History
No explorations yet
What exactly is an AI agent, how is it different from a chatbot, and why did protocols like MCP and A2A emerge?
Explore AI Agent through an interactive 3D knowledge map. This visual guide covers 10 key concepts and 11 relationships, helping you understand the topic structurally.
An autonomous system that uses LLMs to plan, use tools, and complete tasks
Unlike simple chatbots, AI agents set goals, create plans, call external tools, and verify results. The LLM acts as the 'brain,' combined with tool-calling, memory, and planning capabilities to perform complex tasks autonomously.
The core brain of an agent — GPT, Claude, Gemini, etc.
LLMs serve as the reasoning engine for agents. They interpret user intent, decide which tools to use, and synthesize results. The model's capability sets the upper bound for the agent's performance.
The ability for agents to call external APIs, databases, and file systems
Also known as Function Calling. When an LLM outputs structured function calls instead of text, the system executes them and feeds results back to the LLM. This is what makes agents 'act' rather than just 'talk.'
Anthropic's open standard for connecting agents to tools
MCP is an open protocol that lets AI agents access diverse data sources and tools in a standardized way. Like USB-C for AI — one standard to connect everything. It uses a server-client architecture.
Google's protocol for inter-agent communication
A2A enables different agents to collaborate. If MCP is agent-to-tool, A2A is agent-to-agent. Agents can delegate tasks to each other and exchange results.
Multiple agents collaborating by dividing roles
Complex tasks that are hard for a single agent are split among specialized agents. A manager agent distributes work, and each agent contributes to the final output. CrewAI and AutoGen are notable frameworks.
Reasoning + Acting — a think-then-act loop
Agents repeat 'Thought → Action → Observation' cycles to solve problems. More capable than simple prompt chaining, enabling complex reasoning and tool use in tandem.
Mechanisms for maintaining conversation context and long-term recall
Split into short-term memory (current conversation context) and long-term memory (past information stored in vector DBs). Without memory, agents start from scratch every time.
Key differences between traditional chatbots and AI agents
Chatbots only answer questions, but agents act autonomously toward goals. Chatbots generate one response at a time; agents create multi-step plans and execute them.
Coding agents — AI that reads, edits, and runs code
Representative examples of AI agents integrated into developer tools. They understand codebases, find bugs, and implement features, accessing file systems, terminals, and Git via MCP.