Sources

Sources that catalogue, classify, or survey agent / multi-agent patterns. Useful background for our four-domain taxonomy (Thinking / Flow / Collaboration / System Operation).

• Agent Design Pattern Catalogue: A Collection of Architectural Patterns for Foundation Model based Agents

  Liu et al., Journal of Systems and Software (2024) · Peer-reviewed catalogue of 18 architectural patterns grouped by lifecycle phase (goal creation, reflection, collaboration, accountability). The most authoritative academic catalogue of agent patterns as a concept. arXiv ↗︎ Journal version (Elsevier) ↗︎

• Multi-Agent Collaboration Mechanisms: A Survey of LLMs

  Tran et al. (2025) · Comprehensive survey of coordination patterns; directly underpins our Collaboration domain. ↗︎

• A survey on LLM-based multi-agent systems: workflow, infrastructure, and challenges

  Springer Vicinagearth (2024) · Three-pillar taxonomy; their workflow maps to our Flow and their infrastructure to our System Operation. ↗︎

• A Taxonomy of Hierarchical Multi-Agent Systems: Design Patterns, Coordination Mechanisms, and Industrial Applications

  arXiv (2025) · Five-axis taxonomy (control hierarchy, information flow, role/task delegation, temporal hierarchy, communication structure) — orthogonal drill-down into the Collaboration domain. ↗︎

• Large Language Model Agent: A Survey on Methodology, Applications and Challenges

  arXiv (2025) · Methodology-centred taxonomy organised around construction, collaboration, and evolution. ↗︎

• Microsoft Azure — AI Agent Orchestration Patterns

  Microsoft · Vendor-published practitioner taxonomy; splits single-agent vs. multi-agent then sub-patterns. ↗︎

• Google Cloud — Choose a design pattern for your agentic AI system

  Google · Decision-tree-style pattern guide; closer in spirit to our webapp Decision Guide than to the taxonomy itself. ↗︎

• Databricks — Agent system design patterns

  Databricks · Sequential, parallel, supervisor, hierarchical — practitioner-grade patterns covering Flow + Collaboration. ↗︎

• Agent design patterns

  Lance Martin / LangChain (blog, 2026) · Recent practitioner synthesis from a LangChain maintainer. ↗︎

• Agentic Design Patterns

  Antonio Gulli (book; Google engineer; ~400 pp; 21 patterns, MCP excluded) · Practitioner catalogue — each pattern with TL;DR, use, pros/cons, examples. Largely re-frames patterns this project already catalogues across Domains 1–4; primary source for the Exploration & Discovery and Resource-Aware Optimization entries.

• Mark Kashef — "Master ALL 20 Agentic AI Design Patterns"

  Mark Kashef (video, 2024) · Hour-long plain-language summary of Gulli's book; secondary aid. ↗︎

• Top Agentic AI Frameworks

  Omdena (blog, 2024) · Practitioner survey of leading agentic frameworks with use-case guidance. ↗︎

• Agentic AI Framework Comparison

  Moxo (blog, 2025) · Independent benchmark comparing latency, token usage, and developer experience across frameworks including LangGraph, CrewAI, and AutoGen. ↗︎

Sources that frame agent systems along an autonomy axis (workflow → agent; how much the LLM decides on its own) and a control axis (centralized → distributed coordination). Underpins the 2-axis pattern map and any framing that positions systems on a spectrum.

• Building Effective Agents

  Anthropic (2024) · Anchors the spectrum: "A system is more agentic the more an LLM decides how the system can behave." Workflows (predefined code paths) vs. agents (LLM dynamically directing). ↗︎

• What is an AI agent?

