Topics
Cross-cutting themes in LLM agent research
These themes run across the entire LLM agent field — they apply regardless of what kind of agent you’re building or what domain it operates in.
Evaluation & Benchmarks
How do we measure agent capability? SWE-bench, WebArena, GAIA, OSWorld, AgentBench, METR time horizons, and the ongoing challenge of benchmarks that don’t saturate or overfit.
Safety & Alignment
Prompt injection, trust hierarchies, sandboxing, reversibility, the transparency gap, and what responsible agent deployment looks like in practice.
Human-Agent Interaction & Trust
How humans work with autonomous agents — trust calibration, delegation patterns, UX of agentic systems, human-in-the-loop design, and the emerging world of ambient background agents.
Agent Economics
Cost per task, token efficiency, budget-constrained execution, model routing & cascading (FrugalGPT, RouteLLM), prompt compression, and enterprise ROI — the economics of making agents affordable at scale.
Long-Horizon Autonomy
Agents that work for hours, days, or indefinitely — error accumulation, memory architectures for sustained operation, hierarchical planning, METR time-horizon evaluations, and the autonomy spectrum.
Agent Societies & Simulation
Multi-agent worlds and emergent collective behavior — Generative Agents (Smallville), CAMEL, MetaGPT, ChatDev, Concordia, Project Sid, economic simulations, and the open question of whether LLM societies model human behavior.
Governance & Regulation
Legal liability, the EU AI Act, US and UK frameworks, industry self-governance (RSP, Preparedness Framework), cascading delegation, agent identity, and the race between regulation and capability.
Personalization & Digital Twins
How agents learn about users, adopt personas, and represent identities — plus digital twins, evaluation benchmarks (LaMP, PersonaGym), and adversarial attacks including DAN jailbreaks, memory poisoning, and system prompt leakage.
Cognitive Architectures
SOAR, ACT-R, LIDA, Global Workspace Theory, BDI — the cognitive science foundations being rediscovered in LLM agent design. What decades of architecture research teaches us about memory, attention, and planning.
Agents and Cybernetics
Norbert Wiener, feedback loops, Stafford Beer’s Viable System Model, autopoiesis (Maturana & Varela), the free energy principle, and self-evolving agents — the cybernetic lineage of intelligent agents.
Agents and Philosophy
Intentionality, the Chinese Room, consciousness, moral status, the extended mind, free will, and the philosophy of action — the deep philosophical questions raised by LLM agents.
These topics are connected. Evaluation shapes what safety problems we can measure. Safety constrains what autonomy levels are responsible. Governance determines what deployment looks like in practice.
Each page collects primary sources, key papers, and synthesis — use them as reference material rather than introductions.
Looking for agents by application area? See By Domain → · For technical architecture deep dives, see Deep Dives →
Social Intelligence & Human-AI Collaboration
Theory of mind in LLMs, the Kosinski–Ullman debate, social norm reasoning, AI-mediated group collaboration, human-AI teaming, and the foundational vision of human-computer symbiosis from Licklider and Engelbart.