Comment by ammmir

honest question: what use case requires cold starts below 100ms, considering TTFT of major LLMs are in the 300+ms range? presumably sandbox will be driven by an agentic loop, so.. you’re still bottlenecked by what essentially amounts to network I/O.