when i started this project earlier this week, the goal wasn’t really to make a new website. it was to figure out how we, at vanilla, could use claude code and figma’s mcp to help the pm and design teams prototype locally and publish high-fidelity prototypes for feedback earlier in the process.

i started by watching a few videos that showed the workflow in action:

videos

• from design to code with claude code in 40 minutes (meaghan choi on peter yang’s channel)
• building your ideas with claude code and figma mcp (dive club)

anthropic has the kind of product building culture i admire.

before diving in, i installed claude code on my own machine and opened a few old studio and personal projects, some using flat-file cms or other outdated setups. it was pretty wild watching claude code scan through everything, explain the stack, the purpose, and even how to get each one running again.

from there, i wrote a short prompt about what i was trying to do: build a personal site that worked as both a portfolio and a digital garden, and use it as a way to teach myself what i’ve missed from the last decade of web development. i originally thought about using jekyll and github pages, but claude convinced me that just diving into the deep end was the better path.

while it took me a while to get claude code running and to find my way around vscode, i eventually got there, set up an astro project, built the repo, and connected figma to claude via mcp. i was a little disappointed to see figma mcp still relies on screenshots (though the output is much better than before).

in my claude.md file, i included not just technical instructions but a reminder that this project was meant to help me learn. plan mode ended up being the most useful part—it breaks everything down step by step and explains what’s actually happening.

one thing that surprised me was how easy it is to go overboard. early on i got lost in filtering and categorization options before realizing i just wanted something that felt simple, alive, and mine.

james ayres, our design engineer, had already been experimenting with using our own codebase and design system to make prototypes. this week, cree (one of our eng directors), james, and i are starting to roll out that same process with the rest of the team.

speculative artifacts: creating the objects that could have existed in the life i didn’t live.

korean adoptee experience: you can learn your birth name but never know what it felt like to hear it.

k-pop ↔ hardcore: diy vs. corporate. both build alternate realities for belonging.

design leadership: questions around team formation, critique culture, and how to balance craft with strategy. what does it mean to build a creative environment that protects clarity without killing experimentation?

design education: what’s broken in how we teach design, and what still works?

to explore

• how language learning reshapes identity, especially when it’s the language you were meant to know
• what it means to teach while still learning

in taking over product at vanilla, i’ve been trying to keep up with the pace of change, not just in design but in how we actually build things. the tools, the process, the way small teams work. companies like linear keep showing what’s possible when design, engineering, and product are all in sync and that’s the kind of team i want us to be.

as i’ve been thinking about how to get there, ai has become a critical part of it. it’s the thing making that kind of speed and craft possible. and i’m not just thinking about it for work, i’m seeing it firsthand while teaching web and code to students at calarts. the flattening of responsibilities is real, and to me, it’s exciting. it means designers, engineers, and pm’s can all get closer to the actual making. it’s a chance for people to lean into their specialties instead of hiding behind titles.

one of the bigger changes i’ve been pushing is getting both pm’s and designers closer to code through claude. this week i decided it was time to dive in myself. after watching a demo of how anthropic’s design team uses figma’s mcp and claude code to prototype directly in code, i finally sat down and built the portfolio site i’ve been wanting to make for years (literally took a figma project i started on a flight 3 years ago).

i’ve always been obsessed with pkms. i was an early obsidian user, tried roam, even started building my own zettelkasten after reading how to take smart notes by sönke ahrens. the hardest part for me has always been balance. i always get lost in building the system instead of capturing the idea. in a way, platforms like tumblr, are.na, and even x really work for me as the idea was the point not the system around it.

so it’s been a dream for a while to have a space of my own like that again (rip live journal). a place to post loose ideas, more polished writing, or just things i don’t want to forget.

when i teach the beginning web class, we start with a short history of the early internet: geocities, angelfire, myspace, that era when people made weird little websites just for the fun of it. in a lot of ways, this feels like a return to that. as i’ve weened myself off social media, i still have the urge to share, just without the noise or performative aspect of it.

learning to build this site through claude code literally felt magical. like the first time i learned to code, but lighter. it’s familiar but obviously much easier. i don’t feel limited by what i don’t know but just in what i could articulate myself.

this is all really exciting for me, i hope i can bring others along.

