Agentic Patterns
AI 에이전트 실전 설계 패턴 모음
Context & Memory
- Context-Minimization Pattern
- Context Window Anxiety Management
- Dynamic Context Injection
- Filesystem-Based Agent State
- Layered Configuration Context
- Curated Code Context Window
- Episodic Memory Retrieval & Injection
- Memory Synthesis from Execution Logs
- Proactive Agent State Externalization
- Curated File Context Window
- Agent-Powered Codebase Q&A / Onboarding
Feedback Loops
- Background Agent with CI Feedback
- Dogfooding with Rapid Iteration
- Graph of Thoughts (GoT)
- Reflection Loop
- Spec-As-Test Feedback Loop
- Tool Use Incentivization via Reward Shaping
- Coding Agent CI Feedback Loop
- Inference-Healed Code Review Reward
- Rich Feedback Loops
- Self-Critique Evaluator Loop
- Self-Discover Reasoning Structures
- AI-Assisted Code Review / Verification
- Reflection Loop
Learning & Adaptation
- Agent Reinforcement Fine-Tuning
- Compounding Engineering Pattern
- Skill Library Evolution
- Variance-Based RL Sample Selection
Orchestration & Control
- Sub-Agent Spawning
- Action-Selector Pattern
- Autonomous Workflow Agent Architecture
- Continuous Autonomous Task Loop
- Distributed Execution with Cloud Workers
- Feature List as Immutable Contract
- Iterative Multi-Agent Brainstorming
- Multi-Model Orchestration for Complex Edits
- Plan-Then-Execute Pattern
- Self-Rewriting Meta-Prompt Loop
- Three-Stage Perception Architecture
- Agent-Driven Research
- Discrete Phase Separation
- Dual LLM Pattern
- Inference-Time Scaling
- Language Agent Tree Search (LATS)
- Opponent Processor / Multi-Agent Debate
- Progressive Autonomy with Model Evolution
- Specification-Driven Agent Development
- Tool Capability Compartmentalization
- Disposable Scaffolding Over Durable Features
- Explicit Posterior-Sampling Planner
- Initializer-Maintainer Dual Agent
- LLM Map-Reduce Pattern
- Oracle and Worker Multi-Model
- Progressive Complexity Escalation
- Stop Hook Auto-Continue Pattern
- Tree-of-Thought Reasoning
- Inversion of Control
- Parallel Tool Call Learning
- Swarm Migration Pattern
- Conditional Parallel Tool Execution
Reliability & Eval
- Anti-Reward-Hacking Grader Design
- CriticGPT-Style Evaluation
- Extended Coherence Work Sessions
- Lethal Trifecta Threat Model
- Merged Code + Language Skill Model
- RLAIF (Reinforcement Learning from AI Feedback)
- Structured Output Specification
- Versioned Constitution Governance
- Asynchronous Coding Agent Pipeline
- No-Token-Limit Magic
Security & Safety
- Deterministic Security Scanning Build Loop
- Egress Lockdown (No-Exfiltration Channel)
- Isolated VM per RL Rollout
- PII Tokenization
Tool Use & Environment
- CLI-First Skill Design
- Agent-First Tooling and Logging
- CLI-Native Agent Orchestration
- Code-Then-Execute Pattern
- Dual-Use Tool Design
- LLM-Friendly API Design
- Multi-Platform Communication Aggregation
- Patch Steering via Prompted Tool Selection
- Shell Command Contextualization
- Tool Use Steering via Prompting
- Agent SDK for Programmatic Control
- Code Mode MCP Tool Interface
- Dynamic Code Injection (On-Demand File Fetch)
- Progressive Tool Discovery
- Subagent Compilation Checker
- Virtual Machine Operator Agent
- Agentic Search Over Vector Embeddings
- Code-Over-API Pattern
- Visual AI Multimodal Integration
UX & Collaboration
- Proactive Trigger Vocabulary
- Abstracted Code Representation for Review
- AI-Accelerated Learning and Skill Development
- Chain-of-Thought Monitoring & Interruption
- Democratization of Tooling via Agents
- Human-in-the-Loop Approval
- Human-in-the-Loop Approval Framework
- Latent Demand Product Discovery
- Spectrum of Control / Blended Initiative
- Team-Shared Agent Configuration
- Agent-Assisted Scaffolding
- Seamless Background-to-Foreground Handoff
- Verbose Reasoning Transparency
- Agent-Friendly Workflow Design