New creative tools and taste

Agency and taste are buzzwords these days, but their popularity hints at how the creative process is changing across many domains.

As a quick definition, "taste" here means the degree to which someone appreciates the quality and quantity of details that make up excellent work.

Agency and taste matter for creative work because without them, you get what's now popularly known as slop.

Sci-fi writer Ted Chiang recently explained why this is. In an essay in the New Yorker, he noted that when AI models are prompted to write a story or generate an image, they reflect the average choices that a skilled writer or artist would make to deliver on that prompt.

Art, Chiang writes, is the sum of innumerable decisions made during the creative process. It might look something like this:

These decision points amount to moments of expression of the creator's agency and taste. High-agency creators with taste make unique and surprising choices, which often can result in the best art.

But by outsourcing these decision points to AI models, Chiang argues, one creates slop by auto-applying the default or median decision across the board.

While AI tools can certainly make for slop, I would argue that the best creative tools will enhance creative expression.

In a classic talk, Bret Victor noted that creators need direct relationships with the media they create. He distinguished between engineering and authorship:

Engineering is the act of manipulating an indirect representation of the final product, usually code or schematics.

Authorship is the act of manipulating a direct representation of the final product, such as composing a word document.

Authorship, Victor argues, is a superior mode of creation, because the creator can make decisions throughout the creative process with high-resolution perception of the consumer's ultimate experience. They're given a lot of empathy for what the end user will ultimately think and feel about the work.

The real promise of new creative tools, then, is not just to save time for creators. The promise is to give more creators high-fidelity representations of the end work throughout the creative process, from idea to distribution. This will allow creators to better express agency and taste.

In fact, this may explain why new AI tools have excelled in certain domains but not in others. AI writing tools, for example, have not exploded in popularity, as Lex founder Nathan Baschez recently noted. However, AI coding tools (Cursor, Lovable), AI video tools (Runway, Sora), and AI music tools (Suno) have taken off.

I suspect this is because word processors already offer a reasonably high-fidelity representation of end user experiences. Compare that to coding – that's why products like Cursor and Lovable feel like magic.