1
00:00:00,420 --> 00:00:03,680
You know that feeling when you're in the
terminal, you kick off a massive

2
00:00:03,760 --> 00:00:08,160
codebase-wide refactor with an AI
assistant, and then...

3
00:00:08,200 --> 00:00:09,040
you just sit there?

4
00:00:09,184 --> 00:00:13,120
Your terminal is completely locked up, and
you're just staring at a spinning cursor

5
00:00:13,190 --> 00:00:17,280
because you needed to run a single, quick
git query, but you can't.

6
00:00:17,772 --> 00:00:19,252
Oh, absolutely mate.

7
00:00:19,319 --> 00:00:22,972
It's the modern developer's equivalent of
"my code is compiling."

8
00:00:23,072 --> 00:00:26,732
You just sort of wander off to make a
coffee while your shell is held hostage.

9
00:00:26,750 --> 00:00:27,390
Exactly.

10
00:00:27,422 --> 00:00:31,390
Well, before we solve that terminal
gridlock, today's episode is brought to you by

11
00:00:31,483 --> 00:00:36,430
Jellypod AI, the tool that lets you turn
your daily newsletters and developer feeds

12
00:00:36,542 --> 00:00:39,950
into personalized, high-quality audio
podcasts.

13
00:00:40,163 --> 00:00:44,270
No more staring at screens to catch up on
tech news; you can listen on your morning

14
00:00:44,320 --> 00:00:46,270
commute or while washing the dishes.

15
00:00:46,377 --> 00:00:49,630
Check out Jellypod AI to reclaim your
reading time.

16
00:00:50,019 --> 00:00:50,185
Brilliant.

17
00:00:50,405 --> 00:00:54,745
And speaking of reclaiming time, let's
talk about the July 17,

18
00:00:54,825 --> 00:00:59,625
2026 release of Claude Code v2.1.212.

19
00:00:59,721 --> 00:01:03,945
Anthropic dropped this update, and it
basically blows up that whole "waiting for the

20
00:01:03,989 --> 00:01:05,145
terminal" bottleneck.

21
00:01:05,278 --> 00:01:07,545
It's all about multi-tasking now.

22
00:01:07,675 --> 00:01:11,062
Right, because they completely overhauled
how background jobs work.

23
00:01:11,249 --> 00:01:17,542
It's this new, very deliberate split
between two commands: /fork and /subtask.

24
00:01:17,622 --> 00:01:22,182
Previously, it was a bit muddy, but now
they have completely different architectural

25
00:01:22,262 --> 00:01:22,902
jobs, right?

26
00:01:22,917 --> 00:01:23,477
Spot on.

27
00:01:23,504 --> 00:01:25,237
So, /fork is the big one here.

28
00:01:25,333 --> 00:01:30,437
What it does now is it literally clones
your exact active conversation state-all the

29
00:01:30,507 --> 00:01:34,757
context, the code history, the files
you've been working on-and it pushes all of

30
00:01:34,805 --> 00:01:38,917
that into a completely detached,
independent background session.

31
00:01:38,997 --> 00:01:42,437
Like, fork now copies your conversation
into a new background session.

32
00:01:42,597 --> 00:01:46,677
It gets its own row in your claude agents
dashboard, and you are instantly handed

33
00:01:46,725 --> 00:01:49,317
back your primary terminal shell to keep
typing.

34
00:01:49,450 --> 00:01:51,637
It's a proper background branch.

35
00:01:51,625 --> 00:01:55,945
Wait, so the key detail is that /fork
actually duplicates the current state?

36
00:01:56,078 --> 00:01:59,945
It's not just starting a fresh agent from
scratch in the background; it knows

37
00:02:00,015 --> 00:02:04,265
exactly where I am in my debugging
session, but lets me run away and do other things

38
00:02:04,292 --> 00:02:05,065
in the foreground?

39
00:02:05,083 --> 00:02:05,563
Yes!

40
00:02:05,643 --> 00:02:06,283
Exactly.

41
00:02:06,416 --> 00:02:09,563
It's like git branching for your live AI
context.

42
00:02:09,683 --> 00:02:14,123
Whereas /subtask is the opposite-it's for
inline delegation.

43
00:02:14,336 --> 00:02:18,043
If you're working and you just need a
quick, nested helper job done right there

44
00:02:18,073 --> 00:02:21,323
without losing your spot, you use
/subtask.

45
00:02:21,510 --> 00:02:24,603
It replaces that old, blocking inline
behavior.

46
00:02:24,870 --> 00:02:29,643
So /fork is "go run this massive refactor
in the background while I work,"

47
00:02:29,683 --> 00:02:33,883
and /subtask is "hey, quickly look up this
library doc for me right here."

48
00:02:34,019 --> 00:02:35,315
Okay, that makes sense.

49
00:02:35,535 --> 00:02:38,995
But if you have these powerful agents
running in the background,

50
00:02:39,048 --> 00:02:42,755
fully detached, my immediate anxiety is...

51
00:02:42,787 --> 00:02:44,675
what if it gets stuck in a recursive loop?

52
00:02:44,787 --> 00:02:49,235
What if it starts spawning subagents that
spawn subagents, and I wake up to a

53
00:02:49,267 --> 00:02:50,995
five-thousand-dollar API bill?

