The Genesis Architect's Matrix
A unified ontology of ... agentic techniques, collected from the frontier
of AI research and synthesized into a single cohesive architecture.
| Category | Technique | Description | Source |
|---|---|---|---|
| Agent Design | Role Prompting | "You are X" to set persona/style. | Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration |
| Agent Design | Style Prompting | "Write in format Y". | Style Transfer with Large Language Models |
| Agent Design | KABB | Knowledge-Action-Behavior-Brain layering. | KABB: Knowledge-Aware Bayesian Bandits for Dynamic Expert Coordination in Multi-Agent Systems |
| Agent Design | AgentArch | Layered Agent Architecture. | AgentArch: A Comprehensive Benchmark to Evaluate Agent Architectures in Enterprise |
| Agent Design | Mixture of Experts | Specialized prompts per expert type. | Mixture of Reasoning Experts |
| Agent Design | CAMEL | Role-Playing Cooperative Agents. | CAMEL: Communicative Agents for "Mind" Exploration of Large Scale Language Model Society |
| Agent Design | Agent-UML | Formal Spec for Agents. | Representing Agent Interaction Protocols in UML |
| Agent Design | COWPILOT | Cooperative Web Agents. | CowPilot: A Framework for Autonomous and Human-Agent Collaborative Web Navigation |
| Coordination & Signals | Musical Signals | Explicit Cue, Fermata, Caesura. | Harmonic Coordination Theory: A Musical Ontology for Autonomous Multi-Agent Systems |
| Coordination & Signals | Barrier Patterns | Wait for all agents to reach state. | [Distributed Systems Principles] |
| Coordination & Signals | Explicit Handoffs | Agent A explicitly triggers Agent B. | Orchestration Patterns for Agentic Workflows |
| Coordination & Signals | IoA (Internet of Agents) | Distributed heterogeneous agents. | Internet of Agents: Fundamentals, Applications, and Challenges |
| Coordination & Signals | A2A (Agent-to-Agent) | Standardized JSON communication protocol. | A2A: A Universal Protocol for Agentic Communication |
| Coordination & Signals | Supervisory Control | Hold for human approval. | Harmonic Coordination Theory: A Musical Ontology for Autonomous Multi-Agent Systems |
| Coordination & Signals | Test-in-the-Loop | Iterative Repair until Tests Pass. | A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications |
| Coordination & Signals | CONSENSAGENT | Mitigate sycophancy in voting. | CONSENSAGENT: Towards Efficient and Effective Consensus in Multi-Agent LLM Interactions |
| Coordination & Signals | BPMN Extension | Business Process for Agents. | Towards Modeling Human-Agentic Collaborative Workflows: A BPMN Extension |
| Economics & Cost | FrugalGPT | Cascade: Cheap -> Expensive models. | FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance |
| Economics & Cost | Model Routing | Pick model based on query difficulty. | RouteLLM: Learning to Route LLMs with Preference Data |
| Economics & Cost | Token Budgeting | Hard limits on conversation cost. | [Genesis Framework] |
| Economics & Cost | Semantic Caching | Cache similar queries to save inference. | GPTCache: Semantic Cache for LLM Queries |
| Economics & Cost | FinOps for AI | Track cost per user/session. | State of AI Engineering 2024 |
| Evolution & Learning | ACE | Agentic Context Engineering. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Evolution & Learning | Voyager | Lifelong Learning (Code + Resume). | Voyager: An Open-Ended Embodied Agent with Large Language Models |
| Evolution & Learning | SFT / RLHF | Fine-tuning models on best paths. | Training language models to follow instructions with human feedback |
| Evolution & Learning | A/B Testing | Experimenting with prompts/models. | TensorZero: Production LLM Engineering |
| Evolution & Learning | Meta Prompting | LLM improves its own prompt. | Meta-Prompting: Enhancing Language Models with Task-Agnostic Scaffolding |
| Evolution & Learning | Software Repair | Agents autonomously fixing code. | A Survey of Automated Program Repair |
| Evolution & Learning | Agentic Repair | Autonomous Agents fixing code. | A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications |
| Evolution & Learning | RAG-Based Repair | Retrieval-Augmented Bug Fixing. | A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications |
| Evolution & Learning | Vulnerability Repair | Fixing CVEs/Bugs specifically. | A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications |
| Inference & Execution | POMDP Modeling | Treat LLM apps as optimization problems. | TensorZero Blog: Think of LLM Applications as POMDPs |
| Inference & Execution | Inference-Time Optimization | Dynamic in-context learning / search. | TensorZero Documentation |
| Inference & Execution | Prompt Optimization (MIPRO) | Automated prompt engineering. | DSPy: Compiling Declarative Language Model Calls |
| Inference & Execution | ReAct | Reason-Act-Observe loop. | ReAct: Synergizing Reasoning and Acting in Language Models |
| Inference & Execution | Toolformer | Self-supervised tool use. | Toolformer: Language Models Can Teach Themselves to Use Tools |
| Inference & Execution | Dynamic Tool Selection | Filter tools to reduce confusion. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Inference & Execution | Model Context Protocol (MCP) | Universal Agent-Tool Protocol. | Internet of Agents: A Modular Framework for Heterogeneous Agents |
| Inference & Execution | Arbitrage Routing | Real-time cost/quality routing. | [Intelligence Broker] |
| Inference & Execution | LLMLingua | Small model compresses prompt. | LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models |
