Is "Metadata-Driven" the only way to build true AI-Native apps? 🤔

Hey PH community, Jianguo here.

Over the past few months, my team and I have been building ObjectStack, and we hit a major realization about AI and software architecture.

80% of enterprise SaaS (like vertical CRMs or PSA tools) is just forms, tables, workflows, and permissions. When we try to let AI agents write raw code for these, they eventually break the logic or create a mess of unmaintainable legacy code.

We took a different approach: Everything is Metadata. By abstracting the entire application layer into structured metadata, the entire context stays extremely lightweight (well under 150k tokens). Now, our AI engine (ObjectOS) doesn't write code; it manipulates the metadata. The result? Instant application generation with strict data isolation and zero runtime syntax errors.

We are exploring this with a local-first SQLite setup for developers and cloud Postgres for production.

Are any other makers building with metadata architectures to make AI's life easier? Would love to connect and discuss!

P.S. If you want to see our engine in action: https://github.com/objectstack-ai/objectstack?utm_source=chatgpt.com