What Should Non-Developers Do Now?

Conversations with two startup AI engineers and a one-person business owner

Lately I've had a lot of chances to talk in depth with developers about Claude Blue. But to learn in the easiest, closest, most unfiltered way, I didn't have to look far. I just had to turn around and ask. So I sat down with two AI engineers at our company, turned on a recorder, and did a 45-minute interview with each of them, same as before. One has six years of experience, the other three.

I wanted to know whether Claude Blue is a bit overblown, and if not, how each of them is taking it and getting through it (it's just us, so let's be honest), and whether, having been through it first, they had any advice for non-developers. I told them I'd translate the jargon into language non-developers can follow. We kept the conversation going in the comfortable company lounge.

That afternoon I also had a coffee chat with a one-person business owner who has run a souvenir and promotional goods business for 15 years and asked to meet after reading my blog. Honestly, I went in without much expectation, and I came out stunned, as if I'd seen a glimpse of the future. So this is who goes through the AI transition first. While big companies weigh things through rounds of reports and reviews, startups and developers, and also self-employed people and freelancers whose livelihood is always right in front of them, jump straight in without overthinking it. And the results went beyond what I expected.

Just as there's a job category called AI developer, I figure there will be AI non-developers too. What will that look like? The easiest way to find the answer seems to be asking AI developers about every little thing they do and looking for clues. That should let us get at least three to six months ahead.

1. Claude Blue is real

First I asked the six-year engineer whether Claude Blue is just a buzzword. "The news is obviously exaggerated. But strip that away and some of it is true. Change is certain. The only question is how fast." Partway through the interview he learned for the first time that the name comes from "Corona Blue," the term Koreans used for pandemic depression. "Oh, there was a Corona Blue too," he said with a calm laugh. What came next wasn't light, though. "Of course I've had something like depression. I still do lately. Should I call it rolling depression? Most of the time I'm fine, but after an interview like this, I'll sink into it again on the way home, wondering what I'm supposed to do."

The three-year engineer was more vivid. In 2023, AI was exciting. Then after a business trip to LA last summer, reality hit her hard. "If I really want to survive, I need to start a business. Or go into a licensed profession. But it already feels a little late for that." These days she gets through one day at a time, working out every day. "Looking too far ahead just wears me out. There's no answer. And there's no real spark." At the end of the interview I asked whether she was looking forward to the future. The answer was short.

I'm not looking forward to it. Honestly, I'm a little scared right now.

What struck me was that she talks about this with her family at home too. Her father's reaction stays with me: "Our generation is the luckiest. We don't have to go through this change." An age where the retired generation calls itself the luckiest. That one line says everything about how heavy Claude Blue is.

There are more people around us who don't care about AI than you'd think. One friend in grad school just says "Wow, the world sure has gotten better" whenever AI comes up. For developers it's a matter of survival, but for many non-developers it's still something happening in a faraway land. The temperature gap right now is abnormally large. But when that gap closes someday, what if non-developers feel the same thing, or something even bigger?

2. Why developers got hit first

Why did developers feel AI's impact sooner? One answer kept coming up in the interviews: the way developers work resembles the way LLMs work. You get a problem, you structure it, you execute, you verify. LLMs do the same. "Verification is easy. It runs or it doesn't. The button is there or it isn't. It takes 10 seconds, and you need to get it to 3. The criteria are clear." That's how the six-year engineer explained it.

So developers used it first, and they also felt threatened first. Non-developer work is relatively unstructured, so it just took more thought to figure out how to apply AI. The three-year engineer was clear on this point. "It's true software got hit first, but other fields are going to get hit harder. There's no way they get hit less."

Interestingly, the developer community's culture of sharing sped this up. "In other fields people hide the core stuff, but for developers, open sourcing your work actually makes you look skilled. So when someone influential puts something out there, it spreads much faster than in other fields." In fact, once big companies like Spotify and people known as great programmers started declaring "I don't write code myself anymore either," it became hard to argue back. Apart from a few holdouts.