  LangChain blog · Proposes agent autonomy as a spectrum; explicitly draws the SAE autonomous-vehicle L0–L5 analogy. ↗︎

• How to think about agent frameworks

  LangChain blog · Uses the autonomy axis to compare frameworks; closest external precedent for ranking systems along it on a chart. ↗︎

• Workflows and agents

  LangChain docs · Formalises the workflow ↔ agent distinction and where Router / Tool-calling / ReAct sit on the spectrum. ↗︎

• Multi-Agent Collaboration Mechanisms: A Survey of LLMs

  Tran et al. (2025) · Explicit centralized / decentralized / hybrid taxonomy of coordination structures. Already cited in Pattern Taxonomies and Surveys. ↗︎

• What Is Agentic Architecture?

  IBM · Practitioner-grade explanation of centralized vs. distributed agent coordination. ↗︎

• Single-agent and multi-agent architectures

  Microsoft Dynamics 365 docs · Vendor mapping of supervisor / decentralized / hybrid models. ↗︎

• Centralized vs. Distributed Agent Collaboration Models

  Auxiliobits (practitioner blog) · Side-by-side trade-off summary. ↗︎

• Building Effective Agents

  Anthropic (2024) · Workflows vs. agents and the canonical composable patterns (prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer). ↗︎

• A Practical Guide to Building Agents

  OpenAI (2024) · Pragmatic introduction to agent design choices, tools, and guardrails (PDF). ↗︎

• How we built our multi-agent research system

  Anthropic Engineering (2025) · Production lessons on orchestrator-worker MAS, parallelism, and reliability. ↗︎

• Don't Build Multi-Agents

  Cognition AI (2024, opinion) · Counter-position arguing single-agent loops often beat MAS; useful for the anti-pattern section. ↗︎

• Agentic Design Patterns, Part 1: Reflection

  Andrew Ng / DeepLearning.AI (2024, blog) · Practitioner overview of the reflection / self-critique loop as a foundational agentic pattern. ↗︎

• Agentic Design Patterns: A System-Theoretic Framework

  Dao et al. (2025) · Twelve-pattern system-theoretic framework (Integrator, Retriever, Recorder, Selector, Planner, Deliberator, Executor, Tool Use, Coordinator, Reflector, Skill Build, Controller); primary source for Part VI of the knowledge base.

• Understanding the Agentic Reasoning Loop

  Digital Bricks (2026, blog) · Accessible vendor-blog restatement of the four-phase reasoning loop (Think / Act / Observe / Reflect) and supporting framing for Part VI's runtime-loop preamble. Also references Microsoft Copilot Studio's "enhanced task completion" as a productized form of the loop. Higher-level than the Dao et al. framework above; useful as introductory reading. ↗︎

Primary papers behind the Domain-1 thinking patterns.

• ReAct: Synergizing Reasoning and Acting in Language Models

  Yao et al. (2022) · Origin of interleaved reasoning + tool calls. ↗︎

• Toolformer: Language Models Can Teach Themselves to Use Tools

  Schick et al. (2023) · A model learns which API to call, when, with what arguments, and how to fold the result back in; foundational source for the Function Calling pattern. ↗︎

• Gorilla: Large Language Model Connected with Massive APIs

  Patil et al. (2023) · Selecting the right call from a large, changing catalogue of APIs; source for the Tool Registry and Capability Routing patterns. ↗︎

• Self-Consistency Improves Chain-of-Thought Reasoning in Language Models

  Wang et al. (2022) · Sample-and-vote aggregation over CoT. ↗︎

• Plan-and-Solve Prompting

  Wang et al. (2023) · Two-phase plan-then-execute prompting baseline. ↗︎

• Understanding the Planning of LLM Agents: A Survey

  Huang et al. (2024) · Survey of planning approaches in LLM agents; primary reference for the Plan-and-Execute pattern. ↗︎

• ReWOO: Decoupling Reasoning from Observations

  Xu et al. (2023) · Plans tool calls up front to cut token cost. ↗︎

• Reflexion: Language Agents with Verbal Reinforcement Learning

  Shinn et al. (2023) · Self-critique loop for iterative improvement. ↗︎

• Tree of Thoughts: Deliberate Problem Solving with LLMs

  Yao et al. (2023) · Search over branching reasoning paths. ↗︎

• Executable Code Actions Elicit Better LLM Agents (CodeAct)

