My guess is SRE culture is a tough sell at Anthropic. When you’re a frontier lab, almost everything else looks more prestigious and more immediately “impactful”.
Flatbuffers are fine - I think it is used in many places that needs zero-copy. Also outside google, it powers the Arrow format which is the foundation of modern analytics
Absolutely. LLM inference is still a greenfield — things like overlap scheduling and JIT CUDA kernels are very recent. We’re just getting started optimizing for modern LLM architectures, so cost/perf will keep improving fast.
From an enterprise adoption standpoint, remote MCP addresses the connector problem and can be easily retrofitted into enterprise-wide gateway services. In contrast, building tools is significantly more expensive for enterprises with large, existing API surfaces.
Most of the concerns can be addressed by a gateway service
many people especially those with chronic illnesses died because Zero-COVID blocked access to basic medical care and food. Those deaths were policy-driven and avoidable.
Zero-COVID was an absolute disaster. It involved severe human rights violations and caused immense social and economic damage, including unnecessary displacement, homelessness, and even deaths. The number is lower sure, but China had the capacity to do much better
I don’t blame the dev teams who ended up creating mini-frameworks. In many big tech companies, central framework teams are chronically underfunded. What I do blame are ego-driven mini-frameworks built by senior principal+ engineers that don’t actually solve real problems.
China has been a developing country for most the time of the past 25 years. It is indeed a huge problem if it is still rising rapidly. But it is also not fair to limit China’s per capita growth for most of the past two decades