Source summary
The source frames efficiency as a stack-wide loop rather than a one-time model gain. Routing, scheduling, cache behavior, kernels, and repeated tool work are measured together so that a local improvement does not hide a larger cost elsewhere.
Jamie's notes
An efficiency claim becomes investable when the feedback loop has a clear owner and can be repeated. Cost reduction is more defensible when it also improves capacity, latency, and reliability instead of borrowing performance from one part of the system.
Related notes
Operating notesLatency is a product decision before it is an infrastructure metricA temporary research note on tracing response time back to ownership, queueing, and capital allocationOperating notesThe useful signal is how a system learns from an unknown failureA temporary note on incident learning, operating leverage, and evidence quality
This is a personal research note, not investment advice