Two new research papers published on arXiv explore different applications of large language model (LLM) agent systems, addressing memory persistence and automated scientific workflows.
According to arxiv.org, researchers presented PrimeAgentOrchestrator (PAO), a system that spawns new instances of Claude Code—Anthropic’s terminal-based coding agent—pre-loaded with relevant memories from user databases. The system queries two memory backends in parallel (a PostgreSQL entity-observation database and a Cloudflare Worker semantic search index) and delivers compiled briefings through filesystem injection. The paper reports on four months of deployment from December 2025 through March 2026, documenting “three generations of context delivery mechanisms, the failure modes that motivated each redesign, and the engineering tradeoffs of bridging heterogeneous memory systems.”
Separately, arxiv.org reports on MAESTRO, an LLM agent system designed for computational materials discovery. According to the paper, MAESTRO executes “the entire screening pipeline for metal-organic frameworks (MOFs),” processing MOF literature, linking publications to crystal structures, and creating a computation-ready database. The system screens candidates “through a strategy of progressively increasing computational cost.” According to the researchers, promising candidates identified for separation under wet flue gas conditions “all originate from unrelated studies.”
Both papers were published on August 24, 2026, with initial submissions dated May 8, 2026 and August 20, 2026 respectively.