1
00:00:00,100 --> 00:00:04,340
Welcome to the show, and thanks to
Jellypod for helping make this daily show a

2
00:00:04,360 --> 00:00:04,870
reality.

3
00:00:05,640 --> 00:00:10,400
I'm Ethan Park, here with Maya, and today
we are looking directly at some highly

4
00:00:10,520 --> 00:00:15,300
practical updates in Codex 0.142.0.

5
00:00:16,100 --> 00:00:21,240
If you have ever run a multi-agent
workflow and watched in horror as your primary

6
00:00:21,320 --> 00:00:27,120
agent unilaterally decided to spawn five
sub-agents, which then spawned their own

7
00:00:27,180 --> 00:00:32,449
sub-agents, all burning through your API
credits, you are going to appreciate this

8
00:00:32,460 --> 00:00:32,880
release.

9
00:00:33,708 --> 00:00:36,328
Oh, the classic agent sprawl.

10
00:00:36,368 --> 00:00:39,608
It is a genuine nightmare for budget
predictability.

11
00:00:40,188 --> 00:00:44,888
You start a simple refactoring task, and
suddenly you have a recursive tree of

12
00:00:44,948 --> 00:00:48,668
agents discussing variable names on your
dime.

13
00:00:48,828 --> 00:00:52,628
It is basically the software engineering
version of Mickey Mouse and the magic

14
00:00:52,728 --> 00:00:54,008
brooms in Fantasia.

15
00:00:54,775 --> 00:00:55,385
Exactly.

16
00:00:55,915 --> 00:01:00,955
And to fix this, Codex has introduced
three distinct delegation modes in

17
00:01:01,315 --> 00:01:08,015
0.142.0: disabled, explicit-request-only,
and proactive.

18
00:01:08,615 --> 00:01:14,355
They give you direct, granular control
over how and when an agent can hand off tasks

19
00:01:14,415 --> 00:01:15,135
to a sub-agent.

20
00:01:15,694 --> 00:01:17,033
Let us break those down.

21
00:01:17,533 --> 00:01:23,143
"Disabled" is pretty self-explanatory-it
completely cuts off the agent's ability to

22
00:01:23,203 --> 00:01:25,943
spawn or delegate to any sub-agents,
right?

23
00:01:26,423 --> 00:01:29,203
It keeps everything strictly single-agent.

24
00:01:29,765 --> 00:01:29,985
Right.

25
00:01:30,445 --> 00:01:33,785
The agent has to solve the problem with
the tools it has locally,

26
00:01:34,225 --> 00:01:34,725
or fail.

27
00:01:35,465 --> 00:01:37,935
Then you have "explicit-request-only".

28
00:01:38,665 --> 00:01:42,985
In this mode, the agent can only delegate
if it explicitly asks you,

29
00:01:43,345 --> 00:01:48,865
the operator, for permission first, or if
your prompt directly instructs it to spin

30
00:01:48,905 --> 00:01:49,425
up a helper.

31
00:01:50,045 --> 00:01:52,465
It cannot make that call on its own
mid-run.

32
00:01:53,116 --> 00:01:56,136
And then "proactive" is basically the old
behavior?

33
00:01:56,536 --> 00:02:01,276
Where the agent can autonomously decide,
"Hey, this sub-task is better suited for a

34
00:02:01,396 --> 00:02:04,906
specialized code-writer agent, I am going
to spin one up right now"?

35
00:02:05,754 --> 00:02:10,934
Yes, but now it is a deliberate choice you
make, rather than the default wildcard

36
00:02:11,014 --> 00:02:11,454
behavior.

37
00:02:12,094 --> 00:02:14,654
And setting this up is incredibly
straightforward.

38
00:02:15,234 --> 00:02:19,134
It is all managed in your
`.codex/config.toml` file.

39
00:02:19,814 --> 00:02:23,914
You just add a ` ` block and set the
`mode` key.

40
00:02:24,554 --> 00:02:28,563
So, for example, `mode =
"explicit-request-only"`.

41
00:02:29,385 --> 00:02:34,844
That is great for global settings, but
what if I have a specific script where I want

42
00:02:34,854 --> 00:02:39,754
to allow proactive delegation just for
that run, without changing my global config?

