2026年AI办公助手介绍：从LLM到Agent的范式跃迁

 传统流程：纯手动操作
 1. 打开会议录音转文字工具 → 导出文本
 2. 复制粘贴到Word → 人工提取关键任务
 3. 打开Excel → 手动整理待办清单
 4. 打开邮箱 → 粘贴周报正文 → 选择收件人 → 发送
 整个过程涉及4+个软件，耗时约30分钟

 Agent工作流核心循环
class SimpleAgent:
    def __init__(self, llm_model, tools):
        self.llm = llm_model         大脑
        self.tools = tools           工具箱
        self.memory = []             记忆
        
    def run(self, user_query):
         1. 感知：理解用户意图
        task = self.llm.parse_intent(user_query)
        
         2. 规划：拆解任务步骤
        steps = self.llm.plan(task)
        
         3. 执行：逐步调用工具
        for step in steps:
            tool_name = step["tool"]
            params = step["params"]
            result = self.tools[tool_name](params)
            self.memory.append(result)
        
         4. 最终响应
        return self.llm.summarize(self.memory)

import os
import json
from pathlib import Path

 ========== 1. 定义工具集 ==========
class FileTools:
    @staticmethod
    def list_files(path: str) -> dict:
        """列出目录下所有文件"""
        files = [f for f in Path(path).iterdir() if f.is_file()]
        return {"files": [str(f) for f in files], "count": len(files)}
    
    @staticmethod
    def get_file_info(filepath: str) -> dict:
        """获取文件元信息"""
        p = Path(filepath)
        return {"name": p.name, "size": p.stat().st_size, "ext": p.suffix}
    
    @staticmethod
    def organize_by_ext(path: str) -> dict:
        """按扩展名整理文件"""
        moved = []
        for f in Path(path).iterdir():
            if f.is_file():
                ext = f.suffix[1:] if f.suffix else "others"
                target_dir = Path(path) / ext
                target_dir.mkdir(exist_ok=True)
                target_path = target_dir / f.name
                f.rename(target_path)
                moved.append(f"{f.name} → {ext}/")
        return {"moved": moved, "summary": f"已整理 {len(moved)} 个文件"}

 ========== 2. Agent核心 ==========
class OfficeAgent:
    def __init__(self, tools):
        self.tools = tools
         工具调用映射表
        self.tool_map = {
            "list_files": self.tools.list_files,
            "get_file_info": self.tools.get_file_info,
            "organize_by_ext": self.tools.organize_by_ext,
        }
    
    def reasoning_loop(self, instruction: str, context: dict) -> str:
        """简化的规划-执行循环"""
         Step 1: 意图解析（模拟LLM）
        if "整理" in instruction and "文件" in instruction:
            plan = [("organize_by_ext", {"path": context.get("path", "./")})]
        elif "查看" in instruction and "文件" in instruction:
            plan = [("list_files", {"path": context.get("path", "./")})]
        else:
            return "未能理解指令，请使用：整理文件 / 查看文件"
        
         Step 2: 执行计划
        for tool_name, params in plan:
            result = self.tool_map[tool_name](params)
            return json.dumps(result, ensure_ascii=False, indent=2)
        
        return "执行完成"

 ========== 3. 使用示例 ==========
agent = OfficeAgent(FileTools())

 示例：整理当前目录文件
result = agent.reasoning_loop("帮我整理这个文件夹里的所有文件", {"path": "."})
print(result)

 输出示例：
 {
   "moved": ["report.pdf → pdf/", "data.csv → csv/", "notes.txt → txt/"],
   "summary": "已整理 3 个文件"
 }

 模型输出的工具调用格式（伪代码）
{
  "tool": "send_email",
  "parameters": {"to": "boss@company.com", "subject": "Q2报告", "body": "..."}
}