Faster inference won't save you (graphcoder.ai)

The latency table says it all. Cloud-to-cloud is 40ms for 20 turns. Hotel Wi-Fi is 16 seconds. You can halve inference time and still have a broken product on bad connections.

does this mean you can disconnect from the internet entirely with the agent loop still running?