1
00:00:08,060 --> 00:00:13,420
So, if you have been watching the
getgrokbot.com ecosystem over the last few days of

2
00:00:13,480 --> 00:00:16,700
August, things moved, uh, extremely fast.

3
00:00:17,040 --> 00:00:20,640
Remember how we were looking at the
visitor wall and those initial editor rankings?

4
00:00:21,120 --> 00:00:25,580
Well, as of August thirty first, the
distribution model got a massive security

5
00:00:25,700 --> 00:00:26,140
overhaul.

6
00:00:26,900 --> 00:00:29,940
Basically, raw markdown pastes for shared
templates?

7
00:00:30,500 --> 00:00:30,880
Gone.

8
00:00:31,260 --> 00:00:32,380
Closed down entirely.

9
00:00:32,917 --> 00:00:37,097
Now, if you want a template listed, it,
it, it strictly has to be an official public

10
00:00:37,697 --> 00:00:41,477
x.ai/bot share link posted directly on X.

11
00:00:41,977 --> 00:00:45,257
And, and look, the reason for that is
pretty obvious when you think about prompt

12
00:00:45,297 --> 00:00:46,657
injection vulnerabilities.

13
00:00:47,177 --> 00:00:52,157
If anyone can just paste raw text into a
catalog, you risk malicious instructions

14
00:00:52,177 --> 00:00:54,577
getting injected right into someone's
agent prompt.

15
00:00:55,237 --> 00:01:01,317
By locking it to verified x.ai/bot links
on X, there is a clear chain of custody.

16
00:01:01,977 --> 00:01:05,477
Alongside that shift, the UI on the site
got a total refresh.

17
00:01:05,877 --> 00:01:10,597
They added a search UI, category pills,
and these much denser setup cards that give

18
00:01:10,637 --> 00:01:14,497
you direct actions like Copy setup text or
Open listing.

19
00:01:15,297 --> 00:01:20,017
And right in the site header, next to Docs
and Changelog, they added Grok Bot 101.

20
00:01:20,477 --> 00:01:24,757
It gives developers a clear, standardized
onboarding framework so you know how to

21
00:01:24,777 --> 00:01:27,957
set up execution limits before you turn a
bot loose on your system.

22
00:01:28,622 --> 00:01:31,182
And the templates coming out of this are
super specific.

23
00:01:31,662 --> 00:01:32,682
Take Jess, for example.

24
00:01:33,162 --> 00:01:37,562
Jess is structured as an executive
assistant for daily email triage and scheduling.

25
00:01:38,202 --> 00:01:42,702
Then there is Sanity, which handles CMS
draft creation, but with hard permission

26
00:01:42,722 --> 00:01:43,222
boundaries.

27
00:01:43,762 --> 00:01:47,342
It can write drafts, but it literally
lacks the authority to hit publish.

28
00:01:47,862 --> 00:01:50,782
So no accidental live deployments at two
in the morning.

29
00:01:51,445 --> 00:01:54,365
Then you have Eng Table, which is tailored
for engineering work.

30
00:01:54,825 --> 00:01:59,165
It takes an incoming issue, moves to
repro, and then runs debugging scripts.

31
00:01:59,685 --> 00:02:02,905
It builds on the same core patterns seen
in Hermes and OpenClaw,

32
00:02:03,345 --> 00:02:05,825
but tuned specifically for developer
workflows.

33
00:02:06,477 --> 00:02:08,017
Quick break to mention our sponsor.

34
00:02:08,637 --> 00:02:13,337
This episode is brought to you by Jellypod
AI, the easiest platform to create AI

35
00:02:13,457 --> 00:02:15,257
podcasts and generative videos.

36
00:02:15,817 --> 00:02:19,757
If you are looking to turn your audio
content or scripts into studio quality shows,

37
00:02:20,177 --> 00:02:21,537
check out Jellypod AI.

38
00:02:21,683 --> 00:02:25,083
Now, speaking of workflow infrastructure,
on August twenty eighth,

39
00:02:25,303 --> 00:02:30,883
getgrokbot.com launched a dedicated
reviews index page at /reviews.

40
00:02:31,563 --> 00:02:34,223
Before this, setup bot reviews were kind
of scattered.

41
00:02:34,763 --> 00:02:39,283
But importantly, even with the new
standalone index, these user reviews are strictly

42
00:02:39,323 --> 00:02:41,403
decoupled from the locked editor ranking.

43
00:02:42,203 --> 00:02:45,583
That editor ranking is curated by hand
after reviewing the setups,

44
00:02:46,043 --> 00:02:49,903
so separating community reviews keeps the
core benchmarks clean from rating

45
00:02:49,943 --> 00:02:50,663
manipulation.

46
00:02:51,412 --> 00:02:55,952
The real architectural leap here, though,
is the One Machine pattern updated around

47
00:02:56,032 --> 00:02:57,012
August twenty ninth.

48
00:02:57,812 --> 00:03:02,372
The old mental model was running multiple
agent jobs on a shared Grok cloud box.

49
00:03:03,092 --> 00:03:03,652
The new model?

50
00:03:04,392 --> 00:03:06,992
One bot, one job, one machine each.

51
00:03:07,752 --> 00:03:11,772
It is a paste ready setup where a bot
takes a single job on an isolated computer,

52
00:03:12,212 --> 00:03:14,992
finishes it, reports back, and then
terminates.

53
00:03:15,735 --> 00:03:19,535
What makes this so powerful is how you can
stack these single machine bots into

54
00:03:19,555 --> 00:03:22,815
production pipelines without risking cross
contamination.

55
00:03:23,575 --> 00:03:27,755
For instance, you could have a Video
Editor bot process raw media clips in an

56
00:03:27,835 --> 00:03:32,235
isolated container, pass the rendered
output to Sanity for CMS drafting,

57
00:03:32,675 --> 00:03:36,735
and neither bot ever has access to the
other's underlying environment or

58
00:03:36,835 --> 00:03:37,415
credentials.

59
00:03:38,417 --> 00:03:42,337
I actually tested this exact paste ready
template pattern on my own setup last

60
00:03:42,377 --> 00:03:42,677
night.

61
00:03:43,397 --> 00:03:47,357
I set up a single machine bot to monitor
GitHub, pull down new bug reports

62
00:03:47,417 --> 00:03:51,457
overnight, run a reproduction script
inside a sandboxed local environment,

63
00:03:51,897 --> 00:03:54,177
and output a clean log summary by morning.

64
00:03:54,957 --> 00:03:57,817
Because it was completely isolated on its
own target instance,

65
00:03:58,357 --> 00:04:02,817
I did not have to worry about handing out
broad production API tokens or letting an

66
00:04:02,877 --> 00:04:05,237
agent wander around my primary file
system.

67
00:04:05,923 --> 00:04:09,583
It just did the job, reported the repro
status, and shut down.

68
00:04:10,423 --> 00:04:14,763
That shift toward isolated, single task
agent machines is really starting to make

69
00:04:14,823 --> 00:04:18,403
autonomous workflows feel practical for
real engineering setups.

70
00:04:19,183 --> 00:04:21,123
That is all for today, catch you next
time!