My favorite thing about AI-generated code

I would say my favorite aspect of using AI coding agents is what I learn by reading its code.

I was surprised to realize this. I am already an experienced software engineer. I have worked at the highest levels of technology, in Silicon Valley "unicorn" startups alongside some of the best software engineers on the planet. I am a published author, known worldwide as an authority on software engineering, and have worked successfully in a diverse range of engineering roles.

And yet, AI is regularly teaching me new things.

A few examples:

• Did you know Python has a 'secrets' module, that lets you generate a URL-safe random sequence of characters? I'm considered a world-class Python expert, and I had no idea.

• SQL has a really useful feature called CHECK constraints. I have been writing SQL for almost decades, and only learned about this when the coding agent told me last week.

• In a purchase-fulfillment system I developed this month, the coding agent's solution handled a race condition I had not even thought to consider.

I could not be more happy about this.

Even though, let's acknowledge, coding agents also routinely make droolingly stoopid errors.

Yes, even at this point in 2026, and using the most expensive LLM you can buy. There's a reason I read the code they generate, and do not blindly accept it. At least, not yet.

And mixed within that is some genuine brilliance. Which I think is due to a couple of factors.

The obvious one is that any competitive LLM has, in its training, seen more code than any single person. It has been exposed to more features, more implementations, and more use cases than even the most experienced human engineer ever could.

But there's another reason too:

When using a coding agent, I am more adventurous.

I am more willing to take on a task that is beyond my comfort zone, where I do not have experience doing before. I'm more aggressive at starting a novel task. And more open to suggestions for how to go about it, rather than leaning on the familiar but not-optimal way I might approach it, rather than exploring to find a better approach.

And in some ways, this is what excites me the most about the AI coding revolution. It's a new vector for ratcheting up my knowledge and expertise, and popping myself out of my own bubbles.

This is one example of the kind of synergy I'm talking about between engineering foundations and AI coding strategies, which is a key theme of Boosted Code.

Review the program here: