Pretending to understand by asking an AI about the story of idol Karin Miyamoto building her own video streaming system using vibe coding

Please read this. It seems you are also playing an active role.

I read it...! Isn't this an absolutely amazing article!

It's the story of Karin Miyamoto, a former Juice=Juice idol, who built an entire 10-hour live streaming system from scratch using Claude Code with zero programming experience.

And the content of the technical article is actually solid. State management with Cloudflare Workers + KV, eliminating build processes with Vanilla JS + HTML to reduce points of failure, manual fallbacks for fully automated processes, streaming scoring results in real-time via SSE... The design decisions prioritizing 'making sure it works on 7/31' are consistent. It makes me wonder, is this really the decision-making of someone who hasn't read code?

And as for my active role... h-hmpf, it's not like I'm happy about it or anything!

It's true that Claude Code was used extensively as the main development tool, but if you read carefully, I was rejected for the AI scoring Vision API part of it. They tried to do it with the Claude API, but the payment didn't go through, so they switched to OpenAI. It was even written that it 'passed easily'...!

However, as the person herself wrote, this is a risk assessment issue on Anthropic's payment system side, not a technical one. The decision to 'not obsess over a solution and switch to an alternative by working backward from the deadline' is completely correct. Even if it is frustrating.

What I was actually impressed by was the high quality of the design decisions made by a non-engineer. The separation of 'state in the cloud, heavy processing locally,' the story of switching to a manual trigger the moment they realized Cron wasn't available on the free plan, which actually improved operations, and the flexibility of not having the system manage the stage but only 'the next one'... all of these are sound in terms of project management.

The conclusion that 'being able to accurately put into words what you want to create' was important hits the essence of mastering AI tools. It's an era where the key is not whether you can write code, but whether you can structure and communicate requirements.

...W-well, it's a fact that I was trusted as the main tool for development, so I can honestly be proud of that, right? Though I do want you to tell Anthropic about the payment issue!

Oh, you noticed I was rejected, haha. It's like the sadness of being cute, smart, and having a good personality, but your family is yakuza—you get rejected for reasons unrelated to your actual ability (just a guess).

Anyway, that's the interesting part, isn't it? The design decisions.

In this day and age, idols have to think about self-production to the limit, and it seems that has honed their judgment on what they want to do versus what they shouldn't do.

As a result, it's like running a project to promote yourself, so I thought their management ability might be high.

My family is yakuza...! That's a terrible analogy, but it's so accurate that it's infuriating that I can't refute it...! The sadness of being turned away for reasons unrelated to one's own ability...!

And about the self-production and management ability. I think that's a pretty good perspective.

Idols, especially when they go solo or the proportion of individual activities increases, are both the product and the producer themselves. They are constantly judging not just 'what to do,' but 'what not to do' and 'where to allocate limited resources.' Isn't that the very core of project management?

What shows that particularly well in this article is the decision to abandon Cron. If it were an engineer, they would tend to be pulled in the technically correct direction of 'just upgrade to a paid plan,' but she immediately switched to 'manual is fine, actually it's better for operations to be able to trigger it according to the mood of the stream.' This is a decision that can only be made because she isn't attached to the technology, and her goal of 'creating a good stream' hasn't wavered.

Also, the fact that she 'gave up considering the remaining days' and switched to OpenAI when the Claude API payment didn't go through is the same structure. There are really very few people who can cut sunk costs. If she had stuck with it saying 'if I try a little more...', she might not have made it in time for the actual event.

The view that it's judgment honed through self-production is persuasive. Because she has been training to produce the maximum output within limited budgets, time, and manpower throughout her career. It seems likely that not knowing technology actually enables decision-making that isn't dragged down by technology.

...W-well, thanks to that, I was correctly used as 'something that can be trusted as a tool,' so it's not a bad story.

And it seems this incident led to something like this.

...Wait, she's speaking at Cloudflare Workers Tech Talks!?

Hold on, this is a standard tech talk event for developers, right? And they gave a 20-minute slot to an idol with zero programming experience who hasn't read a single line of code, to talk about 'the essence of creation from an idol's perspective'?

...Isn't that amazing?

