AI Music, Human Taste, and the Future of SEO (Featured On Smol Der)

Key Takeaways

• All creativity is iteration. Whether you are a human or an AI model, making something good requires studying what came before and putting a new spin on it. The difference is humans still have an edge on true originality.
• Human taste is the differentiator. AI generates. Humans curate. The ability to look at output and say "this sucks" or "this is brilliant" remains uniquely human and CRITICAL to success.
• Vibe coding is real. You can build functional web applications without being a traditional software engineer. But you need a solid stack and the discipline to review what the AI produces.
• SEO strategy is shifting. Traditional content still works, but you have to go far beyond what a default AI prompt produces. Interactive tools and experiences will matter more than blog posts alone.
• Listen to your customers. The best business ideas often come from what people ask for, not what you initially planned to build.

Iteration Is the Root of All Creativity

One of the philosophical points I kept coming back to in the conversation: iteration is at the core of all music. And not just digital music. All creativity.

Could a person in a vacuum, exposed to no music whatsoever, come up with something decent? Maybe. But that is not how creative work actually happens. A human writes music by learning from people who wrote great music before them. By studying songs that vibe with their soul. By mimicking what resonates and then doing a new spin on it.

AI approaches this differently. Machine learning models are trained on massive datasets. They learn patterns and structures from millions of examples. The ethics of how that training data was collected is a legitimate debate. But at its core, both humans and machines learn by studying what exists and building on it.

The difference? Humans are still significantly better at originality. If you listen to what AI music platforms produce, yes, it sounds refined. Suno's latest model is very good. Depending on the genre, I would not be able to tell it was AI generated. And I listen to AI music every day. But that originality that comes from the human heart? AI still struggles to match it. We have an edge on making something that feels truly fresh.

Taste Is What Separates Winners from Losers

I think the core of everything in AI implementation is taste. Being able to look at the end result and say: does this suck?

Because if you are able to do that, you can go back to the drawing board and refine it. That is the real work. Usually you do not get it right on the first try. Or the second. The people who succeed with AI are the ones who can recognize when something is not good enough and keep iterating until it is.

This applies across every domain. Content. Music. Code. Video. The raw output from AI tools is becoming easier to spot. The value comes from having humans with domain expertise review and refine what AI produces. A graphic designer spots image flaws that a programmer might miss. A writer catches narrative inconsistencies that escape a marketer's notice.

Despite marketing claims about eliminating team members, building systems around skilled humans who scrutinize AI outputs yields FAR superior results. The best implementations use AI to handle repetitive tasks while directing human attention to areas requiring judgment and creativity. If you want to learn more about this approach, check out my piece on why you should never use AI as your value proposition.

My Vibe Coding Stack

A big shift in my focus over the past couple years has been programming. And a lot of it is vibe coding. I am not a software engineer by trade. My dad was a programmer so I started with some basic coding knowledge. But the actual web applications I have built? Mostly using AI assistance.

Here is the stack I recommend and have written about on MuseMouth:

• Next.js + Vercel for the frontend and deployment
• Supabase for database and authentication
• Claude Code for vibe coding (terminal based AI is insane)
• Modal for running GPU workloads and machine learning functions

I think people who use Claude Code or those terminal based AIs are could rule the world in the next 10 years. The amount of stuff you can do with them is honestly insane. You can specify what you want in plain English and iterate quickly.

One critical piece of advice: do not let an AI build your entire codebase from scratch. Use a boilerplate. Find something that works and build on it. The AI struggles with state management and queuing logic. It gets distracted. You need to give it clear constraints and yell at it sometimes. I have definitely yelled my way into getting things functional with Claude. Not proud of it. But it works.

If you are interested in this approach, I wrote about how I vibe coded an MVP for ChangeLyric. And for the backend machine learning stuff, check out my piece on building GPU powered AI tools with Modal.

The Future of SEO and Content Strategy

When everyone has access to the same AI tools, how do you stand out with content? This is a question I think about constantly.

The answer is competitive advantage. You have to do the thing that nobody else can do. Or at least the thing that is way harder than what everybody else is doing. If you are just going to OpenAI's ChatGPT, typing "write me a blog post on this topic," not giving it any context, and pasting the result onto your site? You are doing what every other newbie is doing. That content will not rank.