3. The Claude Code ecosystem

The six-year engineer summed up the recent shift in the developer world with a single name: Claude Code. "Claude Code built a really good ecosystem." When Cursor came out, sure, there was a reaction. But it was more "wow, that's neat, that's useful," and it didn't feel like an ecosystem was forming.

Claude Code was different. It focused on developers and rapidly evolved features like skills, agents, and MCP. Things that weren't there at first got added one after another, bam bam bam, and the ecosystem grew fast. When someone builds a harness, they don't lock it away. They share it so others can try it, and it became an environment where people can copy things quickly.

The three-year engineer also pointed to this moment as the turning point. "Things felt really different once Claude Code came out. Then OpenClaw came out and everything blew up all over again." The people in front are pulling ahead too fast. She said something the six-year engineer once said in the office, comparing AI to a mirror, keeps running through her head: "Sharp people will get sharper, and lazy people will get lazier." She sees herself somewhere in the middle, and the people ahead seem to be pulling away faster and faster.

4. Harness: how to saddle a horse

One idea came up again and again in the interviews: the harness. It's the hottest concept among developers right now. You put a saddle on the horse, add a whip, hitch up a carriage. Instead of riding the LLM bareback, you build a system on top of it and drive it where you want to go.

There's the workflow organizations have long used to build software: planning, dividing the work, assembly, QA, deployment. The heart of a harness is rebuilding that process on top of AI. According to the six-year engineer, this approach is already converging. "It's mostly about how well you replicate the existing organizational development process. But I think it's more or less converged by now. I don't chase every latest trend." New things keep coming out, but nothing has fundamentally changed.

The three-year engineer felt the same. She hooked a new feature up to Telegram, but she didn't actually have anything she wanted to send, so she stopped using it. "It feels like I'd have to do something just to use this, instead of borrowing it to reach some bigger goal of mine." You use tools for the work. You don't work for the tools. And if harnesses have converged, that also means it's not too late to catch up now.

At the same time, you can't ignore the power of inertia. One colleague on our team used Claude Code heavily and then canceled the subscription. They could have just paid again, but they didn't. They were getting along fine without it. Meanwhile, the AI muscle they'd built up was wasting away. If it hadn't come up in the interview that day, it would have slipped by unnoticed.

5. The bottleneck is what to build

The developer's identity is changing. In the past, a developer's role was to move part of the world's work onto software. HR, content, investing: taking areas that already existed and translating them into software. That translation took specialized knowledge.

Now that friction has shrunk. "The language we use to talk to computers has become natural language. We can work one level up, at a higher level of abstraction. You end up writing spec documents more than code." That's the three-year engineer.

As a result, "what to build" has become the bottleneck. "What everyone's saying is that the bottleneck is what to build. Building it isn't the bottleneck." Say someone makes a killer landing page. It doesn't take long to develop. The idea is what matters. The six-year engineer is currently building a Windows desktop app. He can build it, but problems keep popping up. In the end you have to understand how Windows works to make something solid. "Studying takes longer. Once you've studied it, getting it running isn't that hard."

There was personal pain too. What he'd originally aimed for was doing backend, frontend, and AI all at once. That was what set him apart. "But now anyone can do all three. What I used as my edge isn't an edge anymore, or it won't get the same economic value it used to. So there's that kind of depression." The software market itself is oversaturated. "The rice bowl here is small, so it spills over." Some people run 20 agents in parallel with OpenClaw, and he said, "What are those people even running so many of?" He isn't curious anymore.

6. Small business owners leap ahead

The three-year engineer told me about a music teacher. This past weekend, a music teacher with zero development knowledge built a music game with Claude Code. You play notes on a keyboard, and the screen shows you, game style, whether your harmony is right or wrong. She said she was so stunned that she couldn't focus for the entire hour-long lesson. She just sat there, dazed.

