AI 용어집
인공지능 완전 사전
Agent Chaining
Architecture where multiple specialized or non-specialized LLM agents are connected in sequence or parallel to decompose and solve a complex task into simpler sub-tasks.
Agent Orchestration
Process of managing, coordinating, and sequencing interactions between different agents within a multi-agent system, often supervised by a master agent or orchestrator.
Master Agent
Central agent responsible for overall planning, decomposing the main task into sub-tasks, and delegating these to appropriate specialized agents.
Specialist Agent
LLM agent designed and optimized to execute a specific type of sub-task (e.g., research, code analysis, writing) with dedicated expertise and tools.
Task Planning
Initial phase of chaining where the master agent analyzes the user's request to create a structured action plan, defining the order and nature of sub-tasks to be performed.
Problem Decomposition
Method of dividing a complex request into a series of simpler, autonomous sub-problems, each of which can be handled by an individual agent for better efficiency.
Result Synthesis
Final step where an agent (often the master) aggregates, analyzes, and formats the individual outputs of specialized agents to produce a coherent and complete final response.
Parallel Agent Execution
Orchestration strategy where multiple agents execute independent sub-tasks simultaneously to reduce overall resolution time, before their results are consolidated.
Agent Pipeline
A chaining configuration where agents are organized in a strict linear sequence, with each agent's output serving directly as input to the next in the chain.
Feedback Loop
Mechanism that allows an agent to send a result back to a previous agent for correction, iteration, or refinement, thereby improving the quality and precision of the final output.
Dynamic Routing
Ability of a system to direct a subtask to the most appropriate agent in real-time, based on its content, complexity, or current system state.
Agent State
Set of contextual information, current objectives, and memory data that an agent maintains throughout its execution to guide its actions and decisions.
Dependency Management
Process of managing the execution order of agents based on the prerequisites of each subtask, ensuring that an agent does not start before the necessary data is available.
Inter-Agent Negotiation
Advanced process where agents communicate to resolve conflicts, allocate resources, or define a common strategy, often used in decentralized systems.
Agent Supervision
Control and validation mechanism that monitors an agent's actions to ensure it complies with its objectives, constraints, and defined policies, able to intervene in case of deviation.
Multi-Agent System
Computing system composed of multiple intelligent agents that interact to achieve collective goals, with agent chaining being a specific architecture.