The same slide editor concept built with Claude — completing the 3-way AI tool comparison experiment.
The third and final entry in the AI comparison series. Same specification, this time built with Claude. The codebase lands between the other two at 113 KB.
Claude's version includes a proper lib/ directory, comprehensive types/, and both test/ and tests/ directories — suggesting it prioritised architecture and testing. The shadcn/ui component library appears here too (like Codex, unlike Kimi).
What's fascinating is comparing the three side by side. Kimi produced the most code (227 KB), Codex was most concise (87 KB), and Claude fell in the middle but with the most structured architecture. Each AI's personality shows through in its code.
This experiment demonstrated that the choice of AI tool meaningfully affects code architecture, not just syntax. The same brief produces genuinely different solutions.