I'm building Proteus, an open-source multimodal editor (think Figma meets Notion, but AI-native) where *AI writes most of the code* while I focus on architecture, technical decisions, and quality control.
*Why this matters:*
In 2025, tools like Cursor and Claude can write good enough code in 80% of scenarios. The question isn't "Can AI code?" but "What becomes valuable when AI can code?" I believe it's *system design, technical decision-making, and end-to-end ownership*—not just knowing APIs.
*What makes this different:*
- *AI-native from day one*: Every architectural decision prioritizes AI-friendliness. This isn't AI bolted on later—it's designed for AI collaboration from the first line.
- *Fully transparent*: All code, architecture decisions, and lessons learned are public. I'm documenting the entire journey in weekly technical articles.
- *Real editor, not a toy*: Phase 1 is complete with a working demo. You can create shapes, text, images, transform them, copy/paste, undo/redo—all the core editor capabilities.
- *Learning resource*: If you want to understand how editors work (scene graphs, rendering, interaction systems) or how to structure code for AI collaboration, this is a live case study.
*Current status:*
Phase 1: Core editing (scene graph, rendering, interaction, tools)
Phase 2: Multimodal elements (video, audio, web embeds)
Phase 3: AI Agent integration (natural language → editor actions)
Phase 4: Real-time collaboration
*The experiment:* What happens when you stop reviewing AI's code and instead focus entirely on architecture, problem diagnosis, and guiding AI through testing and context-building? That's what I'm exploring here.
Would love feedback from the HN community—especially from those building complex frontend apps or thinking about AI-native development workflows.
I'm building Proteus, an open-source multimodal editor (think Figma meets Notion, but AI-native) where *AI writes most of the code* while I focus on architecture, technical decisions, and quality control.
*Why this matters:*
In 2025, tools like Cursor and Claude can write good enough code in 80% of scenarios. The question isn't "Can AI code?" but "What becomes valuable when AI can code?" I believe it's *system design, technical decision-making, and end-to-end ownership*—not just knowing APIs.
*What makes this different:*
- *AI-native from day one*: Every architectural decision prioritizes AI-friendliness. This isn't AI bolted on later—it's designed for AI collaboration from the first line. - *Fully transparent*: All code, architecture decisions, and lessons learned are public. I'm documenting the entire journey in weekly technical articles. - *Real editor, not a toy*: Phase 1 is complete with a working demo. You can create shapes, text, images, transform them, copy/paste, undo/redo—all the core editor capabilities. - *Learning resource*: If you want to understand how editors work (scene graphs, rendering, interaction systems) or how to structure code for AI collaboration, this is a live case study.
*Current status:*
Phase 1: Core editing (scene graph, rendering, interaction, tools) Phase 2: Multimodal elements (video, audio, web embeds) Phase 3: AI Agent integration (natural language → editor actions) Phase 4: Real-time collaboration
*Try it:* [Live Demo](https://proteus.gezilinll.com/) *Code:* [GitHub](https://github.com/gezilinll/Proteus) *Articles:* [Tech Blog](https://github.com/gezilinll/Proteus/tree/main/articles) (4 articles so far, covering architecture, rendering, interaction design)
*The experiment:* What happens when you stop reviewing AI's code and instead focus entirely on architecture, problem diagnosis, and guiding AI through testing and context-building? That's what I'm exploring here.
Would love feedback from the HN community—especially from those building complex frontend apps or thinking about AI-native development workflows.