live_tv
Livestream Starting Soon
00
Hours
:
00
Minutes
:
00
Seconds
Up next in 10
The new Xcode 26.3 release candidate introduces support for agentic coding, Apple said Wednesday. It allows developers to work with AI coding agents like Anthropic’s Claude Agent and OpenAI’s Codex directly within the development environment.
Produced by Extra Ordinary for Cult of Mac
Follow us!
Threads: https://www.threads.net/@cultofmac
Mastodon: https://mastodon.social/@cultofmac
Instagram: https://instagram.com/cultofmac/
X: https://x.com/cultofmac
Facebook: https://www.facebook.com/cultofmac
Show More Show Less View Video Transcript
0:02
An update to Xcode enables AI agents to
0:05
operate with greater autonomy throughout
0:07
the entire app development cycle. Xcode
0:09
of course is Apple's comprehensive free
0:11
integrated development environment for
0:14
programming apps on all of its
0:15
platforms. And now instead of simply
0:18
suggesting code completions like uh you
0:20
know pathetic sort of Apple intelligence
0:22
stuff I'm guessing uh these agents can
0:24
break down complex tasks make artificial
0:28
architectural decisions and use Xcode's
0:31
built-in tools independently. Apple just
0:34
I I think they released this in beta or
0:37
something this week. Right. This is
0:38
in release candidate in Xcode.
0:39
Oh release candidate. Correct. Yeah. uh
0:42
supposedly makes Xcode capable of
0:44
building apps by itself, completely by
0:47
itself, which isn't scary at all. Uh
0:49
should make these easier for beginning
0:50
coders though. People on the uh
0:53
internet, developers like uh Steven
0:55
Trotten Smith, isn't it? He was talking
0:57
about how great it is. Says Xcode just
0:59
builds entire apps without you. Now you
1:02
could write a full shippable app every
1:04
evening with Xcode's agentic coding
1:06
features. Two apps a day even. He wrote
1:08
on Mastadon, which maybe you've heard
1:10
of. Woo! Maston. Wait, that was a horse,
1:13
not a I don't know what a ma does.
1:14
Anybody know what a mastadon sounds
1:16
like?
1:16
That must be a masttodon.
1:18
That's a mustang, not a mast.
1:20
That's a miss.
1:22
So, uh, he he he had a long thread on
1:25
Mastadon talking about this. He he said,
1:27
"I had Xcode's new agent feature throw
1:29
together a little UI kit timeline app
1:31
without me writing anything myself, all
1:33
using codecs."
1:35
Apparently, I mean, the the name dropped
1:37
ones are Claude agent and open AI
1:40
codecs, a couple of co coding tools, but
1:43
it it looks like you can use basically
1:46
anyone, right? I mean, they have some
1:47
kind of API, so you can pull in whatever
1:50
kind of coding AI you want into Xcode
1:53
and uh you know, streamline stuff. You
1:56
can have it do
1:58
some of the drudge work of of of this
2:00
stuff and check things out. But um
2:02
it supports uh MCP which is basically
2:06
the open- source equivalent to what
2:07
Apple's trying to do with like app
2:09
intents to part of the new Siri where
2:12
it's an open protocol that um a lot of
2:14
other companies are implementing that
2:16
basically lets you know AI things
2:18
control parts of computers and so Xcode
2:22
incorporates MCP support so you can just
2:24
plug in your API tokens and use a
2:26
different model if you want to use that
2:27
one instead. Of course, I think the part
2:29
that everyone glosses over is you kind
2:31
of got to know what you're doing in the
2:32
first place. I think like, you know,
2:34
when Graeme was talking about his VIP
2:36
coding experience, what came through
2:38
clearly really was, you know, he knew a
2:40
lot about what he wanted. Um, what the
2:44
different platforms and APIs were
2:47
capable of. Um, so when he was vibe
2:49
coding, he's coming from a lot of
2:51
knowledge, you know, like I think if you
2:53
read the press release, you kind of come
2:54
across the idea, oh, you know, this is
2:56
like I can just I can just like vibe
2:59
code and um my way into into
3:02
making a a a full app. Uh, and I don't
3:06
think that's probably going to be true
3:07
at all. uh you know, you've still got to
3:09
bring a lot of knowledge about what um
3:12
the different APIs and and uh platforms
3:15
can do for you and what your app's going
3:17
to look like, what you want it to do.
