17h agoResearcher Pushes Model-Harness Co-Design for LLM AgentsSentimentSentimentPos50%Neg50%Positive users agree model-harness co-design is key for LLM agents and highlight bash-based re-in-act approaches, while negative users call ReAct beating CodeAct a costly mistake due to API costs and design flaws.2 comments with sentiment. View comments.