The ADHD Advantage

Honest AI talk with friends in big tech at our book club

I've been in a monthly book club for six or seven years now. The members are colleagues from my first job. We do read books, but the real goal is to stay in touch and keep up with each other's lives now that we've all moved to different companies. These days everyone is a middle manager at a major tech company, so it's become a rare place where we can talk more honestly than anywhere else about how they work in their environments and what's on their minds. Maybe it's because we've known each other so long, yet work at different companies, that we can be so candid.

They told me they'd shared my recent Claude Blue series on their internal company boards. I was really grateful. But at the same time, they all said it gave them massive AI FOMO, and that a lot of it hit home. This meeting got far more heated over AI than over the book: how each of us uses AI at work, and what we're worried about. A lot of honest stories came pouring out. To skip to the conclusion, I was shocked by how big the gap can be from company to company and person to person, even within the same tech industry.

1. The AI temperature at each company

What surprised me was that everyone works at a tech company you'd recognize by name, yet the AI temperature varies that much from one company to the next.

Company A has rolled out Claude company-wide and actively encourages people to use it. It's hooked into Slack, Jira, and Confluence, so basic use is already in place. Other AI tools are open too. The company is clearly pushing it.

Company B bans Claude and only allows Cursor. People know that connecting Claude to Slack is a whole new world, but the company has blocked it, so they can't even try it. Things are more complicated still in the content and creative teams, which push back on adopting AI at all. They're touchy even about translation, so even if you want to use AI, you have to tread carefully inside the organization.

The gap inside the company is extreme too. Someone on the next team over will very cautiously ask, "What AI do you use?" and it turns out they're only using plain ChatGPT. There's almost no internal AI training, so people have to learn on their own, which leaves a huge gap between employees in how they use AI. Especially at higher levels, quite a few people still do everything by hand, step by step. At a tech company. There are definitely people who use it well, but the better they are, the more absorbed they are in their own work, so they don't have time to teach others, and everyone has their own know-how that's hard to pass on. So the gap is widening rather than closing.

Company C is completely team by team. The CS team has adopted AI well and handles repetitive work efficiently, but the data team, maybe out of sheer stubbornness, doesn't use AI and has people do all the data analysis by hand. Their reason is that "the internal data is unstructured, so AI can't handle it," but I suspect it's really just the team lead being stubborn. I don't think that will last long.

If the temperature gap is this extreme inside the tech industry, I can only imagine what it's like outside it. And I suspect this gap will soon turn into a gap in companies' future competitiveness.

2. The biggest hurdle is time

The thing everyone said, more than anything else, was this: "I know AI is great, but to really use it I'd need three or four hours to dive deep, and I don't have that time."

My friend at Company A came up as a PO (product owner), and their calendar is packed with meetings from morning to evening. Their AI FOMO is huge, but they physically don't have time to look into it. Meanwhile, the CTO is piling on the pressure by posting every day in the tech leads' Slack channel about how many lines of code they wrote with Claude. When people on the ground say "I haven't had time to use it," the CTO pushes back hard, along the lines of "So you have time to go home and watch Netflix?" There are tons of internal AI lectures, but again nobody has time to attend them. It's a vicious cycle.

My friend at Company C tried installing an open-source AI tool, OpenClaw, on their personal computer. They installed it without wiping the machine first, got an alert that 72 pieces of personal information had been exposed, and deleted it right away. They didn't really have anything to use it for personally anyway, so they ended up dropping it.

In the end, people are missing out on all this because they lack a trigger and the time. But the difference between those who somehow made the time to try it and those who didn't was already enormous.

3. Walking into work feels reassuring now

The best example of someone who did try it was my friend at Company B. They were using Cursor almost like a personal assistant.

First, they organized all their work data into folders locally, then set up a schedule so Cursor reports only the things they need to do every morning. When a meeting ends, the notes get organized automatically, sent to each person responsible, and updated on the wiki. They said that alone cut their workload enormously.

Their operational follow-up work was impressive too. There was a task that involved entering data into Excel for each country, which people used to do entirely by hand. Now AI does all of it, and they told it to sort everything into three categories: things that are certain, estimates, and blanks for anything it doesn't know. Each one is color-coded, so you can see how reliable the data is at a glance. Recently they even dumped all their tax information in and said, "I have no idea, so tell me what I need to do first in my situation," and got it handled.

Something this friend said stuck with me: "When I get to work and Cursor is open in the window next to me, I feel so reassured." They said it's like having a trusted coworker sitting next to them all the time.

Earlier people said they had no time to use it, but this friend shows that once you take the time to set it up, it actually frees up time afterward. Even within the same tech industry, I could feel how huge the gap already is between people who use it like this and people who only use plain ChatGPT.

4. The 18-page PRD strikes back

More people getting good at AI doesn't necessarily make everything better. At Company A, junior staff have started churning out PRDs (product requirements documents) with AI, and the documents are getting huge. 18-page PRDs are pouring in. The problem is that it's exhausting for the senior people who have to review each one and catch the mistakes.

It's ironic: AI boosted productivity, and the cost of checking the work exploded. Of course they could use Claude for the reviews too, but as I mentioned, they haven't had the chance to dive deep yet, so they haven't gotten that far. We've entered an era where AI makes things and AI checks them, and the seniors stuck in the middle are suffering the most. This ties directly into a shift in what kind of talent companies need.