  Wang et al. (2024) · Code as the action interface for agents. ↗︎

• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

  Wei et al. (2022) · Antecedent for most "thinking" patterns. ↗︎

• Designing LLM Chains by Adapting Techniques from Crowdsourcing Workflows

  Grunde-McLaughlin et al. (2025) · Practitioner-grade reference for the Sequential Pipeline (prompt chaining) pattern. ↗︎

• PromptChainer: Chaining Large Language Model Prompts through Visual Programming

  Wu et al. (2022) · Visual-programming approach to composing LLM prompt chains; referenced in the L2 rung. ↗︎

• AI Chains: Transparent and Controllable Human-AI Interaction by Chaining LLM Prompts

  Wu et al. (2022) · Foundational prompt-chaining paper showing transparency and control benefits; referenced in the L2 rung. ↗︎

• DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines

  Khattab et al. (2023) · Automated optimization of LLM prompt pipelines; referenced in the L2 rung. ↗︎

• Self-Refine: Iterative Refinement with Self-Feedback

  Madaan et al. (2023) · Self-critique and iterative improvement loop; primary source for the Evaluator-Optimizer and Iterative Refinement patterns. ↗︎

• From Decoding to Meta-Generation: Inference-time Algorithms for Large Language Models

  Welleck et al. (2024) · Survey of inference-time strategies including sampling, search, and self-correction; referenced in the L2 rung. ↗︎

• FlowForge: Guiding the Creation of Multi-agent Workflows

  Hao et al. · Workflow-design tool cited under Parallelization (Sectioning / Voting).

• A Universal Modular Actor Formalism for Artificial Intelligence

  Hewitt, Bishop & Steiger (1973) · Introduced the Actor model of concurrent computation: independent agents that communicate solely by asynchronous message passing; foundational precursor to modern multi-agent message-based coordination.

• A Blackboard Architecture for Control

  Barbara Hayes-Roth (1985) · Historical root of shared-state coordination between specialists. ↗︎

• Improving Factuality and Reasoning via Multiagent Debate

  Du et al. (2023) · Foundational paper for the Multi-Agent Debate pattern. ↗︎

• ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate

  Chan et al. (2023) · Debate-as-evaluation method; cited under the Multi-Agent Debate pattern. ↗︎

• Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate

  Liang et al. (2023) · Shows that societally diverse agent personas improve debate quality; referenced in the L3 rung. ↗︎

• Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks

  Fourney et al. / Microsoft Research (2024) · Composite orchestration with planning ledger and replanning. ↗︎

• GPTSwarm: Language Agents as Optimizable Graphs

  Zhuge et al. (2024) · Agent coordination modelled as executable, learnable state graphs; cited under Graph-based Orchestration. ↗︎

• Blackboard Systems

  Engelmore & Morgan (eds.) (1988) · Canonical edited volume on the Blackboard architectural pattern; primary textbook reference for shared-workspace coordination cited in the L3 rung.

• The Contract Net Protocol: High-Level Communication and Control in a Distributed Problem Solver

  Smith (1980) · Historical root of market-based task allocation between agents; origin of the Contract Net coordination pattern.

• AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

  Wu et al. (2023) · Conversation-centric MAS framework paper. ↗︎

• MetaGPT: Meta Programming for a Multi-Agent Collaborative Framework

  Hong et al. (2023) · Role-based crew metaphor / SOPs for agent teams. ↗︎

• The Anatomy of Autonomy: Why Agents are the Next AI Killer App after ChatGPT

  Lilian Weng / OpenAI blog (2023) · Influential write-up of the agent loop (planning, memory, tool use). ↗︎

• An Introduction to MultiAgent Systems, 2nd Edition

  Michael Wooldridge (Wiley, 2009) · Standard academic textbook on MAS; covers coordination, communication, and cooperation in classical (pre-LLM) terms. Author page ↗︎ Publisher ↗︎

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (RAG)

  Lewis et al. (2020) · Origin of RAG as a memory pattern. ↗︎

• Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG

  Singh et al. (2025) · Survey of RAG where the agent decides whether, when, and how to retrieve, reformulates queries, and grades results. Primary source for the Agentic RAG pattern (Tool Integration). ↗︎

• Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

  Asai et al. (2023) · Trains a model to decide when to retrieve and to self-critique retrieved passages; close relative of Agentic RAG. ↗︎

• Corrective Retrieval Augmented Generation (CRAG)

  Yan et al. (2024) · Adds a lightweight retrieval evaluator that scores evidence and triggers correction (re-retrieval, web fallback); cited as an Agentic RAG variant. ↗︎

• From Local to Global: A Graph RAG Approach to Query-Focused Summarization (GraphRAG)

  Edge et al. / Microsoft (2024) · Builds a hierarchical knowledge graph for multi-hop, corpus-spanning questions; cited as an Agentic RAG variant. ↗︎

• RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval

  Sarthi et al. (2024) · Recursively clusters and summarises chunks into a multi-level retrieval tree; cited as an Agentic RAG variant. ↗︎

• Beyond Vector Search: 5 Next-Gen RAG Retrieval Strategies

  Machine Learning Mastery (2026, practitioner guide) · Survey of GraphRAG, Agentic RAG, Self-RAG, CRAG, and RAPTOR; source for the Agentic RAG named-variants note. ↗︎

• 7 Steps to Mastering Memory in Agentic AI Systems

  Machine Learning Mastery (2026, practitioner guide) · The RAG-vs-agent-memory distinction (read-only/content vs read-write/user) and storage-technology selection; source for the Memory Architecture preamble and storage-selection note. ↗︎

• MemGPT: Towards LLMs as Operating Systems

  Packer et al. (2023) · Hierarchical memory for long-running agents; primary source for the Virtual Context Management pattern. ↗︎

• Cognitive Architectures for Language Agents (CoALA)

  Sumers et al. (2023) · Formalises agent memory as short-term working memory plus long-term episodic, semantic, and procedural stores; the taxonomy underlying the Memory Architecture subdomain. ↗︎

• Generative Agents: Interactive Simulacra of Human Behavior

  Park et al. (2023) · Natural-language memory stream retrieved by a recency + importance + relevance score, with reflection; primary source for the Episodic Memory pattern. ↗︎

• Voyager: An Open-Ended Embodied Agent with Large Language Models

  Wang et al. (2023) · An ever-growing skill library of executable code stored and retrieved for reuse; primary source for the Skill Build (procedural memory) pattern. ↗︎

• Agent memory framework landscape (2026)

  Machine Learning Mastery (2026, practitioner survey) · Practitioner survey of dedicated memory layers: Letta (the MemGPT lineage), Mem0, Zep, Cognee (knowledge-graph memory), plus the LangChain and LlamaIndex memory modules. Source for the concrete-framework mentions in the Memory Architecture subdomain. ↗︎

• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena

  Zheng et al. (2023) · Reference paper for LLM-as-Judge evaluation; documents position and verbosity bias in model judges. ↗︎

• G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment

  Liu et al. (2023) · Chain-of-thought + form-filling judging paradigm; also documents the judge's bias toward LLM-generated text. ↗︎

• LLM Evaluators Recognize and Favor Their Own Generations

  Panickssery et al. (2024) · Self-preference bias: an LLM judge scores its own outputs higher than humans do, correlated with self-recognition. ↗︎

• Ragas — Evaluation framework for LLM applications

  Ragas · Reference-free RAG/agent metrics (faithfulness, context precision/recall, answer relevancy). ↗︎

