auto-research style proposal loops should be data driven!
they largely work best only when Data/Evals/Feedback give a useful gradient to hill-climb against
increasingly auto-research is a very good tool for general Agent Optimization
the LangChain ecosystem is here to help running these data driven self-improvement loops with easy tooling so teams can focus on their problems: - a customizable agent harness in deepagents or create_agent - support for any model provider (Open/Closed/Local models) - BYO tools, prompts, skills, you name it - tooling for developing evals, running them, tracing them, and understanding them at scale in OpenEvals & LangSmith
getting started today means getting an earlier feel for the data your specific use-case needs to power the loop
🧠Self-Harness: Harnesses that improve themselves
New paper on agents shaping their own harnesses to improve over time. Not from LangChain, but builds on top of DeepAgents! Three key steps:
1/ Weakness mining: find failure modes from traces 2/ Harness proposal: suggest changes to the harness 3/ Proposal validation: does regression testing on proposals and then accepts
Paper: https://arxiv.org/pdf/2606.09498
Builds on top of DeepAgents: https://github.com/langchain-ai/deepagents
