LangChain, OpenAI, Anthropic, MIT, Microsoft, and Claude Code all point in the same direction: agents are becoming engineered loops with memory, stop rules, observability, identity, and measured work