AI Agent
AI Agent — a system where an LLM autonomously performs multi-step tasks by: (1) reasoning about the goal, (2) calling tools (web search, code execution, API calls), (3) observing results, (4) iterating. Paradigms: ReAct (Reason + Act), Plan-Execute, Reflexion. 2026 frameworks: LangGraph, Claude Agent SDK, AutoGen, CrewAI. Use cases: research assistants, coding copilots, customer support.
Below: details, example, related terms, FAQ.
Details
- Tools: callable functions — web search, SQL query, file ops, API call
- Memory: short-term (scratchpad) + long-term (vector DB)
- Planning: decompose goal → subgoals → actions
- Loops: iterate until completion or max_iterations
- Frameworks: LangGraph (StateGraph), Claude Agent SDK, AutoGen multi-agent
Example
# Claude Agent SDK pattern
from anthropic import Anthropic
tools = [{'name': 'search_web', 'description': 'Search the web', 'input_schema': {...}}]
response = client.messages.create(
model='claude-opus-4-7',
tools=tools,
messages=[{'role':'user','content':'Find latest news on TLS 1.3 adoption'}],
max_tokens=4096
)
# Agent loop: if stop_reason='tool_use' → execute tool → send result backRelated Terms
Learn more
Frequently Asked Questions
Agent vs chatbot?
Chatbot: single-turn response. Agent: multi-step, uses tools, can hunt for information itself. Blurry boundary — modern chatbots increasingly use agentic patterns.
Reliability?
Agent loops can get lost. Best practice: max_iterations limit, human-in-loop for critical steps, observability via traces.
What is Claude Agent SDK?
Anthropic 2025 — SDK for building agents. Includes tools, retrieval, long context (1M tokens), cache. Tight integration with Claude Opus 4+.