Taste Is How You Keep AI From Making Slop

Meng, a designer who has shipped five apps in two years using AI, started his demo with a direct look at the results. On the screen was a vibrant, interactive 3D world, procedurally generated by an AI that had worked for two hours straight. The entire scene, full of graphs and butterflies that react to the cursor, was less than one megabyte.

This, he explained, is what is possible. But he sees a huge problem in the current AI landscape. “A lot of people don't realize that the designs that they create is actually AI slop,” he said. AI can produce anything you can imagine, but it does not have taste until you give it clear directions. The reason is simple. AI is lazy. “And you have to tell it to work really hard,” Meng explained. Many of the new models have “hands and legs,” with capabilities far beyond what many builders are using. But they need a push. How to force a lazy model to produce exceptional work is the central question for applied AI, and Meng’s demo provides the only honest answer: with relentless, opinionated human direction.

What counts as AI slop?

Meng has a clear definition for AI slop. It is the sea of generic design that has become commonplace. “AI slop is like purple radiance,” he stated, “You see this everywhere.” He continued, pointing to the next wave of lazy outputs. “The new generation of AI slop which is like blinking lights everywhere. You know, I browse with huge lab spacing... it's horrible.”

The problem is that AI will always choose the lowest-hanging fruit unless you tell it to work hard. The core issue, he argued, is that most people do not realize they are producing it. They lack the developed taste, the “sixth sense” that comes from years of practice and exposure.

In the Q&A, one audience question asked whether today’s good design is destined to become tomorrow’s slop. Meng’s response was immediate. “That's a great question. I think... everything is going to become AI slop. And it makes sense because as humans, we're always chasing the next best things. We always want to do something else and something better.” This cycle is inevitable. The solution is not a static set of rules, but a continuous application of human taste to stay ahead of the curve.

Your Taste is the Differentiator

If all design eventually becomes slop, the only defense is a builder’s own judgment. During the Q&A, the discussion moved toward taste as the final few percent of the work that makes all the difference. It is the series of micro-decisions at the end of the process, a muscle you develop by seeing a lot of design, knowing the references, and building a vocabulary of keywords to direct the AI.

To get great work out of a model, Meng insists you have to give it strong references and make it work for hours. In his demo, he showed how the 3D world could be entirely regenerated with different themes. The scene shifted from a general theme to one filled with Sakura blossoms, and then again to one with Maple Leafs, all from simple text prompts. The goal is to force the agent to self-verify its work against a high bar. If you put in the effort, people will notice. They are looking for the human factor.

The Human Memory as a Prompt

The ultimate source for unique, non-generic work is not a website or a prompt library. It is your own life. Toward the end of his session, Meng showed another procedurally generated scene, this one of a 3D game board. He had his agent build it because of a personal connection.

“You as a human, you have to look into your memories and your sort of passion and you have to reference that. So, for example, who the hell does a game board, right? But that comes with me growing up with the game board and that gives me the purpose, that gives me happiness.”

He is saying that our most potent prompts are not on our screens. They are in our own histories. The work of creating something meaningful with AI is not just about technical skill. It is about the deeply personal work of translating your own memories into a specific and unique reference that an agent can build from. AI can do the manual work, but it cannot supply the purpose. That must come from the person directing it. It lets you do the thing you have always dreamed of doing, that you could not do before.

Pick one memory, one reference from your own history, and ask an agent to build something from that. It will not be slop.

---

SQ Collective hosts Coworking Fridays for founders, operators, and AI builders working through real product questions in Singapore.

Join an upcoming Coworking Friday: https://lu.ma/ai-labs

Explore SQ Collective: https://www.sq-collective.com

Michael