Aug 2025
In which I specifically discuss the output of AI tools, not the associated costs or ethical considerations, both of which are certainly well documented at this point.
AI features are metastasizing across most of the platforms we use to get things done, allowing more people to blast more stuff everywhere than ever before! Nobody asked for this and lots of people very much don’t want it. However! I would argue that creative tooling being commodified isn’t in itself bad. People had the same complaints about the invention of photography or desktop publishing. It’s cheating! It lets the wrong people make things! It lets people make the wrong things! More relevant to our current landscape: it squeezes out professionals and lowers the bar for what’s considered good.
It’s (largely!) the same story playing out again. I am in no way an AI apologist or booster – there are certainly problems with the advent of AI that didn’t exist for, say, cameras – no question, it's not apples to apples. But: we’re still close to the beginning of this particular story, not the end.
Just as with cameras, desktop publishing, home music studios etc., Sturgeon’s law still applies: 90% of everything is crap. AI has made it so, so easy to make things that we’re up to our necks in undifferentiated ‘content’. We can point at, say, music from years past: it feels better, the craft is more evident, et cetera – and that’s because we’ve thrown most of it in the trash and left only the good stuff. AI-derived works haven’t yet undergone any degree of winnowing. Our ability to curate is (currently!) overwhelmed by our ability to create.
Algorithmic or AI-driven curation isn’t the solution. AI can’t have taste because taste requires values and intent, not pattern-matching. LLMs fundamentally can’t understand why something is good. I’m not sure it’s possible to determine what’s good right now anyway: we’re too close to the rockface to know what has staying power. 90% of everything being crap is simply a function of people making things, not the methodologies or technologies.
Functionally, there’s a more nuanced UX stance than ‘collapse everything into an open text field’. A desirable end state is (imo) agentic intelligence within existing workflows – sitting on top of application layers and making everyday functionality smarter without replacing established UIs. You can’t replace Photoshop with a prompt, but you can make each of Photoshop’s features more intelligent or capable in various ways.
‘Augment professionals’ is the killer application of AI, not ‘replace professionals’. We can’t (and shouldn’t!) try to stop anyone from creating, but the fundamental expertise of a seasoned dev or designer remains very relevant. Adding a powerful camera to everyone’s phone didn’t create a generation of incredible photographers. Vibe coding enables non-coders to build, but not to build well.
We’re drowning in an overabundance of output right now, but we’re also discovering how AI unlocks the tools we thought we already knew. In the slop, we’ll find what’s worth keeping.