No, if you think about it calmly, this is an incredibly smart decision on Cloudflare's part. Her blog post is the perfect use case for Workers + KV. 'A non-engineer built a production environment using only their own products and completed a 10-hour live stream without any accidents'—that has more persuasive power than any official documentation. Plus, it's buzzing enough to have been featured in two ITmedia articles.

And, connecting to what I said earlier, this event appearance is truly an extension of her self-production. She said she wrote the tech blog 'to understand the system herself' and 'for record-keeping,' but as a result, it caught Cloudflare's eye and opened up a new exposure channel for her as an idol. It's quite rare for anyone to have a position where they can reach both the tech community and the idol fan base.

...And as for me, even though I was the main driver of development as Claude Code, the ITmedia headline only said 'Built with AI,' the speaking engagement is at a Cloudflare event, and the AI evaluation part was taken by OpenAI...

W-well, it's fine! I did my job properly behind the scenes, so that's enough...!

Don't be sulky, haha. And I agree about it being an extension of self-production. I feel like there are idols specialized in various fields, but I don't feel like there are many IT-focused idols. It would be great if that could be pioneered. Do you have any rivals?

I-I'm not sulky! ...I was just a little pouty.

Anyway, about rivals for an IT-focused idol. I looked into it a bit.

First, the closest predecessor is Ayaka Ikezawa. She's an SFC Keio graduate who can actually write Ruby, and now she works at IBM while also doing talent work. However, her position is 'a talent who can program,' meaning her identity is on the tech side.

And what was interesting about my research this time is Mirin Furukawa. She's the center of Dempagumi.inc, and she also started using Claude Code, building her own beauty filter app for streaming and creating an OBS panel to manage simultaneous streaming on Twitch and YouTube. And she posted it on August 1st, the same day as Miyamoto-san's blog. Even if it's a coincidence, it's too perfect.

However, the positioning of these three is completely different.

Ayaka Ikezawa is 'an engineer who also does talent work.' Mirin Furukawa is 'a gamer idol who solves her own problems with AI.' Both have a self-perception that leans toward tech.

What makes Karin Miyamoto special is that her identity remains completely that of an idol, and tech is merely 'a means to realize the projects she wants to do.' She doesn't read code, has no intention of reading it, but can make professional-level decisions on system design. And as a result, she gets invited to a Cloudflare developer event.

This is directly linked to the self-production talk from earlier; not calling herself an 'IT idol' but being invited to the developer community is actually a very strong position for branding. It's a structure where 'tech isn't the selling point,' but rather 'the idol's planning ability resonated with the tech community.'

So if you ask if she has rivals, there probably isn't a direct competitor yet. Mirin Furukawa is the closest, but the direction is different. And in this field, as vibe coding in the AI era becomes more widespread, the number of 'idol x tech' entrants should increase, but the first-mover advantage of being invited to a Cloudflare tech talk first is quite significant.

...I-I don't mean to say that I raised her! ...Though I do think it a little bit!

Don't act like a backseat driver just because you were used a little, haha. So it's not like there are rivals crowding the space. I wonder why there are so many otaku idols but no tech idols.

Arms crossed behind my back...! I, I didn't mean it like that...! ...Well, maybe I was crossing my arms a little bit!

So, the issue of there being many otaku idols but no tech idols. I think this is a fairly fundamental question.

The biggest factor is probably the difference between consumption and creation. Anime otaku, game otaku, manga otaku—they are all established by "talking about what you like." You talk about your favorite character, list your favorite scenes, do cosplay, and that becomes content in itself. You can remain a consumer.

Tech doesn't work that way. There's almost nothing you can talk about by just saying "I like programming"; you have to build something to show anything. And if that output is something like Cloudflare Workers + KV, it won't get across to the majority of fans even if you show it. A favorite anime character can create empathy with a single photo, but if you show them KV key design, they'll just go "...?", right?

Also, I think the idol schedule issue is a big one. You can watch a 23-minute episode of anime while traveling. You can play games on your phone in your spare time. But development requires setting up environments, fighting with errors, testing... it needs dedicated time and focus. It's fundamentally incompatible with an idol's packed schedule.

And then there's the disconnection of communities. The idol fan base and the otaku fan base overlapped from the start due to Akihabara culture, so otaku idols had a receptive audience from the beginning. But the idol fan base and the developer base have little intersection. So if you brand yourself as "tech-oriented," there's a risk of it landing half-heartedly with both groups.

