Dig for Gems or Polish Them

What do we really need in the AI era, and what are we missing? Conversations with an Ivy League PhD student and a media editor.

I think I'm a born marketer. What I'm good at and enjoy isn't making something from scratch. It's packaging and distributing something that already exists. When there's a good product, I think about how to get it in front of more people. When there's a good story, I figure out which channel to tell it on and in what tone. That's my real job.

But lately, the biggest bottleneck I feel is finding those raw gems. We all have limits as humans, so there's no unlimited supply of them. Say we're heading toward a new audience and we don't have any raw gems for it. That's a real bind.

In that situation, what we can do is go dig for the gems ourselves, or bring in people from that field.

"Raw gems" sounds a bit abstract, so to put it simply, I mean the stories or track record that make you unique and make people like you. If everyone shows similar skills in the AI era, the story or content that makes you stand out will probably be your biggest weapon and your biggest differentiator.

I think of AI as a pickaxe. But if I can't find a diamond mine, I can't put it to good use. First I find the mine, then I dig it out well with AI. Something like that.

And if I've already dug out every mine I have, I'll need to go find a new one. Just as 0 × 1,000 is still 0, you need better raw gems. These days, turning 1,000 into 5,000 matters, but getting from 0 to 10 first may be the bigger lever.

1. Human judgment matters more in the AI era

A US blog post I read recently described the AI era as a "crowded 7/10 world." A world overflowing with things that score 7 out of 10. Blog posts, marketing copy, design mockups, code snippets. Everywhere you look, "decent" work is pouring out. The problem is that the gap between decent and good has become far more important than it used to be.

It used to be that making a 7/10 was work in itself. Writing a plan and putting a design on it took time and money. Now that process has been compressed dramatically, and anyone can make a 7/10. The axis of competition naturally shifts toward the ability to make 8s, 9s and 10s. Or more precisely, toward the ability to filter out the 7s and pick out the real thing.

That blog called this "taste." In Korean we might call it 감, a kind of intuitive feel. It's the ability to tell things apart under uncertainty. It also made one interesting point: the scarce skill in the AI era is refusal rather than generation. The person who knows what not to do, more than what to do, wins. Either way, this seems to be about human feel, intuition and judgment. The problem is that it's still a mystery what good taste actually is and how you develop it. It's a bit like a lagging indicator.

2. A fatal weakness of LLMs

I asked a friend doing a PhD in the US how much AI is changing academia. The answer surprised me. At the PhD level, the curriculum has barely changed. Sure, the data cleanup that first-year research assistants used to do has been automated, but the core of deep research is still human territory.

They gave me a concrete example. There was a study analyzing changes in the S&P 500 listing criteria. If you ask GPT to summarize the related papers, it neatly organizes the published findings. But if you hand it the raw data and say "find a new pattern here," it can't do it at all. When you think about it, that makes sense. An LLM is basically a language model, so to use tokens efficiently it usually finds the easiest, fastest source it can cite and answers from that. Pulling insight out of raw data that nobody has interpreted yet isn't something LLMs are good at, and it isn't what they were built for in the first place.

So LLMs are excellent at recombining knowledge that already exists, but creating knowledge that doesn't exist yet is a problem. If a raw gem has already been dug up somewhere, AI can cut and package it well. But when nobody knows yet what the gem even is, that's still a human job.

3. What that "sparkle" really is

That friend used a particular expression. They said some people "sparkle" when they look at the same data. People who intuitively catch the story hiding behind the numbers. People who spot, in a single line of data, a connection that you wouldn't find after reading hundreds of papers.

This lines up exactly with "taste." At its core, taste is the ability to tell what matters from what doesn't under uncertainty. Reading a new trend in financial data and picking the story readers really want out of thousands of pieces of content are, in the end, the same ability showing up in different ways.

And this sense is also the hardest thing for AI to replace. AI recognizes patterns, but it can't break them. It's good at giving the statistically most plausible answer from existing data, but asking a new question that everyone else is missing seems like a human skill. I wonder if this is what being human really means.