| Inference & Execution | Few-Shot (ICL) | Provide K examples (Task, Solution). | Language Models are Few-Shot Learners |
| Inference & Execution | Emotion Prompting | "This is critical for my career". | EmotionPrompt: Leveraging Psychology for Large Language Models Enhancement via Emotional Stimulus |
| Memory & Context | Basic RAG | Retrieve -> Generate. | Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks |
| Memory & Context | MemGPT | OS-style memory paging. | MemGPT: Towards LLMs as Operating Systems |
| Memory & Context | Memory-of-Thought | Store CoT in memory. | Memory-of-Thought: Mining and Utilizing Thoughts for Large Language Models |
| Memory & Context | Graph RAG | Knowledge Graph Retrieval. | From Local to Global: A Graph RAG Approach to Query-Focused Summarization |
| Memory & Context | Semantic Chunking | Split by topic/meaning shifts. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Memory & Context | Hierarchical Chunking | Summaries -> Sections -> Details. | Tree of Thoughts: Deliberate Problem Solving with Large Language Models |
| Memory & Context | Late Chunking | Embed full doc first, then chunk. | Late Chunking: Contextual Chunk Embeddings |
| Memory & Context | Agentic Chunking | Agent selects chunking strategy. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Memory & Context | Context Summarization | Compress history into summaries. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Memory & Context | SelectiveContext | Filter tokens via entropy. | Unlocking Context Constraints of LLMs |
| Memory & Context | Context Pruning | Actively remove irrelevant info. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Memory & Context | Query Rewriting | Rewrite -> Retrieve -> Read. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Memory & Context | Query Expansion | Generate multiple related queries. | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Memory & Context | Dynamic Query Construction | Build structured queries (SQL/Graph). | Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models |
| Observability | Distributed Tracing | Track request through agent graph. | LangFuse |
| Observability | LLM-as-a-Judge | Use strong model to grade weak model. | Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena |
| Observability | RAGAS | RAG Assessment metrics. | RAGAS: Automated Evaluation of Retrieval Augmented Generation |
| Observability | Graph Visualization | View agent interactions as DAG. | LangGraph Studio |
| Planning & Score | Least-to-Most | Break down complex tasks into sub-questions. | Least-to-Most Prompting Enables Complex Reasoning in Large Language Models |
| Planning & Score | Chain-of-Thought (CoT) | "Think step by step". | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
| Planning & Score | Tree-of-Thoughts (ToT) | Search/Backtrack multiple branches. | Tree of Thoughts: Deliberate Problem Solving with Large Language Models |
| Planning & Score | Graph-of-Thought (GoT) | Non-linear reasoning DAG. | Graph of Thoughts: Solving Elaborate Problems with Large Language Models |
| Planning & Score | ReWOO | Decouple reasoning from execution. | ReWOO: Decoupling Reasoning from Observations for Efficient Augmented Language Models |
| Planning & Score | Step-Back Prompting | Ask high-level concept first. | Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models |
| Planning & Score | Skeleton-of-Thought | Parallel sub-generation. | Skeleton-of-Thought: Prompting LLMs for Efficient Parallel Generation |
| Quality & Feedback | SVRL Loop | Solution-Verification-Refinement Loop. | Harmonic Coordination Theory: A Musical Ontology for Autonomous Multi-Agent Systems |
| Quality & Feedback | Self-Refine | Iterative "Critique -> Improve". | Self-Refine: Iterative Refinement with Self-Feedback |
| Quality & Feedback | Chain-of-Verification (COVE) | Generate verification questions. | Chain-of-Verification Reduces Hallucination in Large Language Models |
| Quality & Feedback | Metacognitive Prompting | Mirror human metacognition. | Metacognitive Prompting Improves Understanding in Large Language Models |
| Quality & Feedback | DeepSeekMath-V2 | Meta-Verification Paradigm. | DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning |
| Quality & Feedback | HAL (Holistic Agent Leaderboard) | Comprehensive Agent Evaluation. | Holistic Agent Leaderboard: The Missing Infrastructure for AI Agent Evaluation |
| Quality & Feedback | Self-Calibration | Ask "Are you sure?". | Language Models (Mostly) Know What They Know |
| Reference & Ontology | Reference Frame | Immutable shared ontology all agents respect. | Harmonic Coordination Theory: A Musical Ontology for Autonomous Multi-Agent Systems (§4.1) |
| Reference & Ontology | Constitutional AI | Principles for ethical behavior. | Constitutional AI: Harmlessness from AI Feedback |
| Reference & Ontology | Rule-Based Constraints | Hard constraints on agent actions. | Auto-GPT Safety |
| Reference & Ontology | Value Locking | Immutable core values. | Coherent Extrapolated Volition |
| Security & Trust | Prompt Injection Defense | RAG-based attack detection. | Securing AI Agents Against Prompt Injection Attacks |
| Security & Trust | Jailbreak Detection | Classifier for unsafe outputs. | Jailbroken: How Does LLM Safety Training Fail? |
| Security & Trust | Input Guardrails | Filter user input before LLM. | Llama Guard: LLM-based Input-Output Safeguard |
| Security & Trust | Output Filtering | Block unsafe generations. | NVIDIA NeMo Guardrails |
| Security & Trust | RbAC / ABAC | Role/Attribute-based Access Control. | Zero Trust Architecture for AI |
| Security & Trust | PII Redaction | Mask sensitive data. | Microsoft Presidio |