43
00:02:40,254 --> 00:02:41,834
Can we override it dynamically?

44
00:02:42,617 --> 00:02:43,267
Absolutely.

45
00:02:43,677 --> 00:02:49,217
You can pass the mode dynamically on a
turn-by-turn basis through the CLI or the API

46
00:02:49,297 --> 00:02:49,626
wrapper.

47
00:02:50,267 --> 00:02:55,917
This means you can keep a tight global
default like "disabled" to prevent accidental

48
00:02:55,997 --> 00:03:01,117
spend, and only open the gate to
"proactive" when you are running complex,

49
00:03:01,357 --> 00:03:02,257
trusted pipelines.

50
00:03:02,824 --> 00:03:07,853
Speaking of preventing runaway spend, let
us talk about the new rollout token

51
00:03:07,932 --> 00:03:08,342
budgets.

52
00:03:08,832 --> 00:03:12,902
Because even if you allow delegation, you
still need a hard ceiling.

53
00:03:13,573 --> 00:03:17,233
Under the hood, how does Codex actually
track this now?

54
00:03:17,257 --> 00:03:19,397
This is a big architectural improvement.

55
00:03:19,937 --> 00:03:24,617
Previously, tracking was often localized
to a specific agent's run.

56
00:03:24,697 --> 00:03:29,637
Now, Codex tracks cumulative token usage
across entire agent threads,

57
00:03:30,157 --> 00:03:31,757
including all sub-agents.

58
00:03:32,337 --> 00:03:37,106
You set a hard limit-a rollout token
budget-and as the run progresses,

59
00:03:37,517 --> 00:03:40,417
Codex calculates the cumulative burn.

60
00:03:40,437 --> 00:03:45,227
If you start running low, it actually
inserts remaining-budget warnings directly

61
00:03:45,257 --> 00:03:47,487
into the system prompts to warn the agent.

62
00:03:48,169 --> 00:03:52,029
Wait, so the agent itself is made aware
that it is running out of money?

63
00:03:52,679 --> 00:03:54,769
"Hey, you have fifty cents left, wrap it
up"?

64
00:03:56,458 --> 00:03:57,398
Literally, yes!

65
00:03:57,978 --> 00:04:03,538
It tells the agent, "Your remaining budget
is X tokens," which encourages it to wrap

66
00:04:03,578 --> 00:04:06,458
up its work or stop spawning deep search
loops.

67
00:04:07,138 --> 00:04:12,238
And if the agent ignores that and hits the
hard ceiling, Codex doesn't just let it

68
00:04:12,298 --> 00:04:17,758
hang-it automatically aborts the turn
cleanly, saving you from those infinite loop

69
00:04:17,838 --> 00:04:22,848
scenarios where an agent tries the same
failing tool call fifty times.

70
00:04:23,399 --> 00:04:28,488
That is a massive relief for anyone who
has left an autonomous run active while

71
00:04:28,568 --> 00:04:32,118
grabbing lunch, only to return to a
massive bill.

72
00:04:33,148 --> 00:04:37,538
Now, let us pivot to some of the terminal
and network resilience updates in this

73
00:04:37,568 --> 00:04:41,648
release, because there are some really
nice quality-of-life fixes here.

74
00:04:41,728 --> 00:04:45,588
Specifically, the Linux TUI and Ctrl+Z.

75
00:04:45,687 --> 00:04:47,467
Oh, this one was highly requested.

76
00:04:48,027 --> 00:04:52,887
If you were running Codex's Text User
Interface on Linux and wanted to temporarily

77
00:04:53,027 --> 00:04:58,927
pause it-say, to run a quick git command
in your shell-you would press Ctrl+Z to

78
00:04:58,967 --> 00:04:59,587
suspend it.

79
00:05:00,287 --> 00:05:03,357
But when you typed `fg` to bring it back
to the foreground,

80
00:05:03,927 --> 00:05:06,317
the terminal rendering would completely
break.

81
00:05:06,987 --> 00:05:10,807
Text would overlap, and the layout would
turn into a scrambled mess.

82
00:05:11,474 --> 00:05:12,554
I remember that.

