LLM agents
12 mentions across all digests
LLM agents are AI systems built on large language models that execute multi-step tasks using tools and dynamic decision-making, subject to security vulnerabilities like reasoning hijacking and studied extensively for architectural patterns by Anthropic.
Building effective agents
Anthropic distills lessons from dozens of teams into simple, composable agent patterns—prompt chaining, routing, subagent orchestration, evaluator loops—as a practical alternative to complex frameworks.
From Coarse to Fine: Self-Adaptive Hierarchical Planning for LLM Agents
Researchers introduce self-adaptive hierarchical planning for LLM agents, enabling coarse-to-fine refinement of action plans to improve complex task reasoning.
From Actions to Understanding: Conformal Interpretability of Temporal Concepts in LLM Agents
Researchers use conformal prediction to formally interpret how LLM agents understand and execute temporal actions, improving transparency into agentic behavior through statistical guarantees.
Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents
Unified framework consolidates memory, skills, and rules into a single knowledge spectrum, improving how LLM agents organize and apply different types of knowledge.
The cognitive companion: a lightweight parallel monitoring architecture for detecting and recovering from reasoning degradation in LLM agents
Researchers propose "cognitive companion," a lightweight parallel architecture that lets LLM agents autonomously detect reasoning degradation and self-correct in real-time, improving agentic system reliability without external oversight.