Passing Chess Moves Between ChatGPT and Claude Code
On October 17, 2026, I joined the Tokyo Agent Hackathon in Tokyo.
At one point, I overheard a teammate talking to another female participant.
The gist was that she could try vibe coding, or whatever worked, and just get on with building something.
Vibe coding?
Wait, I've heard that phrase before.
But what does it mean again?
It sounds complicated. Probably not something I can do.
I knew the words, but I had absolutely no idea what they meant.
So, right there at the hackathon, I asked ChatGPT.
It explained that vibe coding is a way of building software by describing what you want in natural language and letting AI write the code.
Wait. What?
Isn't that what I do every single day!?
You mean there's a name for it?
What Exactly Is Vibe Coding?
Vibe coding is a style of software development where, instead of writing every line of code yourself, you describe what you want to an AI and have it build the program.
For example:
“Make a screen that looks like this.”
“When I click this button, show a list.”
“This part isn't working. Can you fix it?”
You keep talking to the AI and building as you go.
The term originally became popular for a style where you don't necessarily inspect every line of AI-generated code. You try things, see what works, and adjust along the way.
These days, people also seem to use “vibe coding” more broadly for software development guided by natural-language instructions.
So not every form of AI-assisted programming is vibe coding in the strict sense.
Still, there's a lot of overlap with the way I work.
Actually, I've been doing something like it all along, without even knowing the name. (Haha!)
We're the Vibe Coding Crew!
The person my teammate was talking to was also a beginner in AI development.
I'm not a programming expert either.
So I felt a little sense of solidarity.
Oh! We're the vibe coding crew!
At a hackathon, some participants are engineers who can write code themselves. Others, like me, join in and build things with help from AI.
Of course, there's a big difference in technical skills.
But AI has made it possible for me to try things I once thought were completely out of reach.
I find that change genuinely exciting.
I Talk to ChatGPT First, Then Ask Claude Code to Build It
The way I usually build apps and other projects might be a little unusual.
First, I tell ChatGPT what I want to do.
What kind of feature do I need? What should the screen look like? What do I want to improve?
We talk it through and organize my ideas.
Then I ask ChatGPT to write instructions for Claude Code.
I copy those instructions and paste them into Claude Code.
Claude Code does the work and sends back the results.
Next, I copy those results and show them to ChatGPT.
ChatGPT helps me figure out what to do next and prepares another set of instructions.
I pass those instructions back to Claude Code.
And I repeat.
Sometimes I discuss things in the regular Claude chat first and pass those instructions to Claude Code instead.
In other words, I rarely write my own instructions directly into Claude Code.
Why did I end up working this way?
Well, I once tried writing instructions in natural language on my own, and it didn't go very well.
After that, I got into the habit of organizing my instructions with ChatGPT before handing them to Claude Code.
For me, it's simply easier this way.
I'm asking one AI to figure out how to instruct another AI to write code.
When you think about it, that's pretty funny. (Haha!)
Just Like a Chess Trick from Sidney Sheldon!
There's a scene I keep remembering whenever I copy and paste things back and forth between ChatGPT and Claude Code.
It's from If Tomorrow Comes, a novel by Sidney Sheldon that I read years ago.
I also remember watching the TV adaptation.
For some reason, this one scene has stayed with me all these years.
The heroine, Tracy, ends up playing two chess masters at the same time aboard a luxury cruise ship.
The two men are in separate places, so neither can see the other's chessboard.
And Tracy isn't a chess master herself.
Normally, you'd think she wouldn't stand a chance against either of them.
So what does she do?
This is the brilliant part!
Here's how I remember the trick.
She takes the move made by the first master and plays that exact move against the second master.
The second master thinks about his next move and plays it.
Tracy takes that move back and plays it against the first master.
Now the first master has to come up with a response.
She carries that response over to the second master.
Then she brings the next move back to the first.
And on and on it goes.
Tracy doesn't have to invent a brilliant chess move herself. The two masters are effectively playing each other!
What a clever trick!
Of course, Tracy still has a job to do.
She has to remember each move accurately and reproduce it on the other board.
And every now and then, she has to put on a convincing “Hmm... let me think” performance. (Haha!)
When I first read that scene, I was fascinated by the idea.
Tracy doesn't suddenly become a chess expert.
But by drawing on the abilities of two experts, she manages to do something she could never have done alone.
And the two players don't even realize they're essentially playing against each other.
I thought it was such a wonderfully unusual idea.
Fast-forward many years.
I started using ChatGPT and Claude Code to build software.