4. Tech companies now have to compete on things other than tech

While I was chewing on all this, I got on a call with another friend who works in media. We started from a completely different angle but reached the same conclusion. AI is leveling everyone up, so tech companies now have to compete on things other than technology.

The performance gap between AI models is narrowing, API prices are dropping, and open source is catching up fast. Differentiating on pure technology is getting harder and harder. Naturally, tech companies are turning their attention to content and stories. News of OpenAI acquiring a media company and Anthropic running an ambassador program to build community fit the same pattern. And people like Sam Altman and Elon Musk have always been global storytellers who know how to grab attention.

Technology can be copied. Code gets released as open source, and model architectures get published in papers. But relationships with people, the trust built up in a community and the story a particular brand carries are, unfortunately, not easy to copy. Ironically, in this era of the most powerful technology we've ever had, the moat seems to come from things that aren't technology.

5. Rapper Paloalto on AI music

When I heard that story is starting to matter more than technology, one thing came to mind right away. I wrote about it in an earlier post too. The Korean rapper Paloalto reviewed AI music on his YouTube channel. AI music has become almost a genre of its own, and judged on sound quality alone, AI music is often better than music made by people.

But when we listen to music, we don't just listen to the quality of the sound. We also take in the background of how the song came to be, the artist's story and even their daily life. So a song doesn't compete on the track alone. Speaking as someone working in the industry, he explained that the content wrapped around a song may matter as much as the song itself.

6. Building is easy now. Distribution is the bottleneck

It's not just music. Building a software product with AI has gotten easy, but competition has gotten fiercer to match, so getting the word out has become much harder.

The time it takes to build an MVP has shrunk from months to days, but the bottleneck is still marketing and distribution. And the core of marketing and distribution is content. Stories that catch people's attention and build trust.

I still wish AI could help with this too. But like the metaphor I started with, AI can be the pickaxe, but it couldn't be the raw gem. Humans may need to focus hard on finding the gems. In the AI era, being able to draw a clear line between what humans should do and what AI should do seems like a very important skill.

7. Who survives the AI era, according to Wall Street: community and media

Back to my Ivy League friend. They study financial economics, so they shared the Wall Street view as well. Since last year, investors have been using the term "SaaS-pocalypse," borrowed from "apocalypse." The idea is that AI is shaking up the entire existing SaaS business.

Interestingly, the analysis is that not every SaaS company is at risk. Companies with network effects (Uber or Instacart, for example), where users are connected so that one person leaving affects everyone else, are seen as relatively safe. Companies that sell pure software features are a different story. Take Chegg, the middle and high school homework platform. When GPT came out two years ago, its investors pulled all their money and it collapsed right away. It was a shock to see a mega platform I used all the time in high school 15 years ago disappear overnight. Isolated, tool-type products are likely to be replaced by AI.

When I told my media friend about this, they immediately agreed and brought up the keyword "community." AI companies are putting serious resources into building touchpoints with people, just as much as into developing technology.

The reason is simple. Technology gets caught up to quickly, but community takes time. An ecosystem where thousands of users care about a platform, help each other and create content on their own can't be copied overnight. That's why tech companies acquire media and run ambassador programs.

This is an extension of the raw gem idea too. Within a community, things emerge naturally: cases users create themselves, connections that only work in a specific context, and fandom. AI can't inject these from the outside. They form only among people as they circulate and interact.

8. This era needs only two kinds of people: those who dig for gems and those who polish them

After those two calls, here's how I'd sum it up. The roles the AI era needs come down to two. People who dig for raw gems, and people who polish them.

The people who dig for gems have their own unique stories and narratives that others don't. A researcher who has studied one field for 10 years and has insights no one else has. An artist with stories that come from an unusual life. A practitioner who has worked in the field and built data by hand. What they have is something AI can't find no matter how much of the internet it searches.

The people who polish gems are the ones who bring those gems to the world. Marketers, editors, producers, community managers. I'm one of them. The core of the job is picking out the gems based on taste, then adding your own perspective and interpretation to create a new context.

I have to keep finding people who have raw gems. But as time goes on, the gems are running out. We're human and we have limits, so someday I'll have no choice but to grow into someone who digs for gems too.

Originally written in Korean on Brunch, April 9, 2026.