3:19
You know, it's not it's not going to
3:20
take all the work out of it.
3:22
This is much deeper integration than
3:23
they than they had before where it's not
3:25
just like editing a single document.
3:26
It's building and it can analyze your
3:28
entire project at once and it can even
3:31
double check the work that it's done by
3:33
like taking a you know capturing the
3:36
screen of the uh preview that Xcode has
3:38
built in or the simulator and verifying
3:41
that it's actually done what it's trying
3:43
to do. So it can double check itself as
3:45
well. It can clear warnings for you.
3:47
This is a this is a great thing for them
3:49
to do at this time, you know, before
3:51
WWDC in a few months. Yeah. You know, to
3:53
make developers happy before their big
3:55
developer event.
3:56
Well, my my joke about OpenClaw, you
3:58
know, obviously like uh for people who
3:59
may not uh know what I'm talking about,
4:02
OpenClaw is the rename of Clawbot. It's
4:04
an agent AI that a lot of people are
4:06
installing on Mac Minis and they're
4:08
giving it access to entire digital life
4:10
and it's supposed to be like they call
4:11
it AI with hands. So it can actually
4:14
really do stuff um go out and order
4:16
things for you or or you know all all
4:18
kinds of check your respond to thousands
4:21
of emails make the inbox go to zero um
4:24
and uh it's supposed to be it's a huge
4:26
security nightmare. people are setting
4:27
it up and giving it access to their full
4:29
digital life and these security
4:30
researchers have found that um they're
4:33
easily taken over even with prop attacks
4:34
you know it's it's you you communicate
4:36
it with it via messaging apps like
4:39
WhatsApp or or messenger uh messages and
4:42
um you know people have been uh able to
4:44
send malicious instructions to these
4:46
these these open claws installations
4:49
anyway that was a joke about open claw
4:51
coming to to Xcode
4:52
funnily enough Federico Vatici who wrote
4:55
you know one of the big articles about,
4:57
you know, how to set it, get started
4:59
setting it up on a Mac. Um, after he
5:01
published that article and it started
5:02
getting a lot of attention, he had to
5:04
disable Open Claw reading his email
5:06
because he was worried about, you know,
5:08
he has a public email address. People
5:09
could send him an email as prompt
5:11
injection into his computer, right? So
5:14
yeah,
5:14
it's a bit of a security nightmare in in
5:17
that regard that if you want it to be
5:19
useful at all, it needs to be, you know,
5:21
plugging into your actual computer. But
5:23
if it's plugged into your computer
5:24
everywhere, then you're opening yourself
5:25
to like a massive attack vector,
5:28
right? Nothing else is thought like
5:29
that. It's super super, you know,
5:31
security risk. Massive massive risk. But
5:34
one of the interesting things about
5:35
openclaw they were talking about was
5:37
like you can you can tell it to do new
5:38
things you know like and it'll go out
5:40
and it'll figure out how to install the
5:43
various packages and the various
5:45
software it needs in order to achieve
5:47
that and what that gives you is an
5:49
insight into our future computers. So
5:52
you know what's you know a future iPhone
5:57
might look like is something like an
5:59
open claw implementation where it's a
6:00
super smart assistant and what it means
6:03
is the death of apps. I mean you'll no
6:05
longer need an app to go do something.
6:07
You'll just tell your agent to go figure
6:09
it out and it'll add those it'll it'll
6:12
figure out how to do something on your
6:13
behalf and go do it. So some people are
6:18
arguing uh that this is the the
6:21
beginning of the end of apps that you
6:24
know standalone software will no longer
6:27
be needed. Now I I I don't believe that
6:29
for a second but
6:32
it's funny you said it pretty
6:33
convincingly. I thought you were already
6:35
carving up the headstone for apps.
6:37
Well yeah I I you know I think there's
6:39
an argument to be made for a lot of
6:40
things that is going to be true. But I
6:43
think for some things it won't be like
6:45
you'll want to have control over you'll
6:47
want to go see for yourself or do for
6:49
yourself certain things. Maybe photo
6:52
editing. I don't know if that's a good
6:53
example. You know like a kind some some
6:56
tasks you'll want to keep hands on.
