AI slop for design and dev

Vibe coding, which is a rising trend, pairs a human programmer with an AI like chatGPT. The AI is doing most of what a human used to do, and generating far more lines of code per hour than a person might. The problem is that often, no one knows exactly how the code works, which means it’s going to be difficult to fix when it breaks or needs an upgrade.

...

Get the system architecture right first. Document it, streamline it and test it. Then divide the components into small pieces and let AI finish the work. Fixing a defective brick is far more cost effective than re-architecting an entire building.

Product design is facing a similar issue. Throw strategic design questions into ChatGPT and you’ll get nothing but slop in return. But take a step back, architecture the system as a human, then turn to AI for help on the individual pieces.

Asking ChatGPT (or any of the equivalents) to do the work of designing a whole system is foolhardy. But diving into the work yourself, figuring out how it all fits together, then throwing individual pieces toward an LLM—that has the potential to bring new ideas to the project and speed up the whole thing.

In my numerous tests I’ve found that AI just cannot do what a good designer does on the whole. But in pieces it can think quicker and more broad than even the most seasoned veteran.

This is how I’ve been using it recently. Throwing specific, tactical, questions toward ChatGPT on my Mac, and then moving back into design while waiting for a response.

Via Seth Godin.