5. The rise of ADHD-type talent

Following the PRD story, what my friend at Company A brought up next was the most memorable part of the night. Their strength had always been going deep and getting fully absorbed. The type who digs into one thing all the way to the bottom. But these days, AI does that deep part better.

Instead, the juniors who stand out are the multitaskers who knock out several things quickly at once and switch contexts well. They throw all kinds of things at AI, judge the results fast, and move on to the next task. The people who run that cycle quickly are the ones who fit the AI era best. My friend at Company B using Cursor like an assistant across several tasks at once fits the same pattern.

This friend said context switching is really hard for them, so they're working on adapting. We're moving from the era of the specialist who digs deep to the era of the hyper-generalist who, with AI's help, can cover ground both deep and wide. Exaggerating a little, it felt like "ADHD-type talent" is becoming the ideal profile for the AI era.

I relate to this a lot. Being at a startup, I'm juggling several jobs and have a lot to check and handle, but I imagine big organizations end up in a similar place, since they have even more varied and bigger problems. Regardless of size, I think we're heading into an era where breadth matters more than depth.

6. Access to AI models becomes a social class

But using AI well might not be the whole story. We're heading into an era where which AI you use is itself a competitive edge.

As it happens, something blew up just two days before. A CMS configuration mistake at Anthropic exposed around 3,000 private files, revealing the existence of an unreleased higher-tier model called "Claude Mythos." It's said to be "dramatically" better than Opus 4.6, the current top model, at coding, reasoning, and cybersecurity, and access has reportedly been given to only a handful of organizations, mostly in cyber defense. So there's a model you can't get even if you have the money. It feels like a kind of power.

That led to a conversation about whether people will be split into classes based on which AI model they use. Tiered AI superpowers. When the same prompt produces completely different quality depending on the model, access to AI itself becomes a privilege. Right now the difference comes down to subscription fees, but I think it could keep widening. Which tools you can get your hands on ends up creating a gap too.

7. Half a year of senior developer work in one month

A tech leader recently analyzed how much code they'd written with AI and shared the results. Over about a month, switching back and forth between Claude and Codex, they produced what several senior developers would have needed months to do. Tens of thousands of lines of code changes. Alone.

This leader stressed that every engineering leader should carve out every hour they possibly can to practice AI coding. That way they'll understand what it looks like for the developers on their team to use AI well, they'll be able to guide them, and that's how engineering leaders survive in this industry.

OpenAI's lead developer and many leading experts predict that 90% of code will be written by AI by the end of this year, and that by the end of next year nearly every organization will code 100% with AI. Getting there, they say, requires harness engineering. Earlier, my friend at Company A said they didn't have time to use it, but looking at this data, "no time" is becoming hard to accept as an excuse.

8. Only people with a story will survive

Even so, using AI well may not be enough on its own. The last topic we got into was more fundamental. Someone brought up a book by Cho Soo-yong (former CEO of Kakao, the company behind KakaoTalk), and its core message was this: once AI raises everything we produce to the same high level, the only things that set you apart are personal appeal and personal branding. So everyone has to aim to become a kind of celebrity, or become someone who owns compelling IP.

The bar for content that does well stuck with me too: it should be so good that the audience feels grateful to the point of feeling a little guilty. Audiences aren't fools, so in the end it comes down to sincerity. The creator has to offer "information I'd pay to see" for something to go viral, and then it all comes back to you.

I once watched a video on Korean rapper Paloalto's YouTube channel where he interviewed artists about how they feel about AI music. On quality alone, AI makes music very well. We already enjoy AI-arranged remixes on YouTube without much resistance. But when we listen to music, we're quietly also taking in the story: which artist made it, and what path led them to make it. A song might briefly trend on TikTok, but in the end we become fans because we're drawn to the artist's color and story, and that pulls us deeper into the music.

I felt the same way watching Show Me the Money (a Korean rap competition show) recently. The songs are great, but the growth story of fighting through a brutal competition and surviving several close calls with elimination makes you far more invested in the music. If they'd just shown performance after performance from episode one, I doubt people would have loved that music as much.

In the end, what AI can't replace is each person's narrative, their story. The industry calls it personal branding, but put more simply, I think the most important skill may be for all of us to aim to become a kind of celebrity, or, if you don't have that kind of flair, at least to create that kind of IP and grow it. I suspect that's what organizations will look for in the future, and what individuals will need in order to survive.

As always, the book club started with AI and wandered somewhere completely different. But in the end it all connected. Not having time to use AI, ADHD-type talent getting attention, access to AI models becoming a class marker, only people with a story surviving: it all came down to what organization and environment you're in, and how proactively you move within it.

I'm thinking about holding a Claude Blue community event. A place where people share how they actually use AI in their own fields, and at the same time talk honestly about their own Claude Blue experiences lately. Honestly, there's way too much content out there on how to use AI well, and it's getting a little boring. I think what we need right now is more of a sense of community, where people can share their anxiety and FOMO honestly.

And this came up over dinner with my wife: with young people all flocking to Seoul, there's really nobody left in the rest of the country. Even if young people did move out there, there isn't much for them to do. Vegetables and fruit are expensive probably because farming outside the big cities is dying, and that's a problem AI alone can't solve. This follows on from my earlier post, "Crazy Ideas Beyond AI" I'm thinking about regularly volunteering outside Seoul on weekends to see how people actually live, and doing some market research on how AI could fit in. Lately I keep thinking we need to look outside AI too, instead of searching for answers only inside it.

Originally published in Korean on Brunch, March 29, 2026.