on AI tools and leaving room for the artist
the emergence of software that does too much and tools that help you create versus tools that create for you.
Creative tools have historically been tools of augmentation. They had to be – the technology couldn’t do the creative part for you. Photoshop made retouching faster; it didn’t compose the image. Pro Tools made editing multitrack recordings easier; it didn’t write the song. GarageBand gave a kid access to instruments and loops, but the kid still had to arrange them into something. The tool handled some of the tedious parts and left the interesting decisions to the human.
AI changes this. For the first time, the tool can do the creative work – generate the image, write the song, produce the code. Which raises a question that didn’t exist before: just because it can, should it? And if it does, what happens to the product, the outputs, and the company’s ability to sustain?
A Suno employee’s tweet went viral this week, and not in the way she intended. She shared her story: growing up singing, writing songs in her diary, dreaming of being a musician but unable to afford instruments or lessons. “A dream I had became just a memory, until now,” she wrote. “I am beyond proud and honored to get to work at a company that is enabling music creation for everyone.” The responses were brutal, and I’m not here to pile on. But it got me thinking about why people in both art and tech responded so negatively.
Part of it was the premise: the idea that music creation was inaccessible in 2006 didn’t land for most people. GarageBand shipped in 2004. Bedroom producers have been a thing for two decades. Billie Eilish made her debut album with her brother, a laptop, and a $99 AT2020 mic.
But the deeper issue wasn’t historical accuracy. It was the gap between how Suno frames itself and how people actually use it. The employee talked about enabling the 13-year-old who dreams of being a musician. The product, in its most common use case, takes a sentence or two of input and outputs a finished song. That’s not enabling someone to become a musician. It’s generating music for them. Those are very different things.
In the follow-up, the employee focused on professional use cases: stem separation, producers playing with different takes. That may be an emerging use case, but it’s not what most people think of when they think of Suno. Suno has built impressive models. But I suspect they’ll have to adjust their offering to sustain, and perhaps they already are, given the pivot toward professional workflows in that follow-up.
There’s a concept that Ricky Van Veen shared a decade ago, via Andy Weissman, that’s stuck with me: the best tools make you feel like you’re someone or something else. Instagram made amateur photographers feel like artists. GarageBand made curious kids feel like producers. The tool elevated what you were already doing and closed the gap between your ability and your taste. Suno, in its default mode, skips that entirely. You don’t feel like a musician when you type a prompt and receive a finished track. You feel like someone who ordered something. The creative decisions—melody, arrangement, production, the thousands of micro-choices that make a song yours – were made for you. The IKEA effect requires actual assembly. When the tool does 99% of the work, there’s nothing to attach to.
And you can hear it in the outputs. Suno songs have a certain sound: competent, polished, but ultimately interchangeable. They converge toward a mean. That’s fine for some use cases – background music for a video, a joke song for a friend’s birthday. But the things that move people, the songs a stranger might hear and feel something – those require the artist to actually be there, making choices, imprinting taste.
Compare this to Cursor, which is becoming the default for a lot of developers I know. Cursor doesn’t one-shot full applications. It accelerates the tedious parts: boilerplate, syntax lookups, the “I know what I want but need to remember how to express it” friction. The human is still the architect. The interesting decisions remain yours. People get better with Cursor. Their projects live in it and their judgment and taste still matter. You can’t typically look at a project and say “this was obviously made with Cursor,” and that’s a really important heuristic for a tool’s long-term sustainability.
And that’s the distinction I keep coming back to: tools that do the work for you versus tools that make you better at the work. Replacement tools are novel, shareable, fun for a minute – then they flame out. Augmentation tools compound. The line here won’t always be clean, and our sense of what counts as “you made it” will shift as these tools get normalized. But right now, the distinction matters – especially for the companies trying to build durable businesses around them.
We’ve seen this before. Lensa, the AI avatar app, made $70 million in November 2022. Everyone was posting those ethereal AI portraits. It was inescapable for three weeks. Then it was over. That $70 million was the majority of the money the app ever made. Upload selfies, receive polished outputs. No skill expression, no room for taste, no reason to return once you’d posted your favorite. The output was the product, and once you had the output, you were done. That’s not a business – it’s capturing maximal value in a viral moment. And $70 million is nothing to sneeze at, but the next year they made a quarter of that for the entire year.
I suspect we’ll keep seeing novelty apps from solo developers or small teams that go viral and make real money. But they’re not venture-scale businesses. The tools that last will be the ones that leave room for the human to matter. Not because of some romantic notion about creativity, but because that’s where the ongoing value lives. Augmentation creates retention: you get better with the tool, your work lives in it, your workflow grows reliant on it and switching costs accumulate.
I’ve made music for a long time. Enough to know the difference between the satisfaction of finishing something you labored over and the feeling of receiving something handed to you. The hundreds of orphaned demos in my Logic folder aren’t failures – they’re the residue of actually trying to make something, of closing the gap between what I could hear in my head and what I could get out of my speakers. Some of them got there. Most didn’t. But the ones that did feel like mine in a way that a generated output never could.
AI tools are going to keep getting better at producing creative work. That’s not a threat to tools that augment human creativity; it makes them more valuable. When anyone can generate a competent song or image or piece of code, differentiation shifts to taste, judgment, and the skill of wielding these tools well. The products that understand this will build for professionals and motivated amateurs: people who want to be better at something they already care about. They’ll charge real money, $100-200+ a month at pro tiers, because they’ll be embedded in workflows that create ongoing value. They’ll have higher friction to start and grow more slowly at first, but they’ll last because their users have a reason to stay.
The products that do everything for you will keep having Lensa moments: viral spikes, breathless coverage, impressive revenue for a month. Then the novelty fades, the outputs all look the same, and everyone moves on. Art can serve lots of purposes. A caricature from a carnival is fun. An AI-generated song for a friend’s birthday is fine. But the things that make strangers feel something still require someone to actually make them. The best AI tools will help more people do that. The rest are party tricks.
Great perspective, Charlie!
no credit for the wait what show photo??! what about the creator?!