Papers · Liberalis

2026.07.09Independent research synthesis

Designing and Building Agentic Systems: A 2026 Survey of the State of the Practice

The hard part of building an agentic system has moved out of the model and into the system around it. This survey distills 118 authoritative sources, gathered and independently verified through 9 July 2026, into a single account of how practitioners and researchers actually design agents today: when to build one at all, how to engineer context, tools, memory, and retrieval, how to reason and plan across long horizons, when multi-agent architectures help versus hurt, how to make long-running agents reliable and durable, how to evaluate and secure them, and what separates deployments that ship from the 95 percent of pilots that do not. Every claim carries an inline citation to a source whose quoted wording was confirmed verbatim on the live page. The recurring finding is that context, architecture, and organization, not raw model intelligence, are the binding constraints.

2026.06.07Independent research synthesis

Understand or Predict: A Survey of World Models from Recurrent Controllers to the 2026 Frontier

A world model is a learned stand-in for an environment that an agent can run forward to ask what happens next. This survey organizes the field around a single split, drawn from Ding et al. (2024): a world model is built either to understand the world by holding a compact internal picture of how it works, or to predict the world by generating what it will look like next in full detail. Working from that axis, the paper walks the two branches through their main architecture families, model-based reinforcement learning and the recurrent world model, Dreamer and latent imagination, LeCun's joint-embedding predictive architecture, video-as-simulator diffusion models, and interactive and embodied 3D worlds, then covers how these systems are measured and where they still fail. The consistent finding from recent physics benchmarks is that looking right and being right are separate measurements, and the strongest 2025 to 2026 systems are converging on models that try to do both jobs at once.