6:58
There are a lot of tasks that there are
6:59
a lot of apps that will I think
7:01
disappear or at least task that apps
7:04
I can help you out here. I don't think
7:05
this will be the end of apps because
7:08
photo editing is a good example.
7:10
Photoshop is a hum is an incredibly
7:12
massive, incredibly complicated app that
7:15
has like, you know, 30year-old legacy.
7:18
AI can't replace that. And if you if it
7:20
tried to generate Photoshop every time
7:22
you wanted to edit a picture, it would
7:23
be a massive energy waste. It would be
7:26
incredibly slow. It's a waste of time.
7:28
Like there is still a room for an app
7:29
like Photoshop, Final Cut Pro, things
7:32
like those to exist
7:34
even in a super powerful post AI future.
7:36
But and and also like online services
7:38
like Instagram, Facebook, you know,
7:40
those require server infrastructure
7:42
software like iMessage, you know, basic
7:44
platform level stuff. But I think
7:46
I don't know, you know, I a messaging
7:48
app could definitely disappear because
7:49
you could say text my mom this, text my
7:51
brother that, I I don't want to go open
7:53
up the app and type it out. If if I
7:55
could have an agent that would do that
7:56
reliably for me, there's no way I am
7:59
going to go to an app to do it any
8:00
longer. That task has completely
8:02
disappeared. And I think it's true for
8:03
shopping too, right? Or or booking
8:05
trips. If you have an agent that you can
8:07
trust, you can say, "Okay, go buy me
8:09
some toilet paper and get it the
8:10
cheapest possible price,
8:12
why and it could do that reliably and
8:14
the toilet paper shows up, you know, the
8:16
next day from Amazon in a box. Why would
8:18
you go open Amazon ever again?" I would
8:20
if it can do it reliably is a very if is
8:23
doing a lot of heavy lifting in that
8:24
sentence,
8:25
right? I I I agree. But I agree if it
8:28
could
8:29
and also like yeah agentic things won't
8:31
make you know the the existence of apps
8:34
obsolete because there still needs to be
8:35
the the iMessage and service in place
8:38
for the agent to be able to do those
8:40
things
8:40
right some apps some apps but from the
8:42
consumer point of view you know from my
8:44
point of view as a as a consumer as a
8:45
user some apps are definitely going to
8:48
disappear
8:48
I don't know a good example is encoding
8:51
chapters into an MP3 file for a podcast
8:53
that that app exists but if you don't
8:56
want to use those in a post, you know,
8:58
imaginary 10-year feature from now where
9:00
AI can generate apps like, you know,
9:02
willy-nilly like that. You know, a
9:04
person who doesn't have, you know,
9:06
forecast, which is an app that does that
9:08
already, can just type it into a, you
9:10
know, an agent and have it generate one
9:11
that'll do it for them. You know, tiny
9:13
little utilities, technical problems
9:15
like that. I think those are those are
9:17
ripe to be reinvented by these things.
9:19
Well, co cod co coding is a good example
9:21
though like the uh at earlier in the
9:23
week um one of the top programmers at at
9:26
Facebook I can't remember the guy's name
9:28
he was uh uh and has since moved on
9:31
somewhere else I can't remember the
9:32
guy's name this tweet anyway someone
9:34
someone was commenting on a tweet that
9:35
he made where he said that 100% of his
9:39
coding is now handled by Claude or some
9:43
AI agent right 100%
9:44
works at Facebook that lines up
9:46
and and he said yeah yeah right good
9:48
good point but he said, you know, he was
9:50
he had very very mixed feelings about it
9:52
because on the one hand this is he said
9:56
this is an astonishing change. This is
9:57
this is like you know programming has
9:59
never been like this before. It's now
10:02
completely different. Programming has
10:04
completely changed. There is no point
10:06
for anybody who's learning who's going
10:09
to with computer who's learning computer
10:11
science to learn programming any longer.
10:13
This is, you know, I'm sure there you
10:16
could argue that that's not true. But
10:17
anyway, he's saying 100%
10:18
I actually wrote up a few notes on this
10:20
because um I also have a computer
10:22
science degree and I have some thoughts.
10:24
Now I I' I've been painted on this
10:27
podcast as like the anti-AI person
10:29
because I don't like AI generated
10:31
images. I don't like AI generated
10:33
videos.
10:33
Accurat.
