Against the age of vibes

We found an office that gets good sun. A little roomy for two people. I set up a standing desk from Coupang and brought in a spare monitor from home; two thick books stand in for a monitor riser. With a clacky mechanical keyboard and a trackpad in place, it turned into a fairly convincing workspace.

Every morning I pick up an iced coffee and head to the office — the one with a clear view of Tancheon across the street. I pull my laptop out of my backpack, plug it into the monitor, and open a terminal. These days I can do a passable impression of an engineer. I open tmux and summon Claude Code and Codex side by side. Even typing the most basic commands — mkdir, cd, ls — gives me a small sense of competence. I can even hold forth, plausibly enough, about how Pi is the agent framework to use, or how you really should be on the Anthropic SDK, or how the Codex app server is the better choice.

And that's not all. I've also been doing an impression of something between a designer and a frontend engineer. Fighting through terms I don't understand, I stitched together a patchwork design system and even published a Storybook. My god — I built a design system in code. My shoulders square up on their own.

You can become anything. You can do anything, too. None of it will be whole, but I can no longer deny that we've arrived at an era where you can imitate your way to something plausible — after all, I've become the proof myself. Maybe it really is, to some degree, a matter of time, money, effort, and will.

But there are still things that mere pretending can't reach. Or, to be honest: things I don't want to reach that way. From here on, it becomes a matter of belief, taste, and philosophy. In a world as noisy as this one, the things I want to avoid — the few things I clearly dislike — are these:

• ·Intoxication with, and blind faith in, AI tools themselves
• ·The belief that one click and sheer speed are all that matter
• ·Being satisfied with below-average results
• ·Mass-producing and sharing output with no deliberation and no detail
• ·AI-steered results that trick you into thinking they were your own idea

In the end these all amount to one claim: don't let the LLM do the steering through a few clicks and a few nods of agreement. Lately, looking at the probabilistic, average, inoffensive, opinion-less output LLMs produce, I find myself annoyed for no particular reason. What I keep trying to hold onto instead is this:

• ·Originality
• ·Deliberation and will
• ·Detail
• ·Care

It's a quietly sad thing that these words now carry a nuance of being at odds with using LLMs. In a world whose zeitgeist says everything gets done fast, in a single click, maybe I'm walking the other way. So be it. I can't chase a path I don't believe in just to soothe an anxiety whose purpose I don't even know — or more honestly, I wouldn't know how. In the end I can only go where my heart leads. That has always been the way that worked for me, and I can only hope it does this time too.