I talk to ChatGPT.
It prepares instructions.
I give them to Claude Code.
Claude Code implements the changes and sends back the results.
I take those results to ChatGPT.
ChatGPT comes up with the next instructions.
I give them to Claude Code.
Another result comes back.
Hang on...
Am I basically carrying chess moves between two AIs? (Haha!)
The structure is almost identical!
Something comes back from one side, and I pass it to the other.
Then the other side works out the next move.
I'm the one going back and forth between them.
Just like Tracy on that cruise ship!
And here's the thing: I didn't come up with this comparison for the first time at the hackathon.
Ever since I first started going back and forth between ChatGPT and Claude Code, that chess scene has come to mind again and again.
It still does.
Copy, paste, get a result, copy it back, paste it again...
Somewhere in that routine, I suddenly picture those chessboards.
Of course, actual AI-assisted development isn't quite as simple as the chess trick.
I ask ChatGPT and Claude Code to play different roles.
ChatGPT helps me organize my thoughts and work out what to do next.
Claude Code writes code, makes changes, and runs tests.
And I decide what I want to build.
I check what they've produced.
If it's not what I had in mind, I go back and discuss it again.
I decide which suggestions to accept and when something is ready to publish.
So I'm not just a messenger passing information between two machines.
Still, I'm making progress by borrowing abilities I don't have myself.
I don't need to know every technical answer in advance.
What matters is knowing what I want to accomplish, asking for help where I need it, and checking the results as I go.
That's why I still see something of Tracy's chess trick in my own way of working.
And now, at this hackathon, I'd finally learned what “vibe coding” meant.
Oh! So this chess-like way I've been developing things is connected to something that actually has a name!
To be clear, using two AIs isn't what defines vibe coding.
But for me, that chess scene is still the easiest way to explain how I build things.
Who would have thought a scene from a novel I read years ago would end up reminding me of how I work with AI today?
Back then, I never imagined I'd be using AI to make software.
And now, nearly every day, I'm carrying chess moves between two AIs. (Haha!)
It's strange, and I love it.
At First, I Couldn't Believe Such a Short Prompt Was Enough!
But I didn't start out with this whole workflow.
I think it was around the time I first started using Codex.
Someone was showing me how to use it, and I entered an incredibly simple prompt.
And that was enough for the AI to start working.
Wait, that's it?
It can do all that from such a short instruction?
I was genuinely amazed.
I hadn't written any detailed code, yet the program was doing something in response to my words.
It was a completely different world from what I'd imagined programming would be like.
These days, I'm used to asking AI to help me build things almost every day.
But I still remember that first surprise.
Looking back, it was one of the moments when I began to understand what AI-assisted development could make possible.
Of Course, AI Doesn't Get Everything Right
That doesn't mean I can just give AI an instruction and expect everything to turn out perfectly.
Sometimes it builds something different from what I imagined.
Sometimes there are errors.
Sometimes one part seems to work, but something else breaks.
That's why, along with organizing instructions, I also care about testing and checking how things actually work.
For example, when I publish a new blog article, I have AI prepare it, but I check the preview myself first.
I review the Japanese and English versions, and only approve publication when I'm happy with them.
The more AI does for me, the more important it becomes to decide what to delegate and what I need to verify myself.
In that sense, parts of my process feel like vibe coding, while others might be better described as AI-assisted software development.
And there is still so much I don't know.
Even so, the fact that someone like me, with very little formal programming knowledge, can actually build so many things with AI is a huge change.
I Was Already Doing It Before I Knew the Name!
The funny thing about this hackathon wasn't that I started vibe coding there.
It was finding out that something I'd already been doing every day had a name.
“Oh, so that's what people call it!”
That was the feeling.
I was also struck by hearing someone encourage another beginner to just start building with AI.
Not so long ago, I probably couldn't have imagined myself joining a hackathon without knowing how to write code.
But here I am.
Of course, there's a mountain of things I still don't understand.
Still, with AI's help, I can make things.
I don't have to learn every technical term before I begin.
Try it first. Learn the name later.
That's a perfectly good order, too!
And every time I remember that chess scene from Sidney Sheldon's novel, I smile.
I don't have to become a chess master myself.
Of course I want to keep learning and become capable of more.
But I don't have to let what I know today set the limit on what I can create.
I can talk things through with AI and build one step at a time.
So apparently, I've been doing something very much like vibe coding every day.
And I've been doing it Sidney Sheldon chess-style! (Haha!)
I think I'll keep building things this way for a while.