• DeepEval — LLM evaluation framework

  Confident AI · Pytest-native, CI/CD-oriented evaluation with a broad metric library spanning RAG and agents. ↗︎

• LangGraph — Low-level concepts (State, Nodes, Edges, Conditional Edges)

  LangChain AI · Canonical vocabulary for state-driven agent graphs. ↗︎

• LangGraph — Human-in-the-Loop (interrupt / resume / Command)

  LangChain AI · First-class halt points and resumability. ↗︎

• LangGraph — Persistence & Checkpointing

  LangChain AI · Shared state across nodes and runs. ↗︎

• LangGraph — Multi-agent supervisor

  LangChain AI · Canonical supervisor pattern implementation. ↗︎

• LangGraph — Hierarchical agent teams

  LangChain AI · Reference for the Hierarchical Supervisor pattern. ↗︎

• LangGraph — Swarm / handoffs

  LangChain AI · Decentralized handoff coordination. ↗︎

• LangGraph — Recursion limit

  LangChain AI · Bounds for unbounded loops. ↗︎

• LangGraph — PostgresSaver checkpointer

  LangChain AI · Production-grade checkpointer; replaces SqliteSaver under concurrency. ↗︎

• LangChain — Frameworks, runtimes, and harnesses

  LangChain AI · Three-layer stack model (Framework / Runtime / Harness) for categorizing agent tooling. Vendor-origin and explicitly exploratory — the companion blog post concedes the categories are fuzzy; the project frames it as a contested lens (see KB Part III). Primary source for the stack-layer concept used in Block 4 and Cross-Cutting View 3. ↗︎

• Agent Frameworks, Runtimes, and Harnesses, oh my!

  LangChain blog (Harrison Chase) · The companion post that introduces the split and openly admits "there is still murkiness and overlap." Cited as evidence the taxonomy is exploratory, not settled. ↗︎

• The AI Agent Stack in 2026

  The Nuanced Perspective (A. Reganti) · Counter-grouping that classifies CrewAI, AutoGen, Google ADK and the OpenAI Agents SDK as runtimes, using a bipartite runtime/harness split. Cited as evidence the framework↔runtime boundary is contested. ↗︎

• LangGraph — GitHub repository

  LangChain AI · Source code and issue tracker. ↗︎

• LangGraph: Building Stateful, Multi-Actor Applications with LLMs

  LangChain AI (blog, 2024) · Original launch post introducing LangGraph's stateful graph model; referenced in the L4 rung. ↗︎

• AutoGen — GroupChat / message-passing design pattern

  Microsoft · Conversation-first coordination model. ↗︎

• AutoGen documentation home

  Microsoft · Framework concepts, agents, and orchestration. ↗︎

• AG2 (community fork of AutoGen)

  AG2 maintainers · Active community fork referenced in framework comparison. ↗︎

• AG2 — GitHub repository

  AG2 maintainers · Source code and issue tracker. ↗︎

• CrewAI — Flows (state, listeners, routers)

  CrewAI · Flow-based analogue to the graph vocabulary. ↗︎

• CrewAI — Crews and roles

  CrewAI · Role-based "crew" abstraction. ↗︎

• CrewAI — GitHub repository

  CrewAI · Source code and issue tracker. ↗︎

• Google ADK — Workflow agents (Sequential / Parallel / Loop)

  Google · Composable workflow primitives. ↗︎

• Google ADK — Overview

  Google · Apache 2.0 agent dev kit, Gemini-first but model-agnostic. ↗︎

• Google ADK — GitHub repository

  Google · Source code and issue tracker. ↗︎

• AWS Strands Agents — Introducing post

  AWS Open Source Blog (2025) · Model-driven loop, three-line agent example. ↗︎

• Strands Agents documentation

  AWS · Concepts, model-driven loop, AWS integrations. ↗︎

• Strands Agents — GitHub repository

  AWS · Source code and issue tracker. ↗︎

• OpenAI Agents SDK

  OpenAI · Successor to the experimental Swarm framework; handoff primitives. ↗︎

• OpenAI Agents SDK — GitHub repository

  OpenAI · Source code and issue tracker. ↗︎

• OpenAI Swarm (experimental, archived)

