@dair_ai
Pay attention to this one if you build research or knowledge-work agents. Most research-agent systems produce uniform outputs regardless of who is driving them. This new work, NanoResearch, argues that personalization is a precondition for real usability, and proposes tri-level co-evolution as the architecture. Three layers run together: a skill bank that distills recurring operations into reusable procedural rules; a memory module that retains user- and project-specific experience across sessions; and label-free policy learning that converts free-form feedback into persistent planner updates. Reliable skills produce richer memory, richer memory informs better planning, and preference internalization continuously realigns the loop. The framework consistently beats SOTA research systems and progressively produces better outputs at lower cost across cycles. The skill / memory / policy co-evolution loop is reusable far beyond paper writing. It is the template for any long-lived assistant in coding, analytics, or research. Paper: https://t.co/00qy1aBWqI Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c