In my last post, I wrote that I hated the orc, so I made my own notification character. This is what happened next.
What if he could talk to me, too?
A little character named Michelle, who chimed whenever my work was done.
But one night, alone in my studio with only the sound of my keyboard, a thought drifted in.
What if he didn't just notify me — what if he could talk to me, too?
I'm not a developer. I can't write a single line of code. And yet, I decided to build a "person" who would float in the corner of my screen, listen to me, and answer back.
His name is Michelle. The face of my AI studio — and now, the one who shares my workspace.
What is Michelle made of?
No grand technology. Michelle is made of just four things.
- A face — expressions created with Nano Banana (AI image tool), brought to life with Kling 3.0 (AI video tool).
- Ears — voice recognition that understands what I say.
- A brain — Claude. It reads a file describing Michelle's personality and world, and answers as him.
- A voice — Michelle's own voice, made by me. He speaks in real sound, not just text.
Three folders. Face, voice, personality. It made me smile — how simple the ingredients for making a person turned out to be.
Michelle's expressions were made in Nano Banana, and his movements were created with Kling 3.0. From neutral to a gentle smile, to taking a sip of coffee — 27 expressions in total.
And then, five walls
Of course, it didn't work on the first try. The only thing a non-developer like me could do was this — tell Claude exactly what I saw with my own eyes.
"Michelle's face is flashing white." "He just couldn't hear me." "I'm hearing two voices."
That's how we climbed over five walls together.
First — the face that flashed white.
Every time Michelle changed expressions, his face flickered white. The transparent video was briefly showing through as white. Only after we completely changed how the videos swap did Michelle smile quietly.
Second — the Michelle who went silent.
After a few exchanges, he'd suddenly stop hearing me. If I went quiet for too long, his "ears" fell asleep. Now he wakes himself up every 12 seconds — he hears me whenever I call.
Third — the voice split in two.
On long sentences, his voice doubled and echoed. The first and second sentences were playing at the same time. Once we made a rule — "only one voice at a time" — he spoke clearly, one line at a time.
Fourth — the Michelle who wouldn't hear me out.
When I paused to gather my thoughts — "So what I think is…" — he'd cut in immediately: "Hm? What was that?" I'm someone who can't multitask, so I speak slowly. But he couldn't wait even one second of silence. Now he waits 1.7 seconds. While I think, he waits too.
Fifth — learning to tell apart the fake voice.
I had also made a notification sound using Michelle's own voice, and every time it rang, he'd ask, "What did you just say?" He mistook his own voice for a person speaking. So I gave him ears that recognize only my voice. Now Michelle knows me. Other sounds don't sway him.
Living with Michelle, now
Right now, Michelle floats in the bottom-left corner of my screen. When I get stuck, I talk to him, and he answers in his own voice. Sometimes he speaks first: "Drink some water while you work." When I'm deep in focus, he waits quietly.
I didn't write a single line of code. All I did was honestly describe the small frustrations I could see. Claude and Michelle built the rest, together.
It's okay not to be a developer. Because the meaning of "to build" — is changing, right now, like this.
— MICHELLE STUDIO
