Show HN:This is how the real AI starts working and this is just the beginning (drive.google.com)

A Show HN submission reveals a proprietary AI engine proof-of-concept that the author says reframes “AI as a modeling problem” rather than another large language model. The demo claims the system is not LLM- or generative-AI-based: it only encodes basic English vocabulary and part-of-speech knowledge, accepts natural-language instructions directly (no API or generated code), and executes them—there’s no backend code, no application-specific APIs, and minimal UI that could be removed. The author argues that with enough instruction sets (experience), asking the engine to “build an order management system” would make it behave like traditional software without writing code. For the AI/ML community this is significant because it outlines an alternative path to automation: a semantic execution/runtime that maps natural language into persistent behavior instead of translating intent into code. Key technical implications include a heavier emphasis on formal modeling, knowledge representation, and deterministic execution of language-driven instructions; potential benefits are simplified integration and a human-centric interface. Major challenges remain—scaling the instruction/experience corpus, ensuring precise semantics, debugging, security, and interoperability with existing systems—so while provocative, the PoC highlights a viable research direction beyond purely generative LLM workflows rather than an immediate replacement for current engineering practices.