Let’s Stop the Slop

When LLM based design and code tools first showed up, we were all wowed. Look at the potential, look at how fast it can work! But as I and many other people have used and tested them, we’ve kind of arrived at them same spots independently.

These tools are indeed fast,a nd quite capable - but without care and intentional input, and ideally human tuning, you can end up with something that looks and sounds just like someone elses site.

Your copy might be written in the same voice, use the same patterns, or maybe your color scheme or even generated logo might look exactly the same.

When you give AI the reigns, you are locking yourself in to being average.

Which for code, is arguably not necessarily the worst thing ever. But for design, you can land on some questionable and jarringly samey decisions without even realizing.

It all looks the same

One of the things I noticed right off the bat is that there is tremendous variety in the outputs you can get. This gave me a false sense of security, thinking that that variety went deeper and was providing me with somewhat unique direction.

This couldn’t have been further from the truth.

Not only does the copy quite often sound exactly like someone elses, the core design of the site can be near identical as well.

I let Claude do the heavy lifting on a low stakes site and ended up finding another site that look almost identical, bar for bar, design decision for design decision, branding detail for branding detail. The similarity to the initial mockup I had made in Claude Design was uncanny.

How it happens

This was a sharp and revealing discovery that realistically I should have understood was likely all along. This is a statistical large language model, obviously it is going to give similar results for similar inputs. Duh.

But in the excitement of speed and novelty, I had been blinded to the slop I was getting ready to ship.

This is really the core of AI slop that makes it so viscerally unappealing. Everyone using minimal effort can get the exact same results, which is honestly by design. The system does the average unless instructed otherwise.

How to stop it

Once you understand the nature of the system and stop outsourcing decisions, you can achieve much better results. These tools are force multipliers - if you do not provide sufficient force on the system though, you will always get a mean response. Slop.

By moving in smaller steps, providing richer and more intentional input and feedback, and by actually getting your hands in the code or design files, you can measurable move that needle.

Sure it won’t be quite as fast, and maybe it won’t be quite as hand made, but the tradeoff is much more marketable.

Don’t be so damn lazy

Really, how much harder is it to take 30 minutes to make a logo, compared to spending an hour and dozens of prompts to generate a handful of mediocre logos?

Do you really need to tell your AI tool of choice to “Make this copy sound more like me” or can you … just type like you?

Let’s be better

Many of us that have been around for a little while remember learning in depth, for the sheer joy of it. Immersed in the skills. Design challenges, little studies, side tangents, creative ideas and trying to make them real.

No tie to a rapid and fruitful outcome, no hustle faster imperative. Just a desire to make stuff, and to figure things out for ourselves.

Don’t let AI rob you of that and make you revert to the mean it resides at itself.

Why this site

I and this site are not actually anti-AI, instead I want to help real designers and developers understand how to use these tools to their advantage where possible.

The speed of iteration and concepting is so much faster now, that you can explore random ideas that were never practical before. Even if you throw them out afterwards.

These tools have the potential to really magnify your taste and skill, but it’s like driving a race car with the traction control off. Sure it is faster, but if you don’t know what you are doing you are going to wreck.