to it.

It also does not change the duties around data. Where it goes, who can see it, what is kept. I treat those as my problem, not the tool's.

Useful versus frivolous

AI should mostly be used for useful things, not just the frivolous.

A thing is useful when it solves a problem someone actually has, for people who would otherwise go without. It is frivolous when its main purpose is to be seen using AI, or to keep someone scrolling. I say mostly because I am not against fun. A silly song for a friend's birthday is fine. A business built on keeping millions of people making them, given what each one costs to run, is not. Most people can tell the difference within a minute of using something.

Why thirty-five years matter more now, not less

When the typing is cheap, the deciding is what you are paying for. Knowing how a business actually runs, what a user will and will not put up with, where a schema will hurt in three years, and when to say no: none of that comes from the model. It comes from having built it, taught it and fixed it for a long time.

The evidence

Two things I have built this way so far, working with AI coding agents, alongside a day job. I designed them, reviewed every line, tested them and shipped them.

How I work with it

I write the requirements with AI first, with IDs, and the tests that prove them. The agent builds against those. I read the diff like I would read a junior developer's, except faster and with less patience. Anything that touches data, money or accessibility gets a second look by hand. The commit carries the requirement it serves, so a year from now I can tell why a line exists.

That is the whole method. It is not clever. It is the same discipline I taught at Centennial College for ten years, with a much faster pair of hands.