1
00:00:00,000 --> 00:00:01,600
LangChain 刚发了一篇文章

2
00:00:01,600 --> 00:00:04,400
Agent 架构，正在走向异步

3
00:00:04,500 --> 00:00:06,200
让 Agent 干个大活

4
00:00:06,200 --> 00:00:07,400
它就埋头跑

5
00:00:07,400 --> 00:00:09,400
你再说什么，它都听不见

6
00:00:09,500 --> 00:00:11,600
因为老架构只有一条循环

7
00:00:11,600 --> 00:00:13,400
一个慢工具没回来

8
00:00:13,400 --> 00:00:15,400
后面的事全得排队

9
00:00:15,500 --> 00:00:17,600
新做法，是把 Agent 拆成两半

10
00:00:17,600 --> 00:00:19,600
前台一个，专门陪你说话

11
00:00:19,600 --> 00:00:21,400
后台一群，专门干活

12
00:00:21,400 --> 00:00:23,400
干完自己回来汇报

13
00:00:23,500 --> 00:00:24,900
中途改主意也行

14
00:00:24,900 --> 00:00:26,400
说「换成 Postgres」

15
00:00:26,400 --> 00:00:27,600
它打断重来

16
00:00:27,600 --> 00:00:29,300
说「做完再跑个测试」

17
00:00:29,300 --> 00:00:30,400
它排在后面

18
00:00:30,500 --> 00:00:32,400
所以 Agent 能干多少活

19
00:00:32,400 --> 00:00:34,200
不再看一条循环多长

20
00:00:34,200 --> 00:00:35,900
看后台有几个帮手

21
00:00:36,000 --> 00:00:38,300
Claude Code、OpenAI 的 dots

22
00:00:38,300 --> 00:00:40,400
DeepSeek Harness，都已经这么干了

23
00:00:40,500 --> 00:00:42,900
你的 Agent，还在一条道跑到黑吗？

24
00:00:42,900 --> 00:00:44,500
评论区说