https://kirancodes.me/posts/log-distributed-llms.html
Article
-
Multi-agent coding systems face classic distributed systems problems: consensus, ordering, consistency
-
Agents lack shared state and reliable communication, mirroring distributed compute challenges
-
CAP theorem and Byzantine fault tolerance apply directly to agent coordination
Discussion
-
One commenter runs 3-4 agents with a supervisor agent; sees supervisor as bottleneck at scale
-
Counterargument: humans built Linux without solving these theoretically — iteration fixes inconsistency
-
Conway’s Law still applies; architecture and collaboration patterns don’t emerge from agent count alone
-
Skeptic notes no real model capability improvement in years, just better tooling
Discuss on HN