  OpenAI · Original handoff-based swarm reference implementation. ↗︎

• Pydantic AI — Documentation home

  Pydantic / Samuel Colvin · Concepts, agents, RunContext dependency injection, structured output, and Logfire integration. ↗︎

• Pydantic AI — GitHub repository

  Pydantic · Source code and issue tracker. ↗︎

• Pydantic AI — Harness overview

  Pydantic · The optional capability library ("the batteries for your agent"): context management, guardrails, filesystem, code execution, multi-agent orchestration, shipped as a separate package whose capabilities graduate into core once stable. Cited in Part III (contested-lens note) and Part VII as evidence the "harness" label is overloaded. ↗︎

• LlamaIndex — Documentation home

  LlamaIndex · Data connectors, indexing, query engines, and agent workflows; hub for the LlamaHub connector ecosystem. ↗︎

• LlamaIndex — GitHub repository

  LlamaIndex · Source code and issue tracker. ↗︎

• Semantic Kernel — Documentation home

  Microsoft · Skills (plugins), planners, multi-language SDK (C#/Python/Java), and Azure-aligned governance. ↗︎

• Microsoft Agent Framework — Overview

  Microsoft · Unified successor to Semantic Kernel and AutoGen; converging roadmap for enterprise agentic systems. ↗︎

• Semantic Kernel — GitHub repository

  Microsoft · Source code and issue tracker. ↗︎

• LangChain4j — Documentation home

  LangChain4j maintainers · Java/Kotlin LLM abstractions (AiServices, chains, tools, RAG), Spring Boot and Quarkus integrations, and MCP module. ↗︎

• LangChain4j — GitHub repository

  LangChain4j maintainers · Source code and issue tracker. ↗︎

• Model Context Protocol (MCP) — Specification

  Anthropic / MCP working group · Open standard for agent ↔ tool integration. ↗︎

• Function calling — OpenAI docs

  OpenAI · Reference for the tool-calling primitive shared across providers. ↗︎

• A2A Protocol — Announcing the Agent2Agent Protocol

  Google (2025) · Open standard for agent ↔ agent communication. ↗︎

• A2A Protocol — Specification

  A2A maintainers · Agent card, message exchange, framework neutrality. ↗︎

• Was ist das Agent2Agent-(A2A-)Protokoll?

  Caballar & Stryker / IBM Think (2025) · German-language IBM Think explainer covering both MCP and A2A; cited under MCP in the Tool Integration subdomain.

• ACP (Agent Communication Protocol) — merged into A2A

  IBM / BeeAI · The former REST-native agent-to-agent messaging specification has merged into A2A under the Linux Foundation; the welcome page carries the notice and migration guide. ↗︎

• AGNTCY (Agent Connect) — Project home

  AGNTCY collective (Cisco) · Directories, identity, and discovery for cross-organization agent cooperation. ↗︎

• ANP (Agent Network Protocol) — Project home

  open source · Decentralized agent identity and discovery on W3C DIDs. ↗︎

• AG-UI — Documentation

  CopilotKit · Event-streamed protocol standardizing agent-to-front-end communication. ↗︎

• AP2 — Protocol home

  Google · 60+ partners · Agent-initiated payments using signed user mandates, built on A2A. ↗︎

• Agentic Commerce — Project home

  OpenAI · Stripe · Standardized programmatic checkout for agent-driven purchases. ↗︎

• x402 — Coinbase Developer Platform

  Coinbase · Revives HTTP 402 for per-call stablecoin payments between agents and APIs. ↗︎

• MapReduce: Simplified Data Processing on Large Clusters

  Dean & Ghemawat (2004) · Foundational distributed-processing model underlying parallelization patterns; referenced in the L2 and L4 rungs. ↗︎

• Time, Clocks, and the Ordering of Events in a Distributed System

  Lamport (1978) · Established logical clocks and causal ordering as the formal basis for reasoning about event sequences in distributed systems; referenced in the L4 rung on runtime coordination and trace stitching.