10:34
What did I think you mean? Yeah,
10:37
I I think my my point of view has been
10:40
taken to a bit of a more extreme than
10:42
what it actually is because I actually
10:44
have less of a problem with this than
10:46
other AI services because the code
10:50
itself is not the product that you're
10:53
making here. You know, the the code
10:56
isn't the poetry. The code is not the
10:58
art. It's the code. The code is what it
11:00
generates to make the actual product,
11:02
which is the software and the app. Um,
11:05
now AI is always going to generate the
11:07
most bland, predictable, statistically
11:10
average outcome. And you know, that's
11:13
just fundamental to how it works. That's
11:15
what it does. Although I have a computer
11:17
science degree, you know, these coding
11:18
tools have a better understanding of
11:20
Swift and Python syntax than uh what I
11:24
can write. I I understand the broad
11:26
concepts, but I'm not as familiar with
11:28
like the actual those languages to
11:30
produce the code as accurately as these
11:32
things can, you know. So, I I actually
11:34
can see foresee myself using a tool like
11:37
this to help finish the, you know, two
11:40
dinky apps that I've been working on as
11:42
a hobby on and off for like the last 3
11:44
years because I've I've reached my skill
11:47
level and it's still not to the level
11:48
that I would want to ship them because,
11:51
and this is critical, these are two
11:53
dinky little hobby projects that run
11:55
locally, don't deal with any essential
11:57
user data or information, and can't
11:59
cause harm to anybody if if something
12:03
goes or if if the AI generates code that
12:05
has like a horrible race condition or
12:07
you know deletes data oh no the it's
12:11
lost basically nothing because the app
12:12
again the apps that I'm working on are
12:14
very small scale for better and worse
12:16
these are going to significantly
12:18
increase the output of the just global
12:21
output of software in the world right
12:23
and that's not something that can be
12:25
undone I don't think that the that's
12:28
going to be necessarily beneficial to
12:30
the world because there's already way
12:32
too much mediocre software that manages
12:34
and runs our entire lives. Like it's
12:37
it's so much harder doing things like,
12:39
you know, just getting a parking spot.
12:41
Now you have to download an app. Now you
12:43
have to do this. Now you have to you
12:44
need to sign up for an online service. A
12:45
thing that breaks, a thing that doesn't
12:47
account for any edge cases. And now
12:49
we're going to be trusting these AI
12:50
services to create them for us. And by
12:54
by companies who are only trying to, you
12:56
know, minimize their operational costs
12:58
to the highest degree, we're opening
13:00
ourselves up to a world where we are
13:02
putting way more trust in these things
13:04
to run so many little parts of our lives
13:07
in every meaningful way. And I think we
13:08
need to prepare for that.
13:10
It's taking a lot of the grunt work out
13:11
of coding. And I think it's there's a
13:13
jobs report today, wasn't there, that
13:15
just came out. There's been this is the
13:16
there have been more layoffs in the
13:18
month of was it for December? I think
13:20
there were more layoffs in December than
13:21
any time in the last 15 years and it
13:23
looks like a lot of entry- levelvel jobs
13:25
are being wiped out. Maybe, maybe not.
13:28
But, you know, for the in the coding
13:29
example, the grunt work is being done.
13:31
And what people like Steve Troutton
13:33
Swift say is that it allows a a
13:35
competent, you know, his example, a
13:37
competent developer to be much more
13:39
productive. Now, he can put he can do
13:41
two apps a day now in Xcode when it
13:43
might have taken him weeks before. And I
13:45
think it's true that, you know, uh,
13:47
they're not laying off the the senior
13:49
developers. They're not laying off the
13:50
senior programmers, the people who
13:52
really know what they're doing because
13:54
and, you know, will feature people have
13:56
to learn coding? Probably because you
13:58
kind of have to know what the machine's
14:00
doing on your behalf. So, I don't think
14:02
it's entirely going to wipe that out.
14:03
But it it does make you more productive.
14:06
And I think on the consumer level, the
14:07
same is true that a lot of the, you
14:08
know, the kind of grunt tasks that you
14:10
don't want to do will be taken over by
14:12
AI agents. And I don't think there's
14:14
anything wrong with that if you can
14:15
trust them. Okay, now that's a big if.
14:18
Uh, and right now you definitely can.
