1
00:00:08,180 --> 00:00:08,800
So, okay!

2
00:00:09,120 --> 00:00:15,480
On September 21st, 2026, SpaceXAI
officially dropped Grok 4.7,

3
00:00:15,920 --> 00:00:16,160
and...

4
00:00:17,100 --> 00:00:20,740
wow, this is a massive leap for persistent
background agents.

5
00:00:21,485 --> 00:00:26,825
The main headline here is that Grok 4.7 is
twice as fast and half the price of

6
00:00:26,845 --> 00:00:28,345
comparable frontier models.

7
00:00:28,905 --> 00:00:32,205
Now, if you are running continuous Grok
Bot workflows in the background,

8
00:00:32,645 --> 00:00:37,625
latency and compute costs are, uh, they
are basically the two biggest roadblocks you

9
00:00:37,665 --> 00:00:37,885
hit.

10
00:00:38,518 --> 00:00:41,418
When your bot is sitting there 24 seven
monitoring threads,

11
00:00:41,758 --> 00:00:46,478
checking code, or running background
checks, cutting that latency in half completely

12
00:00:46,518 --> 00:00:48,858
changes what you can hand off to an
autonomous team.

13
00:00:49,692 --> 00:00:55,212
But honestly, what gets me even more
excited as a developer is a less flashy update

14
00:00:55,252 --> 00:00:57,112
from the September 12th release notes.

15
00:00:58,012 --> 00:01:00,992
They solved what I call the silent killer
of agent runs.

16
00:01:01,562 --> 00:01:06,522
Picture this: your bot is halfway through
a complex task, and it hits a tool or app

17
00:01:06,582 --> 00:01:07,902
that you haven't authorized yet.

18
00:01:08,522 --> 00:01:13,061
Previously, it would just error out, crash
the run, or throw some obscure error.

19
00:01:14,012 --> 00:01:18,252
Now, when Grok needs a connector you
haven't linked, it literally stops right in the

20
00:01:18,272 --> 00:01:23,792
conversation and asks you directly:
connect, skip, or continue without leaving the

21
00:01:23,852 --> 00:01:24,192
chat.

22
00:01:24,963 --> 00:01:29,623
And on top of that, when a connector tool
actually fails, it doesn't dump a wall of

23
00:01:29,703 --> 00:01:33,383
raw JSON output or a messy stack trace
into your workspace.

24
00:01:33,943 --> 00:01:37,603
It explains what went wrong in plain,
human readable language.

25
00:01:38,533 --> 00:01:42,473
That means the bot can actually self
diagnose missing credentials or rate limits on

26
00:01:42,513 --> 00:01:45,393
the fly without breaking the whole
execution flow.

27
00:01:46,373 --> 00:01:48,113
That is huge for reliability.

28
00:01:49,135 --> 00:01:53,655
Before we jump into how SpaceXAI is
putting this to work in real production systems,

29
00:01:54,115 --> 00:01:55,435
a quick word from our sponsor.

30
00:01:56,215 --> 00:02:00,815
This episode is brought to you by Jellypod
AI, the absolute easiest platform to

31
00:02:00,855 --> 00:02:03,475
create AI podcasts and generative videos.

32
00:02:04,155 --> 00:02:07,835
If you want to turn your content into
engaging audio and video effortlessly,

33
00:02:08,315 --> 00:02:09,715
check out Jellypod AI.

34
00:02:10,852 --> 00:02:14,672
Okay, so how does all of this come
together in real company workflows?

35
00:02:15,532 --> 00:02:18,672
SpaceXAI published two fascinating
breakdowns recently.

36
00:02:19,392 --> 00:02:23,872
First, on September 22nd, they detailed
how they use Grok Bot to scale customer

37
00:02:23,952 --> 00:02:24,432
support.

38
00:02:25,212 --> 00:02:27,892
These bots handle tier one tickets from
start to finish.

39
00:02:28,452 --> 00:02:32,872
They verify account details, pull system
logs, and answer common issues.

40
00:02:33,692 --> 00:02:38,412
And if a complex ticket does need human
escalation, the bot hands it over seamlessly

41
00:02:38,452 --> 00:02:42,632
without losing the thread history, so the
customer never has to repeat themselves.

42
00:02:42,788 --> 00:02:46,988
Then back on September 4th, they
highlighted setting Grok Bot loose on procurement.

43
00:02:47,548 --> 00:02:51,728
Teams are using bots to scrape vendor
quotes, audit line item discrepancies across

44
00:02:51,768 --> 00:02:55,608
invoices, and stage purchase orders
straight into ERP platforms.

45
00:02:56,187 --> 00:02:56,907
Think about that!

46
00:02:57,287 --> 00:03:02,107
That is multi step, high stakes
operational work that used to require dedicated ops

47
00:03:02,167 --> 00:03:04,127
staff or complex manual scripts.

48
00:03:05,003 --> 00:03:06,983
Here is my practical takeaway from all
this.

49
00:03:07,843 --> 00:03:11,823
When you combine two times faster
reasoning at half the cost with strict,

50
00:03:12,023 --> 00:03:17,143
friction free inline authorization,
autonomous AI teams finally become viable for

51
00:03:17,203 --> 00:03:18,283
early stage startups.

52
00:03:19,027 --> 00:03:23,187
You do not need a massive devops team or
custom infrastructure to manage background

53
00:03:23,247 --> 00:03:23,987
agents anymore.

54
00:03:24,567 --> 00:03:28,667
You can literally just deploy a bot team,
grant permissions right inside the chat as

55
00:03:28,707 --> 00:03:32,627
needed, and let them handle support and
operations while you focus on building.

56
00:03:33,360 --> 00:03:36,440
It is a massive shift in how software gets
built and run.

57
00:03:37,360 --> 00:03:39,900
That is all for today's quick take, catch
you next time!