The second product

For most of the last year, LuvAI was one product. PromptCraft — a working library of AI prompts, plus a handful of free generators — was the whole studio. When people asked what LuvAI was, I pointed at PromptCraft and said "that."

This month there are two. The second product is a Taiwan stock-analysis site. It has a name now — 股譯 GUYI — and a tagline I'm fonder of than I expected to be: 台股，翻成你看得懂的話. Taiwanese stocks, translated into language you can actually understand.

A prompt library and a stock screener are, on their face, unrelated products. One helps you talk to image models; the other reads the Taiwan market. Friends who've watched the studio for a while have, gently, asked the obvious question: isn't this a loss of focus? Wasn't the whole point of a one-person studio to do a small number of things well? Building a second product in a completely different domain looks, from outside, like exactly the kind of sprawl I spent an entire essay arguing against.

I want to defend the decision, because I think the thread connecting the two is real and not a story I told myself afterward.

The thread is the word *translate*. PromptCraft, underneath the library and the generators, does one thing: it takes what a person wants — a mood, a scene, a song — and translates it into the specific grammar a specific model needs to produce it. The value isn't the model. The model is free and available to everyone. The value is the translation layer: the accumulated knowledge of how to phrase the thing so the machine does what you actually meant.

The stock tool makes the same move in a different domain. The Taiwan market produces an enormous amount of raw data every day — prices, three-major-investor flows, margin balances, financial statements, disclosures. All of it is public. The official sources publish it for free. And almost none of it is *legible* to an ordinary person. A retail investor opens a quote screen, sees forty numbers, and has no idea which three matter, or what they imply, or whether the stock is expensive or being quietly accumulated by the people who tend to know first. The data is already there. The translation is what's missing.

So the second product is, structurally, the same product. Public raw material that's freely available but illegible, plus a translation layer that turns it into something a person can act on. Prompts in one case, market data in the other. Once I saw it that way, the second product stopped feeling like a loss of focus and started feeling like the studio's actual job, pointed at a second domain.

The honest part — what's AI and what isn't.

I should be precise about this, because the AI-tools space is full of products that wave the letters "AI" around to borrow credibility they haven't earned, and I've written before about how much I distrust that. 股譯 is not, for the most part, an AI product. The bulk of it is unglamorous: official market data ingested on a schedule, technical indicators computed with the same arithmetic chartists have used for half a century, a scoring model, a screener with fifty-odd conditions, a few calculators. None of that is AI. It is data plumbing and ordinary math.

There is an AI layer, and it's the part I'm proudest of, but it sits *on top* of the data, not in place of it. The headline feature is a set of twenty-seven investing masters — Buffett, Peter Lynch, Howard Marks, Cathie Wood, and so on, down through several Taiwanese investors most foreign readers won't know. Each one is distilled into a distinct analytical lens. Ask any of them about a stock and you get that master's actual point of view: the deep-value investor refuses to look at the chart, the trend-follower refuses to look at the financials, and they disagree with one another in the specific ways the real people disagreed. The hard part of building that was never the model call. The hard part was the distillation — reading enough of each investor's real writing to encode *where they diverge from each other*, because the divergence is the entire point. A panel of twenty-seven advisors who all say the same thing is worth nothing.

That's the rule the second product inherited from the first: the AI is a layer, not the substance, and it earns its place only by being honestly scoped. Every analysis on the site says, in plain language, that it is an information tool and not investment advice — because that's what it is. The translation is the product. The model is just one of the things doing the translating.

Why this can be one person's work and not two.

The reason a single person can run two products in different domains is the thing I keep returning to in these essays: the plumbing was expensive once, and now it's nearly free. The second product shares the first one's accounts, its credit wallet, its session system, its admin tooling, its hosting bill. A reader who signs up on one side is signed up on both, with one balance that spends in either place. The architecture that made adding a small subtitle tool cheap is the same architecture that made adding an entire stock site cheap — not free, the stock site was months of domain-specific work, but cheap in everything *except* the domain-specific part.

I don't yet know whether the second product will find its people. It's at the stage PromptCraft was at a year ago: built, polished, quietly online, mostly unvisited. If the essay I wrote about launching to crickets is right, that's precisely where it's supposed to be. The work from here is the same work as before — keep using it myself, keep fixing what bothers me, and wait for the slow arrival of the people who find it useful enough to stay.

Two products. One job. The studio translates things that are public but illegible into things a person can actually use. That's all it has ever done. It just does it in two domains now instead of one.