• Sagas

  Garcia-Molina & Salem (1987) · Long-lived transaction pattern for distributed systems; foundational reference for durable execution and compensation chains in agentic runtimes. ↗︎

• Dapper, a Large-Scale Distributed Systems Tracing Infrastructure

  Sigelman et al. / Google (2010) · Foundational distributed-tracing architecture underlying modern agent observability stacks; referenced in the L4 rung. ↗︎

• LangSmith — Tracing for LLM applications

  LangChain AI · Distributed tracing and token cost tracking. ↗︎

• OpenTelemetry — GenAI semantic conventions

  OpenTelemetry · Cross-vendor observability for LLM / agent calls; standard gen_ai.* span attributes. ↗︎

• Arize Phoenix — Open-source AI observability & evaluation

  Arize AI · OpenTelemetry/OpenInference tracing paired with LLM-based evaluations over the captured traces. ↗︎

• LangGraph — Durable execution & resumability

  LangChain AI · Checkpoint-based recovery for long-running runs. ↗︎

• Actor Model of Computation: Scalable Robust Information Systems

  Hewitt (2010) · Actors as the universal primitives of concurrent computation — private state, message passing, no shared memory; foundational source for the Actor Model pattern. ↗︎

• Enterprise Integration Patterns

  Hohpe & Woolf (2003) · The canonical catalogue of messaging patterns (publish-subscribe channel, event-driven consumer, choreography); source for the Event-driven Choreography and Pub/Sub Agent Mesh patterns. ↗︎

• OWASP Top 10 for Large Language Model Applications

  OWASP (2025) · Canonical catalogue of LLM and agent security risks: prompt injection, data and model poisoning, vector and embedding weaknesses, excessive agency. Underpins Part IX §7 (Cascading Security Vulnerabilities). ↗︎

• Implementing Statistical Guardrails for Non-Deterministic Agents

  Machine Learning Mastery (2026, practitioner guide) · Semantic-drift detection via cosine-distance z-scores and confidence gating via Shannon entropy; primary source for the Statistical Guardrails pattern (Governance & Safety). ↗︎

• The Protection of Information in Computer Systems

  Saltzer & Schroeder (1975) · Origin of the principle of least privilege and separation of privilege; foundational source for the Least Privilege Agent pattern. ↗︎

• NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Rails

  Rebedea et al. (2023) · Programmable input/dialog/output rails independent of the underlying LLM; source for the guardrail patterns. ↗︎

• Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations

  Inan et al. (2023) · A safety classifier applying a risk taxonomy to both prompts and responses; source for the Multimodal Guardrails pattern. ↗︎

• Defeating Nondeterminism in LLM Inference

  Thinking Machines Lab (2025) · Most authoritative recent source on LLM non-determinism: shows that outputs vary even at temperature=0 because of batch-invariance failure in inference kernels, not floating-point folklore. ↗︎

• Non-Determinism of "Deterministic" LLM Settings

  arXiv 2408.04667 (2024) · Empirical study quantifying nondeterminism across providers and settings. ↗︎

• Be Clear, Direct, and Detailed

  Anthropic prompt-engineering guide · Frames prompts as instructions requiring engineering rigor, not natural-language requests. ↗︎

• Define your success criteria

  Anthropic empirical-evaluation guide · Frames LLM verification as an open eval-design problem. ↗︎

• OpenAI — Temperature and sampling

  OpenAI · Reference for non-determinism even at temperature=0. ↗︎

• Moffatt v. Air Canada (BCCRT, Feb 2024)

