Asked how agents use a design system differently from people, Kamminga says humans slowly become fluent in a company's components and conventions, while an agent starts from scratch every session. Unless the system is one with lots of public training data, it has to understand it end to end and use it correctly the first time.
Yep.
…I would love for you to kind of riff on that for a second. So when you think about the difference between a human going into like, you know, mydesignsystem.com and browsing around versus an agent doing that, like what is the fundamental difference in action and how they use that structured data?
…correctly immediately the first time so that's where i think the fundamental challenge is there's no that there's no incremental learning there's no Uh, self, you know, there's no self-learning and remembering in the model, and that's what we still don't forget it anymore. Everybody knows it, but it's still hard to remember it every single time when you start a process that this is the reality. This is the new guy on the job, is first day, never done it, never seen it, now do a perfect job. That's what we're asking. If I want to sort of humanize that, right? So, yeah, so I think that is what I see as a challenge: that our users don't learn about our tools, the new experience every time. So, they are the first-time customer every single
time. Yeah, but what about the idea of context, right? And I hate to throw this like as a whataboutism there, right? But, like, isn't this the idea of an agent brain or agent memory? Is the idea of how do you serve up the right information to that agent to make it so that that new guy has a better…
Yeah, that's very interesting. And so, if you if you ask me that, right, D, um, so, so, what, what that agent really does is, you know, it does have knowledge and experience, all his past knowledge and experience up to the training moment, right? Right. Whatever that is. So, that's where I think…
uh, Tailwind is very easy to write in React with uh most LLM tools is because it's a lot of the data,