Materializer - is a set of conventions to establish a precise line-by-line binding between the specification you write - and the code AI generates. Iterate on specs free text - having AI mirror changes into the code seamlessly, rather than "ask" it to "make no mistakes" :) in countless prompts. It explores world where code is a material that is molded by AI on-demand, same way CPU-level bytecode is a material produced by compilers.

HiBird — Smarter communication for growing businesses.

Smarter communication for growing businesses.

After a couple of months of vibecoding, I realized it's so annoying to "onboard" the AI Agent into the project each time I want to make a change. A memory problem arose, and I was poking one solution after another.

I dreamed about a system where strong bindings between docs/schemas would serve both guardrails for the AI to cover everything but not to overcode, and persistent shared memory between humans and AIs where humans work (update specs) and AIs remember.

What I did is Materializer - a set of source map formats, AI prompts, and a VS Code Extension to allow zero-prompt code generation from specs and the other way, line-to-line VS Code hyperlinks, and git diff to apply code updates from spec updates deterministically.

I'm thrilled to see if this is something other people encountered too, and maybe to help solve this issue broader than for myself.