83
00:05:13,054 --> 00:05:18,434
You had to basically close the terminal
window or run a manual `reset` command to

84
00:05:18,494 --> 00:05:20,374
get your shell looking normal again.

85
00:05:22,334 --> 00:05:23,064
Exactly.

86
00:05:23,554 --> 00:05:29,264
In 0.142.0, they fixed the signal handling
for the Linux TUI.

87
00:05:29,914 --> 00:05:34,754
Now, when you bring it back with `fg`, it
triggers a clean redraw of the interface

88
00:05:34,814 --> 00:05:36,054
automatically.

89
00:05:36,114 --> 00:05:41,043
It sounds minor, but for command-line
purists, it makes the environment feel so much

90
00:05:41,094 --> 00:05:42,854
more native and stable.

91
00:05:43,253 --> 00:05:47,153
And speaking of stability, what about
those network disconnects?

92
00:05:47,813 --> 00:05:50,933
I saw something about stdio session
survival.

93
00:05:51,690 --> 00:05:58,650
Yes, Stdio MCP-Model Context
Protocol-sessions and the underlying

94
00:05:58,790 --> 00:06:02,290
exec-servers are now resilient to
transient disconnects.

95
00:06:03,050 --> 00:06:06,410
If you are on a flaky Wi-Fi connection or
switching networks,

96
00:06:06,910 --> 00:06:12,820
and the stdio connection drops
momentarily, Codex doesn't just crash the session and

97
00:06:12,830 --> 00:06:14,230
throw away your active state.

98
00:06:14,950 --> 00:06:19,530
It keeps the session alive in a holding
pattern and cleanly reconnects once the

99
00:06:19,550 --> 00:06:20,850
pipeline is restored.

100
00:06:21,868 --> 00:06:25,808
That is huge for anyone working on laptops
while traveling.

101
00:06:26,258 --> 00:06:31,058
You do not lose your entire context just
because the train went through a tunnel.

102
00:06:31,998 --> 00:06:34,958
Now, what about the `/usage` command
update?

103
00:06:35,790 --> 00:06:38,319
They have completely refactored `/usage`.

104
00:06:38,999 --> 00:06:42,649
It is no longer just a static output of
your current token consumption.

105
00:06:43,259 --> 00:06:47,559
You can now inspect your usage-limit reset
credits directly from the terminal.

106
00:06:48,179 --> 00:06:51,319
And more importantly, you can redeem those
credits in-line.

107
00:06:52,094 --> 00:06:53,624
In-line redemption?

108
00:06:54,194 --> 00:06:58,054
So you do not have to open a browser, log
into a dashboard,

109
00:06:58,454 --> 00:07:01,634
click three buttons, copy a code, and come
back?

110
00:07:02,455 --> 00:07:02,965
None of that.

111
00:07:03,535 --> 00:07:06,375
You just type `/usage` in your terminal.

112
00:07:07,055 --> 00:07:10,215
If you have earned reset credits, it will
show them to you.

113
00:07:10,875 --> 00:07:14,635
You can trigger a redemption right there
with built-in confirmation prompts.

114
00:07:15,215 --> 00:07:19,715
And they even built in retry states, so if
the billing or credit server has a

115
00:07:19,795 --> 00:07:25,235
momentary hiccup during the redemption
request, it won't fail silently or break your

116
00:07:25,275 --> 00:07:25,635
session.

117
00:07:26,684 --> 00:07:28,974
That is incredibly clean.

118
00:07:29,034 --> 00:07:33,794
It really feels like Codex is maturing
from an experimental tool into something

119
00:07:33,854 --> 00:07:37,634
designed for highly resilient,
professional local development.

120
00:07:38,615 --> 00:07:39,365
It really is.

121
00:07:39,975 --> 00:07:44,805
Well, that wraps up our quick look at
Codex 0.142.0.

122
00:07:45,515 --> 00:07:50,725
Dynamic delegation modes, hard token
guardrails, and much-needed terminal

123
00:07:50,795 --> 00:07:51,375
resilience.

124
00:07:52,055 --> 00:07:54,415
Thanks for listening, and we will see you
tomorrow.

125
00:07:56,003 --> 00:07:56,004
See you then!