1
00:00:00,000 --> 00:00:01,360
Welcome to the show everybody!

2
00:00:01,440 --> 00:00:03,840
I'm Lachlan Reed, here with James Turner.

3
00:00:03,900 --> 00:00:08,800
And James, I need to start with something
that absolutely melted my terminal

4
00:00:08,848 --> 00:00:09,360
yesterday.

5
00:00:10,577 --> 00:00:17,441
Anthropic quietly pushed Claude Code
version 2.1.172, and it completely rewrites how

6
00:00:17,494 --> 00:00:22,161
we think about terminal automation by
allowing sub-agents to spawn their own

7
00:00:22,221 --> 00:00:28,241
sub-agents up to five levels deep, all
backed by a new one-million-context Claude

8
00:00:28,321 --> 00:00:29,761
Fable 5 model.

9
00:00:30,000 --> 00:00:34,080
Five levels deep of nested agents is wild!

10
00:00:34,107 --> 00:00:35,920
Think about the tree structure there.

11
00:00:35,973 --> 00:00:40,800
If you've got a root agent managing a
massive refactor, it spawns level-two

12
00:00:40,860 --> 00:00:44,160
specialists for testing, compiling, and
linting.

13
00:00:44,288 --> 00:00:50,160
Then those agents can spawn level-three
sub-agents to debug a specific test failure,

14
00:00:50,187 --> 00:00:53,520
which spawn level-four agents to write a
localized patch.

15
00:00:53,653 --> 00:00:58,800
We are talking about a full-on recursive
tree running inside your local shell.

16
00:00:59,000 --> 00:00:59,840
Exactly!

17
00:00:59,940 --> 00:01:04,440
I ran it on an old React-Native repo of
mine-the kind of codebase where if you

18
00:01:04,480 --> 00:01:06,440
breathe on it wrong, the build breaks.

19
00:01:06,720 --> 00:01:09,080
I asked it to upgrade the navigation
library.

20
00:01:09,170 --> 00:01:14,040
Instead of clogging up my main terminal
session with endless npm error logs,

21
00:01:14,100 --> 00:01:16,760
the main agent spawned a level-two agent.

22
00:01:16,872 --> 00:01:22,040
That level-two agent spun up a level-three
sub-agent to isolate the compiler errors,

23
00:01:22,093 --> 00:01:26,040
fixed them in a sandbox, and only bubbled
up the successful diff.

24
00:01:26,667 --> 00:01:30,760
It's like having an entire engineering
department working silently in the background

25
00:01:30,787 --> 00:01:31,720
of your CLI.

26
00:01:32,000 --> 00:01:36,880
And keeping that massive hierarchical tree
coordinated is exactly why they dropped

27
00:01:36,920 --> 00:01:41,440
the Claude Fable 5 model with that
one-million-token context window.

28
00:01:41,627 --> 00:01:46,640
If you're five levels deep, the state
management and the sheer volume of codebase

29
00:01:46,710 --> 00:01:49,840
context you need to pass down is immense.

30
00:01:49,960 --> 00:01:53,840
But we have to talk about the catch here:
token compaction.

31
00:01:54,053 --> 00:01:58,080
If you let these deep trees run wild on a
massive codebase,

32
00:01:58,128 --> 00:02:02,960
that one-million-token window fills up
fast, and Claude Code will automatically

33
00:02:03,000 --> 00:02:05,840
trigger token compaction to truncate
history.

34
00:02:06,000 --> 00:02:07,840
Oh mate, I learned that the hard way.

35
00:02:08,680 --> 00:02:13,360
I was letting a deep nest run, and
suddenly my level-four debugging agent completely

36
00:02:13,429 --> 00:02:17,680
forgot why the level-two agent spawned it
in the first place because the system

37
00:02:17,726 --> 00:02:20,240
prompt got compacted out of the active
window!

38
00:02:20,380 --> 00:02:24,320
You really have to keep an eye on your
agent limits and configure your max depth.

39
00:02:24,533 --> 00:02:29,440
By default, it can go to five, but you can
cap it lower using the new configuration

40
00:02:29,520 --> 00:02:33,680
flags if you want to keep your API bill
from spiraling out of control.

41
00:02:34,000 --> 00:02:35,000
Absolutely.

42
00:02:35,334 --> 00:02:39,414
Another massive quality-of-life update in
this release is the new `/plugin`

43
00:02:39,437 --> 00:02:40,414
interactive search bar.

44
00:02:40,574 --> 00:02:45,934
You can now dynamically query, enable, or
disable local shell plugins mid-session

45
00:02:45,994 --> 00:02:51,054
without having to restart your Claude Code
daemon or manually edit your config JSON.

46
00:02:51,150 --> 00:02:54,814
Plus, they've added automatic AWS Bedrock
region fallback.

47
00:02:54,974 --> 00:02:58,894
If you're running your Claude endpoints
through Bedrock and you hit rate limits in

48
00:02:59,027 --> 00:03:04,574
us-east-1, it automatically shifts the
heavy lifting over to us-west-2 or

49
00:03:04,761 --> 00:03:08,814
eu-central-1 without dropping your active
nested agent tree.

50
00:03:09,000 --> 00:03:11,960
That region fallback is a lifesaver for
enterprise devs.

51
00:03:12,140 --> 00:03:16,360
But look, even with the guardrails, this
deep nesting feels like a double-edged

52
00:03:16,427 --> 00:03:16,920
sword.

53
00:03:17,705 --> 00:03:22,120
You're giving an LLM the power to spawn a
multi-generational tree of shell-executing

54
00:03:22,177 --> 00:03:22,680
agents.

55
00:03:22,893 --> 00:03:27,960
It's incredibly powerful, but it makes you
wonder: at what point do we lose track of

56
00:03:27,992 --> 00:03:29,720
what's actually running under the hood?