A mechanism-first reading of why LLM-based agents need explicit architectures, communication rules, incentives, norms, trust models, and institutional supervision before they can become reliable business systems.
A case-first reading of SRA-CP, a risk-aware cooperative perception framework that treats vehicle communication as a scarce safety resource rather than a permanent data buffet.
A mechanism-first look at how Bayesian networks, neural text classifiers, virtual evidence, and consistency nodes can turn messy clinical notes into auditable probabilistic patient features.
A practical reading of HTAM and EarthAgent, showing why domain-structured agent hierarchies can outperform generic agent workflows in specialized planning tasks.
A mechanism-first reading of how formal BDI ontology can make AI agents more auditable, interoperable, and explainable without pretending that ontology magically solves reliability.
A practical reading of why LLM behaviour probes can look accurate in synthetic tests yet fail when the monitored behaviour depends on hidden generative context.
A mechanism-first reading of Mutual Intrinsic Reward, a simple reward-coupling idea for helping cooperative reinforcement-learning agents explore actions that change each other’s worlds.
A mechanism-first reading of CLOZE, a zero-shot pipeline for turning clinical notes into candidate Disease Ontology extensions without pretending that LLMs have become unsupervised medical librarians.
A mechanism-first reading of why trustworthy AI agents need lakehouse infrastructure built around branches, sandboxed compute, declarative I/O, and atomic merge.
PersonaDrift shows why long-term AI monitoring needs personalised baselines, anomaly-specific detectors, and calibration before it can credibly notice human change.