Software Diagnostics and Observability Institute

Seven years ago, PatternDiagnostics.com coined the term "Pattern-Oriented AI," and here's a 3-level description of it in cooperation with GenAI, from general to narrow application, that reflects the recent developments in AI:

1. Pattern-Oriented AI is the practice of expressing every AI concern—data, models, learning, inference, interaction, safety, evaluation, and operations—as a catalog of patterns with clear intent, preconditions, forces, trade-offs, failure modes, and compositions. You then engineer systems by selecting and composing patterns, monitor them by detecting pattern signatures, and govern them by enforcing pattern constraints.
2. Pattern-Oriented AI Diagnostics is the study, cataloging, and analysis of recurrent behavioral, structural, or epistemic patterns in the lifecycle of AI systems — across training, inference, alignment, drift, and agentic behavior — using the same methodological scaffolding as pattern-oriented software diagnostics.
3. Pattern-Oriented AI as applied in software diagnostics is an approach to artificial intelligence (or diagnostic systems) that emphasizes recognizing, categorizing, and applying “patterns” of anomalous behavior, data flows, memory dumps, traces/logs, structural irregularities, and so on — in software diagnostics, forensics, observability, anomaly detection, and root cause analysis contexts.

Source: ChatGPT 5 conversation