Turnstile, Seattle, WA, 10/07/25

Seoul, South Korea, April 2023

Seoul, South Korea, August 2022

these photos came from a workshop i took by janna ireland through the southland institute.

at the time, i was living in downtown los angeles and spending most of my free time just walking around looking for interesting visual / graphic patterns out in the wild.

i’ve started adding photos from other places that carry the same feeling.

The footage is grainy and bulbously distorted, fisheye lens from a skate video. Posted eleven years ago: “Masshysteri plays live in Mullsjö at Dennis Lyxzén’s birthday!” In what looks like Dennis Lyxzén’s backyard, the singer for Refused, surrounded by trees, a small back house, one of those helicopter clotheslines. The band furiously knocks out two songs to a small crowd. The audio is crisp, brighter than the studio recording on Spotify. I find myself transported to this backyard, watching a Swedish garage punk band that barely played outside their home country. I’ve never been to Sweden. But I’ve been to shows like this many times.

I came of age in the post-”punk is dead” era, in the suburbs of D.C. I discovered hardcore through chance: a Vans skatepark grand opening, an Epitaph comp with Madball and Agnostic Front. In true teen angst spirit, I globbed onto early 2000s mall punk, convincing my mom to let me dye my hair, spending my allowance on band t-shirts from Interpunk.com. During high school I kept a LiveJournal, writing sub-posts to crushes and finding others who listened to the same music. I learned HTML and Photoshop customizing my profile. I frequented Pheer.com, a site maintained by the semi-mysterious D.C. legend Pablo. Black and white, minimal formatting, all HTML tables and non-breaking spaces. Show listings, messageboard, video archives. The messageboard read like a digital high school cafeteria: cliques, posturing, occasional bullying. Still an improvement over an actual cafeteria. Pablo built exactly what the scene needed and nothing more. No growth ambitions. The gatekeeping was the feature.

Then, overnight, everyone migrated to MySpace. Your top 8 friends, featured prominently. Songs pinned to your page, the original SoundCloud. Secret house show locations disseminated via DM. MySpace made the scene’s social dynamics visible and portable. It scaled what Pheer did by hand. In undergrad, we moved to Facebook. We waited eagerly for our small art school to be added during the university rollout. All of MySpace’s charm wiped out for convenience. Customization traded for uniformity. Community traded for network. I still go to shows, but with every passing year, I feel less responsibility to participate digitally, other than to amplify what my friends are doing on Instagram.

I’m not suggesting subculture is dead. If anything, it’s the opposite: white nationalism, once a persistent problem in the scenes I participated in, is having a resurgence thanks specifically to YouTube. In college, I saw Agnostic Front in Fredericksburg, Virginia, a town that still believes the South will rise again. In the bathroom of a pool hall turned venue, a white nationalist skinhead was actively recruiting kids. When word broke, he was confronted and physically removed. Complete with flying chair. The scene had an immune system. Someone doing something wrong got dealt with, immediately, by people who were there. What’s YouTube’s immune system? Algorithmic moderation that can’t tell the difference between a basement show and a recruitment video. A platform optimizing for watch time has no incentive to distinguish between community and audience.

I’m watching Dennis Lyxzén’s backyard from my apartment. I can leave a comment, “I was at that show,” except I wasn’t. The video transports me somewhere I’ve never been, to a moment I wasn’t part of, in a scene I have no stake in. It feels like community. It isn’t. Every platform promises to bring people together. Each one trades away the thing that made the last one work.

---

Written for my application to SFPC’s Critical Writing Workshop, a week-long program on essays, reviews, and personal writing about art, technology, and society.

Teaching at Otis, we took the MFA Design students on a trip to Mexico City. We visited the Central University City Campus of UNAM, Casa Barragán, participated in studio visits, and explored the city.