What if open-source AI focused less on giant models and more on reusable capabilities?

Instead of everyone building another general-purpose model, the community could distill open models into domain specialists—biology, Python, accounting, OCR, and more. Developers could combine these capabilities into local tools: small model + OCR + accounting → local accounting assistant Like Linux, open-source AI could grow through shared components rather than complete systems. Could domain capabilities become the fundamental unit of contribution?