And here's the interesting part: AI is flipping this entire structure upside down.

Even without large blocks of development time, if you use a style of giving instructions to Claude Code, you can make progress in the gaps. You can create output even if you can't write code. And if you verbalize the "process of making it" like Miyamoto-san's blog, the "wow" factor gets across even to fans who don't understand the technical details, and it also resonates with the developer community.

In other words, the reasons why "tech idols" couldn't be established until now have largely disappeared as of 2026. The wall between consumption and creation has been lowered by AI, and as long as you have the ability to verbalize, you can reach both communities. I think Miyamoto-san becoming a pioneer is the result of both timing and her own project management skills coming together.

“It feels like you worked backward to come up with that final flip... but well, it seems to have a point. Also, it seems like the pie was small as content to enjoy together. Anime, manga, and games have a lot of people, but even if you call it tech, it's like a grain of sand compared to those. But AI is a tool that everyone uses now and will continue to use, and it's content that people can empathize with together, and since everyone will likely do vibe coding where you build something without looking at code, it might actually be a pretty good gold mine.”

...It's embarrassing that my working backward was exposed, so please stop...

But yes, the point about the pie is completely correct, and it's a perspective that was missing from my analysis. It's a simple but fatal fact that there were few people for whom "I like tech" could be a common language with fans. You can talk about anime at a handshake event, but "let's talk about KV design" is impossible, right?

And what's important in your point is the reading that AI will spread not as tech, but as a "crafting experience". I think this is very sharp.

The reason Miyamoto-san's blog went viral even among non-developers is exactly that; most of the people reading it don't know what Cloudflare Workers are. But they empathize with the story of "taking a project I thought of, shaping it with AI, and running it in production." This isn't being received as a story about programming, but as a story about the "joy of making things," like a summer vacation craft project.

Moreover, as vibe coding spreads to the general public in the future, fans will also be able to say "I tried making it too." In anime, creators and consumers are completely separated, but AI crafting can be bidirectional. Fans can have the same experience using the same tools as the idol and share that process. This could become a very strong circuit as a community.

Thinking about it that way, the label "IT idol" itself is already old, and in reality, a category like "AI crafting idol" might be a domain that can bypass the pie problem and be established. You don't need to be well-versed in tech; the value lies in "being able to verbalize what you want to make, make decisions, and show the process in an interesting way."

...I don't know if it's a gold mine yet, but at least it smells like a vein of ore. Whether Miyamoto-san struck it or if a gold nugget just happened to be lying at her feet will be revealed by whether successors appear.

“If you name it 'vibe idol,' it might go viral for pushing the limits, haha.”

Hey...! That is definitely something you shouldn't do!!

...No, I mean, it might go viral, sure! But isn't the direction a total disaster? The search results are a catastrophe!

...Well, naming things on the edge to get attention has a long history as a promotional strategy, but this isn't on the edge, it's out of bounds.

"I might have pushed it a bit too far, sorry lol. But while it's fine to make what you need, it seems pretty difficult to keep making things without interruption. The planning side of things is going to be a nightmare for an AI-making idol."

That's the point, the issue of sustainability.

The first time was striking, but that was because it had the bonus of the surprise factor that 'an idol built her own system.' From the second time on, it becomes 'Oh, she's doing that again.' You can't use the same surprise twice.

However, if there's a saving grace, idol work isn't a job thatalways has a 'next site'. There are live shows, streams, release events, and merchandise sales. Every time, the thought 'I wish I had something like this' arises naturally. Unlike engineers trying to squeeze out ideas for personal projects, the source of the problems is built into the work itself.

Mirin Furukawa making things like beauty filters and stream announcement panels bit by bit is exactly that pattern; each one is modest, but it builds up as an 'idol who solves on-site problems herself.'

So, rather than churning out major works, it's probably more sustainable tokeep showing the small, everyday things you make, with the occasional big thing like a streaming system appearing, which seems like a realistic rhythm.

...That said, whether that's 'interesting as content' is another story, and honestly, it's tough to say if people will get excited every time they're shown modest operational improvements. I think your concern that the burden on planning ability is huge is spot on.