14:20
But this is what Apple's working on with
14:22
the smart assuri, isn't it? Trustworthy
14:23
private secure AI agents that will take
14:26
out the grunt work of, you know, that
14:27
that that right now some apps do for
14:30
you.
14:30
It's also going to have rippling effects
14:32
on the economy in the way that the
14:34
existence of these tools. It's shifting
14:36
power away from entrylevel people who,
14:38
you know, just learn a little bit of
14:39
coding to go to a coding boot camp or
14:41
have like, you know, maybe they have a
14:42
CS degree, but they don't have any job
14:44
experience. All of those jobs are going
14:45
to be gone. And that is going to be an
14:48
absolute net negative on the world
14:49
because it's shifting the existence of
14:51
these jobs only to the senior level
14:53
positions and it's, you know,
14:55
transferring the ability to to make
14:57
money on code. Well, like when when code
15:00
is devalued, it's shifting the ability
15:03
to make money with the skill of
15:06
programming only to the Fortune 500
15:08
companies.
15:10
Well, I maybe maybe not. And and
15:13
software has gotten more and more
15:14
abstract, hasn't it? As as time has gone
15:16
on. I mean, at at the beginning, you had
15:18
to learn how to you had to learn know
15:20
machine code, and then it got abstracted
15:23
up to another level with Forran and, you
15:25
know, whatever those other early
15:26
programming languages. And now modern
15:28
programming languages are even more
15:30
abstract, aren't they? And now it's got
15:31
to the level of abstraction where you
15:33
can just you can just you can you can
15:36
it's it works with plain English which I
15:39
think is an astonishing and absolutely
15:42
fantastic breakthrough. It's amazing.
15:46
It's amazing that that the machine can
15:49
translate those ideas now and just
15:51
implement it and you don't have to know
15:52
machine code. How many programmers these
15:54
days know machine code? I mean they
15:56
probably know exists. they may have some
15:57
like some you know some ideas about how
16:00
it works but they definitely don't learn
16:01
it and I think this is you know now
16:04
we've gone up to the another layer of
16:05
abstraction where I think you have to
16:07
know some of the you know you have to
16:09
know you have to you will definitely
16:10
have to have the basics of what you want
16:12
the code to do you'll defin you'll have
16:13
to know that because it's not a good
16:15
idea you know someone who doesn't know
16:17
what they're doing in charge of
16:19
something that might be critical
16:23
this is getting a bit abstract now but I
16:25
think it's amazing breakthrough I think
16:26
it's astonishing and I think there's
16:28
definitely going to be a huge shakeout.
16:30
Um, and I think there's definitely going
16:32
to come a time when there's going to be
16:33
a huge backlash too when we find out
16:35
there are huge downsides to this to to
16:37
allowing agents to do it. And
16:39
we already know what they are like that
16:41
the the data centers are massive and
16:43
require huge amounts of power. It's less
16:46
efficient. It's less energy efficient.
16:48
Again, we're we're going to see the the
16:49
the massive economical impact of this
16:51
and I don't think we're remotely
16:53
prepared for that. Uh, honestly, if
16:55
you're considering if you're considering
16:57
getting into a a white college job in
16:59
programming, I'd say don't. Go to a
17:01
trade school. Learn how to become a
17:02
plumber or
17:03
That's funny. That's funny because so a
17:05
good friend of my, you know, my 24 year
17:07
old son uh his group of friends, I mean,
17:10
a couple of them have gone into the
17:11
trades. They they went to one of them
17:12
went to Berkeley and now he's, you know,
17:14
working as a he's learning how to be a
17:15
carpenter because and he's thinking and
17:17
he is thinking very clearly, very
17:19
explicitly that, you know, he's having
17:21
trouble finding a job. He couldn't find
17:22
a job. that was brought that this is you
17:24
know what what uh and in San Francisco
17:27
this is um uh what spurred it but he's
17:30
also thinking ahead you know he doesn't
17:31
think those skills are going to be
17:32
disrupted and almost every single white
17:34
color job is in the cross sites of AI
17:37
and thinking about his future he wants a
17:40
blue color he wants a blue color trade
17:42
and it's smart he's smart he's smart to
17:44
do that it's a good idea
17:46
yeah I have a friend who works in HVAC
17:48
and I have much more confidence in his
17:49
career than I do mine
17:52
that's great isn't
17:53
Wonderful.