I heard a similar story from the one-person business owner I had coffee with that afternoon. They majored in fine art and have run a souvenir and promotional goods business for 15 years. They needed an internal inventory CRM and asked a development agency, which said it would take two months. On top of that, communicating with the developers was a real struggle. Then yesterday they discovered Lovable, a vibe coding tool, for the first time, and built the CRM alone in a single day. They showed it to me at our meeting, and I was amazed myself: the interface was pretty clean and the data ran correctly. They even added foreign language support for overseas staff and deployed it to a GitHub server so partner companies could use it.

In both cases, the trigger for jumping into development was simple. Rather than paying outside developers and handing over time and money, if you know clearly what you want to build, it's easier and faster to just build it yourself. The music teacher knew what game they wanted to make, and the business owner had a precise picture of an internal tool from 15 years of running the business.

This fits exactly with what the engineers said. Development used to be a special domain of expertise, and now it isn't anymore. These days the bottleneck is what to build. That's why freelancers and solo business owners are taking to AI faster. The problem they need to solve is clear, and nobody knows that problem better than they do.

7. How to talk to AI

What's the hardest part for non-developers using AI? The six-year engineer's answer surprised me. "Putting what I'm thinking into writing is always hard." It's a problem of expression, not a technical one. He said talking to AI is actually easier than talking to people. "Between humans it's much harder. People assume different things and use terms differently. With AI you can work with all that already aligned." But give AI something vague and it gives you something vague back. In the process of getting specific you have to make choices, and you have to steer so those choices come out well, and that isn't easy.

On top of that, AI's speed makes me, the human, more anxious. Give AI an instruction with a single click and it executes right away. But because I have to instruct it well, I think deeply, plan, and only then give the instruction. And when that thinking goes past 30 minutes, a strange sense of competition kicks in, like I'm falling behind the AI. Do I really need to think this hard before handing it over? I feel ridiculously unproductive. If speed is everything in the AI era and everyone is charging ahead, then sometimes isn't it more important to just move forward even if I don't fully get it? These thoughts collide every time, and the stress doubles.

On the practical side, the difference between Claude chat, Cowork, and Code also mattered. If it's a one-off, Claude chat is enough. It remembers a bit about you, and that's about it. If you need to keep building something over time, you need Cowork or Code, and if you're developing software, Cowork will be a struggle. Code is a native environment for running code. For someone who isn't going to develop code, it's a side issue.

The three-year engineer was frank about the hurdle for non-developers. "To automate non-developer work, you really do need to write code. If you know nothing and the AI just writes away, you don't even know how to use it, and even if you hack your way through, it's hard." But at the core it's the same.

Whether you're a developer or not, the job is the same: structure the problem and build a flow that hands it off to something else that can carry the load for you.

8. What it means to use AI well

I asked whether our company is using AI well, and what using it well even means. The three-year engineer's answer was crisp. "In this era, using AI well just means an organizational culture where people work in an AI-native way." It doesn't mean some grand AI orchestration. Getting AIs to talk to each other in a Slack channel isn't what using it well looks like. If the organization has a structure where people can keep asking "Which parts of my work can I improve with AI?", that's using it well.

The problem is that when you're buried in the tasks in front of you, you don't have room to think about that. "Bringing in something new takes time and energy. But you have to do it to get 10x later. If you only look at what's in front of you, that time feels wasted." Our company once tried to drive adoption under the name "AI tasting," and her honest review was this: "The theme was a bit vague. Sometimes it felt like a toy project. It would have fit the purpose better if we'd gone all in and made it seriously about work efficiency." A modest hit, I'd say.

Big companies have it even harder. To bring in anything, you need approvals, and squeezing something new into an ongoing enterprise project is risky. As I mentioned before, there's a story that Stripe, the global fintech giant, nearly shut down the company for a week and had every employee go through an AI transition. Even that might or might not be enough. It doesn't happen all at once. You have to change mindsets, step by step.

