[mvr123456]

This is going to happen and is real stuff we could be working towards with the tools that we already have. No need for AGI vaporware, no waiting around for the perfect agentic playground. Not even necessarily a big requirement for excellent reasoning. Just using LLMs for what they are actually good at, i.e. fuzzy translators, stylistic filters, and compilers.

Gradual-typing was practice and hints. Like gradual typing, gradual spec'ing could be an iterative and kind of parallel annotated representation, ignored by the main runtime unless called for, and ignored by developers that aren't interested in it. But when there's enough of it to hit some kind of critical mass, then it's suddenly very powerful and lots of very interesting stuff is possible