I have been building Research automation with LangGraph for the past 2 months. We always put a human in the loop checkpoint after each critical step, might be annoying now but I think it will save us long-term.
the implementation we are building is open source: https://github.com/giatenica/gia-agentic-short-v2