Sasisundar09·geçen ay·discussReliability and observability seem harder than execution itself once agents are in production. How are you detecting silent tool failures today?
Sasisundar09·geçen ay·discussCurious how you are handling benchmark reliability. Have you seen cases where evaluations pass but production behavior fails?
Sasisundar09·geçen ay·discussModular hardware is a great concept, but local model development lives and dies by memory bandwidth. Apple’s unified memory architecture is a physical moat that current modular buses cannot bridge.