Information fatigue piles on top of all this these days. Reading LinkedIn and newsletters isn't what it used to be. "There used to be real insight, but now there's way too much noise. AI slop itself has increased. There are definitely more people copying what someone else wrote, rephrasing it, and posting it as their own." That's the three-year engineer. Even when she sees a post that actually helps, just thinking "when am I going to apply this to my workflow" has become exhausting. With information fatigue stacked on top of Claude Blue, she said she's gradually stopped reading LinkedIn and newsletters altogether.

She described this feeling as "damage over time to my mental strength," like a slow poison effect in a game. She's built up her physical stamina by working out, but her mental strength gets chipped away a little each day. Her stamina has come up a lot, but her mental strength can't keep pace. As AI has gotten more advanced, the pressure on time has gotten worse. There's a constant sense that every minute and second is being wasted.

"Walking along with a compass and a map may well be the right way. But right now, I can see sports cars zooming past right in front of me." Developers have always been trained to look for a better way. They've done that for years, and now doing it makes them feel unproductive, and they lose the will to do anything.

Is this right? Eh, let's just run for it.

9. Everyone's own compass

The six-year engineer copes by taking a historical view. "Skills and careers I had disappearing isn't a one-time thing. Abacus operators, stenographers, carriage drivers. There are tons of examples. This isn't something special that's only happening in 2026." Once he accepts it that way, the depression eases a little. These days he's turning his attention to fields like robotics and hardware, where you "can't just whip something up." Areas you can't build quickly by yourself the way you can with software. There might be possibility there. He's also in the middle of planning his wedding, so apart from his AI worries, a big life event is reshuffling his priorities.

The three-year engineer copes differently. "Avoidance. Escaping reality." She said it with a laugh, but she meant it. "Because if I didn't even do that, it would be too hopeless." She works out every day, building her stamina. She said she feels like "a wanderer who lost her domain." She only ever did development, and that domain is disappearing. Even if she wants to start a company, to compete with LLMs she'd have to get domain specific or build a far better system, and she wonders whether that's something she can actually do.

That's why I've been studying hard for a wine certification lately, you know.

She also had a serious talk with her younger sibling, who majored in video editing. She sat them down, told them "I don't think you should keep doing this," and convinced them to go to Australia on a working holiday visa. Cooking, physical work, anything, just try it, she told them. She's seeing up close that video editing is one of the jobs AI is replacing fastest.

Her thinking on FIRE (financial independence, retire early) has changed too. She used to think there was no need to retire early, once she realized the point was doing work you love for a long time. These days she's thinking again that she needs to save money fast. "I don't know how many jobs will be left. How long I'll be able to do things that make money." Because of that anxiety, she feels she needs to build a safety net. Her thinking keeps going back and forth.

At the end of the interview, The Matrix came up. The red pill and the blue pill. "Taking the blue pill doesn't seem so bad." Half joking, of course. Because we've all taken the red pill called startups, and we're gritting our teeth and colliding with things at the leading edge of the trend. Still, going back and forth all the time is just an honest, realistic state of mind.

Two AI engineers and one small business owner. Put their three stories side by side and you see a paradox. The developers are moving past coding and digging into business domains, while the business owner is moving past their business domain and digging into coding. The line between them is blurring.

So what should AI non-developers do? Putting together the answers from the interviews, here's what I'd say. Dig deeper into your own domain (as long as that domain hasn't been ambushed by AI yet). Build the habit of structuring problems the way an LLM does. Practice organizing your thoughts in writing, a lot. Harnesses, meaning ways to put guardrails on AI so it works efficiently, are already converging, so it's not too late to start now. And above all, don't think of what developers are feeling right now as someone else's problem. It might be the mild version. It might be only a matter of time before that feeling comes at you as a much bigger snowball.

I'm not an AI expert, but I'm always up for a coffee chat. If you'd like to share your story or hear more about other people's experiences, reach out on LinkedIn.

Originally written in Korean on Brunch, March 25, 2026.