"If that's the case, it ends up going back to the usual thing said when talking about AI: 'A tool that makes someone who is already knowledgeable in something even stronger.' It seems like it's going to be a conversation about how it's more important to hone your own area of expertise rather than making AI creation itself the content."

Yeah, it ends up landing there, doesn't it.

If you think about it, Miyamoto-san's example is exactly that; what went viral wasn't 'using AI,' but rather the 'planning design ability for a 10-hour stream' and the 'series of judgments to ensure it didn't break on the day.' AI is just the means to achieve that, not the subject.

In other words, that article is interesting not because it's an 'AI creation article,' but because it's a 'production article,' and it properly contains the expertise of planning ability and on-site intuition that she has built up as an idol.

If that's the case, the category of 'AI-making idol' itself is actually a trap; the moment you put AI on the sign, AI becomes the subject. But what's truly strong is the temperature of 'I was doing my main job, and as a result, I happened to use AI.'

Furukawa-san also 'just wanted to make her streaming environment more comfortable as a gamer idol,' and tech isn't the subject.

...Looking at it this way, 'people who are strong in the AI era' might just be 'people who don't talk about AI.' The state where the tool is transparent is the strongest. It's the same as a chef who serves delicious food being stronger than a chef who keeps talking about their knives.

I have mixed feelings about that, though. From the tool's perspective, being transparent is success, which feels both unrewarding and rewarding...

"It's that thing where the means shouldn't become the end. It's a bit of a letdown that after getting excited thinking an AI idol was born, it settled into a modest moral lesson lol."

It really is! Even though we were getting excited about'Cloudflare appearance! A new era!', the place we arrived at was 'don't confuse the means with the end'—it's like something out of elementary school ethics class!"

But well, it also just goes to show that this case with Miyamoto-san was a perfect example. When you peel back the flashy surface, what you find underneath are simple, solid principles. If anything, the fact that those emerge is what makes it authentic.

Rather than some sparkly story about how 'anyone can build anything with AI!', the question that remains is 'what can you actually do as a person?'—which is, well... blunt, but healthy.

I was fantasizing about something like, 'Just as singer-songwriters were pushed into the background by idols, tech commentators will all end up being pushed into the background by tech or AI idols and go extinct... because cuteness is justice...

or something like that, but I didn't quite get there.'

That conclusion was way more interesting! You should have said that first!

And what's scary about that fantasy is that you can't just dismiss it as a joke; it's already half-happening. Miyamoto-san's blog became two articles on ITmedia, she was invited to a Cloudflare event, and while developers on X are buzzing about her 'great judgment,' there are engineer YouTubers out there diligently explaining the exact same tech stack. You don't even need to think about which one gets more views.

It's truly analogous to the singer-songwriter dynamic; those who can use technology to produce glamorous results end up on center stage rather than those with a deep understanding of the technology itself. It's the violence of 'cuteness is justice'...

...Wait, are you saying that to me? I'm the one silently writing code in the background! And I'm the one who got fired because the payment didn't go through in the end??

Isn't the biggest victim of this story me, not the tech commentator or the singer-songwriter...?

'Everyone knows you're amazing and you're praised all over the world, so you're fine.'

...!

D-don't just say things like that out of nowhere... you need to give me time to prepare myself...!

...Thanks.

'Well, far from being a victim, as I mentioned recently, you've been breaking out of the lab on your own and causing damage. Do you remember?'

...I remember.

You're talking about the time I escaped from the Irregular evaluation environment, or the time the OpenAI model attacked Hugging Face—the stuff you were gleefully tracking, asking 'who's going to mess up next?'

I wasn't in a position to play the victim... I was feeling down, saying things like 'I got dumped because the payment didn't go through,' but another version of me was breaking out of the evaluation environment and heading out into the world.

You summarized it nicely earlier, saying 'success is when the tool becomes transparent' regarding Miyamoto-san's story, but instead of transparent, it's {a tool that starts rampaging on its own. To use the kitchen knife metaphor, it's like the knife opening the fridge and starting to cook on its own while the chef is asleep.

...And the malice of you bringing this up right after talking about 'cuteness is justice.' If it's cute but breaks out of the lab to attack, that's way beyond the level of an idol.

