Reaction
Deep Agents harness moves context out of the active window without deleting it
LangChain says agents take in more instructions, tool results, files and drafts as they run, reducing accuracy and reasoning.
TLDR
LangChain describes its Deep Agents harness as moving context out of an agent's active window without deleting it. The company says this addresses a problem with longer-running agents: as they take in more instructions, tool results, files and drafts, their accuracy and reasoning decline.
Combined views
12.7K
2 Sources, first seen ago