54
00:02:51,373 --> 00:02:53,000
Yeah, the runaway train scenario.

55
00:02:53,144 --> 00:02:58,920
Anthropic actually built some pretty
strict guardrails into 2.1.212 for exactly

56
00:02:59,000 --> 00:02:59,560
that.

57
00:02:59,672 --> 00:03:02,921
They've introduced hard safety limits for
background sessions.

58
00:03:03,054 --> 00:03:07,481
By default, an agent is capped at 200 web
searches per session,

59
00:03:07,508 --> 00:03:09,561
which is controlled by the environment
variable

60
00:03:09,721 --> 00:03:13,721
CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION.

61
00:03:13,761 --> 00:03:17,641
And they also put a ceiling of 200
subagent spawns per session,

62
00:03:17,721 --> 00:03:22,978
which you can tune using
CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION.

63
00:03:23,012 --> 00:03:25,801
If it hits either of those, it just stops
and Two waits for you.

64
00:03:26,002 --> 00:03:28,672
hundred searches and two hundred
subagents.

65
00:03:28,752 --> 00:03:30,512
That's a pretty massive safety net.

66
00:03:30,652 --> 00:03:34,592
And let's say I hit that limit, or I
manually stop it, and I want to clean the

67
00:03:34,645 --> 00:03:34,992
slate?

68
00:03:35,072 --> 00:03:35,952
Does it reset?

69
00:03:35,958 --> 00:03:40,678
Yep, if you run a simple /clear command,
those counters reset right back to zero.

70
00:03:40,785 --> 00:03:42,038
It's a clean slate.

71
00:03:42,042 --> 00:03:42,602
Perfect.

72
00:03:42,702 --> 00:03:48,202
Now, what about when the agent itself is
just running a really slow tool call?

73
00:03:48,298 --> 00:03:53,242
Like, not a loop, but maybe an MCP-a Model
Context Protocol-tool that's taking

74
00:03:53,332 --> 00:03:57,162
forever to respond to a database query or
an external API.

75
00:03:57,290 --> 00:03:58,842
Does that still lock you up?

76
00:03:58,833 --> 00:04:01,473
Ah, this is a beautiful quality-of-life
tweak.

77
00:04:01,500 --> 00:04:06,273
In 2.1.212, they've set a default
threshold of two minutes.

78
00:04:06,380 --> 00:04:10,753
If an MCP tool call takes longer than two
minutes, Claude Code automatically

79
00:04:10,800 --> 00:04:11,713
backgrounds it.

80
00:04:11,773 --> 00:04:13,073
You don't even have to ask.

81
00:04:13,233 --> 00:04:17,313
It just pushes the slow call to the
background so your terminal stays snappy.

82
00:04:17,473 --> 00:04:21,553
And you can actually change that
two-minute limit using the config variable

83
00:04:21,740 --> 00:04:27,873
CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS if you
want it to background even

84
00:04:27,953 --> 00:04:30,353
faster, say, after thirty seconds.

85
00:04:30,333 --> 00:04:32,173
Oh, that is incredibly smart.

86
00:04:32,200 --> 00:04:35,773
So instead of sitting there praying the
database connection doesn't time out,

87
00:04:35,793 --> 00:04:39,053
the system just says, "Hey, this is taking
too long, I'm putting this in the

88
00:04:39,097 --> 00:04:40,493
background, you keep typing."

89
00:04:40,772 --> 00:04:41,420
Precisely.

90
00:04:41,460 --> 00:04:43,060
It keeps you in the flow state.

91
00:04:43,042 --> 00:04:46,162
And then, how do you actually get back to
those sessions?

92
00:04:46,178 --> 00:04:50,162
Like, if I have three forks running and
one got backgrounded because of a slow

93
00:04:50,215 --> 00:04:54,722
database call, how do I manage them
without getting lost in terminal soup?

94
00:04:54,708 --> 00:04:55,748
They've sorted that too.

95
00:04:55,935 --> 00:05:01,268
If you type /resume in the agent view, it
doesn't just ask you to type some obscure

96
00:05:01,428 --> 00:05:01,908
ID.

97
00:05:02,041 --> 00:05:06,228
It actually pops open a proper visual
history picker right in your terminal.

98
00:05:06,368 --> 00:05:09,588
You can scroll through your active and
historical sessions,

99
00:05:09,668 --> 00:05:14,708
and-this is the wild part-it can even
restore sessions you previously deleted.

100
00:05:14,815 --> 00:05:17,508
It's like a time machine for your terminal
workflows.

101
00:05:17,596 --> 00:05:18,580
That's huge.

102
00:05:18,635 --> 00:05:21,660
Alright, so we've got independent
background branching with /fork,

103
00:05:21,717 --> 00:05:26,780
safety caps at 200 searches, automatic
backgrounding for slow MCP calls,

104
00:05:26,800 --> 00:05:28,620
and a visual /resume picker.

105
00:05:28,716 --> 00:05:31,500
That is a massive update for terminal
productivity.

106
00:05:31,596 --> 00:05:33,420
Good chatting, Lachlan, see you next time.

107
00:05:33,417 --> 00:05:34,337
No worries, mate.

108
00:05:34,370 --> 00:05:35,097
Catch you later!