  Civil Resolution Tribunal of British Columbia (2024) · Civil Resolution Tribunal of British Columbia held Air Canada liable for negligent misrepresentation by its chatbot, which had hallucinated a bereavement-fare refund policy. Awarded CAD $812.02. McCarthy Tétrault legal analysis ↗︎ ABA Business Law Today ↗︎ CanLII case comment (2025) ↗︎

• Chevrolet of Watsonville chatbot — "$1 Tahoe" (Dec 2023)

  Business Insider coverage (2023) · Customer prompt-injected a dealership's GPT-powered assistant into agreeing to sell a 2024 Chevy Tahoe for $1 with the binding response "That's a legally binding offer — no takesies backsies." ↗︎

• Mata v. Avianca (SDNY, June 2023)

  Southern District of New York (2023) · Two lawyers were sanctioned after submitting a federal court filing in which ChatGPT had hallucinated six fictional case citations. Court order ↗︎ NYT coverage ↗︎

• What is the Agent2Agent (A2A) Protocol?

  Caballar & Stryker / IBM Think (2025) · IBM Think article covering the A2A protocol.

• Agent2Agent Protocol Specification

  Linux Foundation · Agent2Agent Protocol Specification under the Linux Foundation.

• Model Context Protocol Specification

  Anthropic · Model Context Protocol Specification from Anthropic.

• Pydantic v2 documentation

  Pydantic · Pydantic v2 documentation. ↗︎

• LangGraph Persistence (PostgresSaver)

  LangChain AI · LangGraph Persistence documentation covering PostgresSaver. ↗︎

• Langfuse documentation

  Langfuse · Langfuse documentation. ↗︎

• Small Language Models are the Future of Agentic AI

  Belcak et al. / NVIDIA Research (2025) · Argues SLMs are the right default for the repetitive, well-scoped calls that dominate agentic workloads; primary source for Model Tiering and Small Language Models (Part VIII). ↗︎

• Introduction to Small Language Models: The Complete Guide for 2026

  Machine Learning Mastery (2026, practitioner guide) · SLM definition (<10B params), the SLM-vs-LLM routing heuristic, and efficiency methods (distillation, quantization, sparse attention). ↗︎

• 5 Production Scaling Challenges for Agentic AI in 2026

  Machine Learning Mastery (2026, practitioner guide) · Orchestration, observability, cost (kill switches, retry storms), evaluation (behavioural / simulation testing), and governance as the five production gaps; supports Part VIII and Part IX. ↗︎

• 7 Agentic AI Trends to Watch in 2026

  Machine Learning Mastery (2026, practitioner guide) · Graduated / bounded autonomy, governance and security agents, and FinOps model tiering; supports the HITL graduated-autonomy framing (Part V §6) and Model Tiering (Part VIII). ↗︎

• Temporal — Documentation

  Temporal Technologies · Durable workflow engine documentation; covers deterministic replay, automatic retries, and saga support for long-running stateful orchestration. docs ↗︎ repo ↗︎

• Inngest — Documentation

  Inngest Inc. · Event-driven durable execution documentation; step functions, retries, and flow control on a managed event backbone without running your own queue. docs ↗︎ repo ↗︎

• Restate — Documentation

  Restate Systems · Durable execution documentation for distributed services; covers durable promises, virtual objects, and exactly-once handlers across microservices. docs ↗︎ repo ↗︎

• Deep Agents SDK — Documentation

  LangChain AI · Opinionated, batteries-included harness on LangGraph; documents built-in planning, subagents, and virtual file system for long-horizon autonomous agents. docs ↗︎ repo ↗︎

• Claude Agent SDK — Documentation

  Anthropic · Agentic harness documentation; covers the gather-context / act / verify loop behind Claude Code, with tools, subagents, and first-class MCP support. docs ↗︎ repo ↗︎