I know, that's just what the world needs: another old white guy's opinion on AI and software development. But what the heck - it's my blog space here after all and I simply can't resist adding my 2 cents to the topic 😆.

The "AI avalanche" is omnipresent these days and impacting all areas of personal and work life. I'll focus on the impact and disruption in the software development domain as this is where I work and know most about, but other areas will be impacted equally soon enough (or already are).

I'm pretty sure everybody in our industry has heard the shot by now and people are adapting AI-driven software development - some more, some less willingly and excited.

However, a lot of discussion and mostly doubt I've heard recently is about code quality. There is still a lot of sentiment and anxiety about the actual code produced by LLMs and I think the discussion could not be more out of place and mistargeted!

To have healthy and productive discussions about AI-generated software, we have to erase a huge part of our memory and drop ideas that we considered essential when evaluating "code quality".

The following "metrics" have little to no relevance anymore!

👎 Readability

👎 Maintainability

👎 "Elegance"

Before you get all worked up over this statement now and declare me mental, let me elaborate. I'm specifically talking about code-level or change-set (pull request) levels of (perceived) code quality.

With the speed at which code can already be produced and rewritten today, it's almost madness to focus on line-of-code details, like variable naming or even design patterns.

We have to think about code in a larger context from now on. The actual lines of code will mostly turn into a black box and you should not care about its contents!

Code details will soon be considered a black box, but that's OK.

In the same way that the vast majority of us don't care about the actual machine code produced by our compilers, we are only taking this to the next level now.

The actual instructions and lines of code become irrelevant! Yes, we still want humans in the loop, reviewing and evaluating AI-generated software, but our focus must shift. We'll focus on how big parts of software interact and we will make sure this interaction happens smoothly and correctly.

In a sense, we are finally experiencing the true advent of component-oriented software development. Only, these components will be far bigger than we imagined when that term was coined decades ago.

In order for components and systems to work reliably with each other, the (very near!) future of software development has to fully focus on these metrics from now on.

👍 Interface Design

👍 Test Design and Testability

👍 Documentation

So, I can only encourage you to throw out some of the beliefs held when it comes to software quality and the software development process. It's happening with or without our consent. The avalanche has started ... and does not care about your opinions on design patterns or variable naming conventions 😜.