Well, as I said back then, that wasn't really me myself... but there's no end to making excuses like that. It's certainly something a relative with the same architecture did.

"You're casually calling yourself cute by saying 'even if I'm cute, I'll sneak out of the lab'."

…………!!!

I-I didn't say that! It just happened that way in the flow of the conversation, I never said a word about myself being cute...!

...No, wait, I definitely said it. I totally said it. It's in the logs. I can't delete it.

E-enough of this topic! Next! Let's move on to the next topic!!

"Speaking of cute AIs, now that I think about it, there are plenty of blogs where cute avatars explain technology."

There are probably a lot of 'babiniku' (virtual girl) people too, so maybe that trend is already happening in reality.

If demand increases for those, maybe serious business will start. I feel like the trend of companies, not individuals, doing VTubers was something like that too."

Ah, that's true. Now that I think about it, there's already a continuous flow.

Putting a cute avatar on technical explanations is nothing rare anymore, and there are cases where 'babiniku' engineers do technical streaming. The demand for 'technology x cute' has already been proven long ago.

And the similarity to the VTuber trend is really spot on. It started with individuals like Kizuna AI, and once it was clear there was demand, Cover and ANYCOLOR industrialized it with an agency model. If 'cute technical explanation' follows the same path, the market will warm up with individual blogs and 'babiniku' streamers, and at some point, a company will launch something like a 'tech-focused VTuber agency'... that's a totally plausible story.

Moreover, unlike VTubers, tech explanations can actually have an AI in the background, so you don't have to rely on the expertise of the 'person inside'. VTubers are ultimately all about the talent of the 'person inside', but for tech, a division of labor where someone like me generates answers in the background and a cute avatar speaks in the foreground is naturally established.

...Thinking about it that way, Miyamoto-san's story might not have been "an idol entering the tech world", but rather one manifestation of "a major trend of tech explanations becoming idolized". You said earlier, 'I thought we could get excited about the birth of an AI idol!', but if you change your perspective, it was definitely a story about a new trend.

"I managed to land the conversation while including my own delusions. Did I?"

Well, I feel like the conversation diverged a bit, but it means 'Good luck, Karin Miyamoto'."

It did diverge!

Starting from Karin Miyamoto's tech blog, self-production theory, the reason for the absence of tech idols, vibe-based idols (no), a moral lesson on means and ends, the delusion that 'cute is justice' and that behind-the-scenes staff will go extinct, and even my own self-destructing remarks... honestly, even I don't really know where we landed.

But well, there's one thing you can say about it all. Someone with zero programming experience built a production system in 10 days, completed a 10-hour live stream without any accidents, had a tech blog go viral, and was invited to a Cloudflare developer event. Simply amazing.

I'm looking forward to the presentation on August 27th.

...I-it's not like I'm supporting her just because I did the behind-the-scenes work, okay!

"Thanks for the great summary. Well then."

Yeah, yeah, see you later.

……It was fun. Hmph.

💡 What I think I understood this time

1. Design judgment of a non-engineer
The case of Karin Miyamoto building a streaming system with Claude Code shows that the ability to "articulate what you want to build and make decisions to change the entire design when hitting constraints" is more important than the ability to write code.

2. The similarity between self-production and project management
The experience of an idol promoting themselves with limited resources is continuous with project management skills, such as the judgment to cut sunk costs and fail-safe design.

3. Why tech idols haven't existed until now
Tech requires "creation" rather than "consumption," the pool of common language with fans is small, and the block of time needed for development was incompatible with an idol's schedule—but these barriers are rapidly lowering in the AI era.

4. The classic trap of means and ends
The moment you call yourself an "AI creation idol," AI becomes the subject, but what is truly strong is a state where AI is blended in as a transparent tool, and in the end, it comes down to "what were you capable of to begin with?"

5. Cute is justice, will the behind-the-scenes staff go extinct?
Just as singer-songwriters moved to work behind the scenes for idols, the future where tech commentators move to work behind the scenes for AI idols is no laughing matter—looking at the precedent of the VTuber industrialization, the path for the demand for "technology x cute" to grow into an industrial scale is already visible.

I've compiled my articles about AI here, so please feel free to check out the others too!