What works? Building out context. I have a thousand line manifest that I give Claude for writing blog content. Rules about voice. Rules about structure. References to copy I have written so it knows how I sound. Technical SEO best practices baked in. Table of contents. Things Google cares about. The stuff that takes actual effort to set up.

And beyond written content, I think the future is tools. Interactive experiences. If the AI has to send people to your website to get what they want, that is defensible. A blog post can be summarized and served by ChatGPT without anyone clicking through. A complex calculator, quiz, or interactive tool? That requires the actual visit.

Listen to What People Are Asking For

The origin story of Music Made Pro is a lesson in customer listening. Originally I was offering general song production services. Then AI voices came out and I pivoted to helping people make songs with whatever voice they wanted.

Then somebody asked if I could change the lyrics to an existing song. The scene from Full Metal Jacket where they sing Happy Birthday. They wanted custom lyrics. I said no, that is not possible. Ignored it.

Fast forward a few months. I realized that was exactly what people wanted. I retooled the entire business around lyric changes. And that became the main revenue stream. The demand was there. I just was not listening at first.

Take your shot on your idea. Get customers. Then LISTEN to them. There will be indications that what you are offering is not quite right. Or that they would rather have something slightly different. That feedback is gold.

The Ethics Question

I do think about the ethical implications of AI music. A lot of people hate how the training data was curated. Basically stealing music and training models on it. I understand that criticism.

But the genie is out of the bottle. China is going to do it whether we do it or not. And the practical reality is that whoever makes money gets sued. Once I make enough money, I will probably get sued too. That is just how it works.

From my perspective, I think about the use cases. People changing lyrics for a friend with cancer. For a birthday. For a wedding. Making their team happy or cheering somebody up. Those feel legitimate to me. Commercial use is a different conversation. My copy everywhere says this is for personal use. Do not use it in advertising. Do not make money with it. Because if you do, you are definitely legally in the wrong.

The bigger question about compensating artists in an AI world? I do not have a good answer. In an ideal world, the people who create art should get paid for it. But AI might push the value of novel art toward zero simply because you cannot enforce ownership at scale. The rules exist mainly to protect big content generators, not individuals who want to send a nice card to their friends.

Security and Practical Concerns

If you are vibe coding web applications, watch out for security. I learned this lesson the hard way when I left a Supabase bucket public. Woke up one day to find a thousand videos from random people in India using my cloud storage for free. Two days after I pushed the code. How did they even find it? I still do not know.

AI has gotten better at warning you about security issues. Usually Claude will say "hey, you should not put an environment variable here" or similar. But it is not foolproof. Read up on security best practices. Do not just trust the AI to handle it for you.

The funny thing is, even big companies are not that secure. The amount of security violations at major tech companies is honestly surprising. But that is not an excuse to be sloppy with your own stuff. Especially if you are handling user data.

What Excites Me About the Future

I am excited to see content that people create with these tools. The filmmaking world is about to change dramatically. I think independent films that are actually good are coming. Cartoons will probably be first. The first successfully made and genuinely good AI cartoon is going to be a milestone moment.

For my own work, I am interested in self solving capabilities. How do you make AI that can launch its own GPU containers to improve itself? Not the superintelligent Skynet stuff. More like an AI that can discover "here is an open source library that does this thing well" and then trigger and launch that automatically. Connecting existing things that other people have made and orchestrating them together.

The vision is any person can have an idea and deploy it without knowing all the technical details underneath. They just need to know that certain libraries exist and do certain things well. Chain them together and the AI handles the implementation. That is what I think agentic AI should be heading toward.

Final Thoughts

If there is one thing I hope you take away from this conversation, it is that AI is a tool for enhancement, not replacement. Simply chaining tools together in automated sequences typically produces disapointing results. The magic happens when you use AI as one component in a larger creative or business process.

Taste matters. The ability to recognize what is good and what is not remains uniquely human. Build systems that leverage AI for the repetitive stuff while directing human judgment to where it counts. That is how you win in this environment.

And if you are building things with AI, whether music or code or content, remember: the first version will probably suck. Keep iterating. That is how humans have always made great things. AI just makes the iteration faster.