one of the really funny things about ai is how there was almost no alpha in being early. all that expertise prompt engineering? better instruction tuned models just do what you ask now. those carefully scaffolded rickety agentic flows in 2023? u were basically trying to force a squad of retarded children into flying a plane with really well written picture books. now u just have models that are smart enough to handle complex long-time-horizon tool use no problem. ur hypertuned RAG with perfect context selection and inscrutable chunking techniques? who cares man just throw it in gemini, or better yet throw a plain old keyword search tool at ur LLM and it'll find what it needs on its own.

the only thing that really mattered is if u built a customer base and a good product.