Research

Papers and frameworks on agent architecture — the thesis the Manning book makes the engineering case for, in formal form. The central claim: agent architecture is bounded resource allocation under uncertainty. The model spends; the harness budgets.

Frameworks

• v1 · 2026 The Dual-Axis Framework for AI Agent Design Patterns

  Two-dimensional classification of agent architectures along Cognitive Function (7 categories) × Execution Topology (6 archetypes). The 7×6 matrix yields 27 distinct agent design patterns, 13 of them introduced and named in the work. Five empirical laws of pattern selection close the framework.

• v1 · 2026 Pattern Evolution: GoF → Distributed → Agent

  Interactive lineage of 48 software design patterns — the 23 GoF object patterns (1994) and 25 distributed-systems patterns (2000s–2020s) — traced into the Agent era. Each pattern is clickable; the panel reveals original problem, default assumption, and Agent-era counterpart. The thesis: patterns did not become obsolete, they were lifted one layer.

• v1 · 2026 Engineering Ancestors of Agent Design Patterns

  A hoverable map from 28 Agent design patterns to their software-engineering ancestors: cache hierarchy, priority queues, Pipes-and-Filters, Command, workflow DAGs, durable execution, Event Sourcing, RBAC, Saga, MapReduce, OpenTelemetry, and more. Each pattern shows both the Agent-era intent and the older engineering idea it inherits.

• v1 · 2026 Harness Anatomy: cross-cutting view of 8 agent harnesses

  Eight production agent harnesses (Claude Code, Codex CLI, Aider, OpenCode, OpenClaw, Hermes Agent, DeerFlow, OpenHands) analyzed across the 7 cognitive functions. The output is a coverage heatmap built by a fixed five-step method (Detect → Classify → Filter → Map → Verify). Each cell is anchored to a file path and a line range. The list is open — the same method extends to any new harness.

• v1 · 2026 Paradigm Evolution: Structured Programming → Agent

  One layer above patterns: paradigms. An interactive timeline across six eras and fifteen major paradigms — Structured Programming → OOP / Pattern Language / GoF → Distributed / CAP / REST / DDD / Event Sourcing → Microservices / Cloud Native / Reactive → Data-Intensive Thinking → LLM / Agent Architecture / Harness Engineering. Each card opens a three-part panel — what it solved, the key insight, and the shift it produced. Only paradigms that changed engineering vocabulary at the CAP / Microservices level are admitted.

Reading Maps

• v1 · 2026 Classic Technical Books and Papers: Defining Insights

  A public reading map for technical works that changed how a field thinks: the problem each book or paper answered, the defining insight, the reversal it introduced, and why the idea travelled. This page is deliberately not an Agent framework; it is a way to study what makes an idea field-defining.

• v1 · 2026 Long-Lived Thinking Principles

  A cross-era map of durable human thinking moves: Socratic questioning, first-principles reasoning, wu wei, Occam's razor, methodic doubt, falsifiability, systems thinking, the Pyramid Principle, deliberate practice, dark time, and more. Each entry is framed as a problem, a principle, a reversal, and a practice question.

Papers

• May 15, 2026 A Two-Dimensional Framework for AI Agent Design Patterns: Cognitive Function × Execution Topology

  Huang & Zhou, 2026. arXiv preprint arXiv:2605.13850 (cs.AI) · archival DOI 10.5281/zenodo.19036557. Coauthor: Joey Tianyi Zhou (A*STAR CFAR). We argue that neither a topology-only nor a cognition-only axis is sufficient to distinguish architecturally different agent systems, and propose the two-dimensional taxonomy that organizes 27 patterns and yields five empirical pattern-selection laws.

Citations and further readings live in Designing AI Agents (Manning, MEAP launches May 26, 2026).