In this episode of Simplyblock's Cloud Commute Podcast, Bruno Borges, Principal Product Manager at Microsoft, discusses millions of Java instances at Microsoft and the role of the Microsoft Java team. He introduces Semantic Kernel, a Java library for integrating AI into business applications, using large language models. Tune in to learn how Microsoft is shaping the future of development and AI integration.
In this episode of Cloud Commute, Chris and Bruno discuss:
- Java’s evolution at Microsoft and OpenJDK initiatives
- AI integration and the role of the Semantic Kernel in intelligent applications
- The rise of remote development environments
- The future of AI and LLMs in software development
Interested to learn more about the cloud infrastructure stack like storage, security, and Kubernetes? Head to our website (www.simplyblock.io/cloud-commute-podcast) for more episodes, and follow us on LinkedIn (www.linkedin.com/company/simplyblock-io/). You can also check out the detailed show notes on Youtube (www.youtube.com/watch?v=GrS8LmPVolE).
You can find Bruno Borges on X @brunoborges and Linkedin: /brunocborges.
About simplyblock:
Simplyblock is an intelligent database storage orchestrator for IO-intensive workloads in Kubernetes, including databases and analytics solutions. It uses smart NVMe caching to speed up read I/O latency and queries. A single system connects local NVMe disks, GP3 volumes, and S3 making it easier to handle storage capacity and performance. With the benefits of thin provisioning, storage tiering, and volume pooling, your database workloads get better performance at lower cost without changes to existing AWS infrastructure.
👉 Get started with simplyblock: https://www.simplyblock.io/buy-now
🏪 simplyblock AWS Marketplace: https://aws.amazon.com/marketplace/seller-profile?id=seller-fzdtuccq3edzm
00:00:01
I was running three
00:00:02
different Java binaries
00:00:03
of Microsoft Build
00:00:04
of OpenJDK, one for
00:00:05
Windows, one for
00:00:06
Apple Silicon, and
00:00:07
one for Mac Intel.
00:00:08
You know, just to show
00:00:09
that we want every
00:00:10
developer, no matter
00:00:11
where they are, to be
00:00:13
able to take advantage
00:00:14
of all the technology
00:00:14
that we produce.
00:00:15
I actually compiled an
00:00:16
embedded Linux kernel
00:00:18
in WSL2, wrote it right
00:00:20
to our embedded system.
00:00:21
Embedded Linux built
00:00:22
inside of Docker,
00:00:23
inside of WSL2, inside
00:00:24
of Windows, which
00:00:25
was kind of freaky.
00:00:27
But also pretty cool.
00:00:30
We humans are like that.
00:00:31
We tend to search for
00:00:32
things when we have
00:00:33
a problem to solve.
00:00:35
We tend to learn when we
00:00:37
have the need to learn.
00:00:38
If we can learn
00:00:39
faster and AI and LLMs
00:00:41
allow us to do that,
00:00:43
then we definitely
00:00:43
should leverage that.
00:00:47
You're listening to
00:00:47
Simplyblock's Cloud
00:00:48
Commute Podcast,
00:00:49
your weekly 20
00:00:50
minute podcast about
00:00:51
cloud technologies,
00:00:52
Kubernetes, security,
00:00:54
sustainability,
00:00:54
and more.
00:00:57
Hello everyone.
00:00:58
Welcome back to this
00:00:59
week's episode of
00:01:00
Simplyblock's Cloud
00:01:00
Commute Podcast.
00:01:01
This day, this week,
00:01:04
you know, I have another
00:01:05
incredible guest.
00:01:06
I say that every
00:01:07
time, it's every,
00:01:07
it's true every time.
00:01:09
This time, somebody
00:01:10
I actually know
00:01:11
for a very, very,
00:01:12
very long time.
00:01:14
He's been with Oracle
00:01:16
for many, many years.
00:01:17
He's for, with Microsoft
00:01:19
for also quite a
00:01:20
few years so far.
00:01:23
Bruno, welcome
00:01:24
to the show.
00:01:26
Hey Chris, thanks
00:01:27
for having me.
00:01:28
Yeah, and you're,
00:01:29
you're right.
00:01:30
I've been at Oracle
00:01:31
for about five
00:01:33
years and a half.
00:01:34
And then, I was
00:01:37
at Oracle for
00:01:37
about that time.
00:01:38
And then now I've
00:01:39
been with Microsoft
00:01:40
since 2018.
00:01:42
So you're almost,
00:01:43
well Six years now.
00:01:46
Yeah.
00:01:47
It's been-
00:01:48
I've been at Microsoft
00:01:49
more time than I
00:01:50
was at Oracle now,
00:01:51
just completing
00:01:52
my extra year.
00:01:54
So, I can say that I
00:01:57
did my dream of working
00:01:59
as a job engineer for
00:02:00
the company that does
00:02:01
Java and now I'm at the
00:02:04
company that does the
00:02:06
alternative to Java, but
00:02:08
I'm on the Java team.
00:02:09
So, we can talk
00:02:11
about interesting
00:02:12
things there.
00:02:13
All right.
00:02:14
As long as Microsoft
00:02:15
is never going to
00:02:16
go back to J sharp.
00:02:17
I think we're all
00:02:18
good with that.
00:02:19
No, no, no, no, no, no.
00:02:22
When we announced the
00:02:24
Microsoft Build of
00:02:25
OpenJDK, we had lots
00:02:26
of jokes about that.
00:02:28
We thought about
00:02:28
doing a prank.
00:02:30
April Fool's joke.
00:02:32
We thought about
00:02:32
announcing Microsoft
00:02:33
Build of OpenJDK
00:02:34
on April 1st.
00:02:38
And we didn't do
00:02:39
anything like that,
00:02:41
but one thing that
00:02:42
did happen when we
00:02:44
announced Microsoft
00:02:45
Build of OpenJDK was
00:02:47
that an anonymous
00:02:49
group of Microsoft
00:02:51
employees, and some
00:02:52
of them ex employees,
00:02:54
we don't know who
00:02:55
are these people, but
00:02:58
honestly, I don't know
00:03:00
who are these people,
00:03:01
but, of course, they saw
00:03:02
the news that Microsoft
00:03:05
announced Microsoft
00:03:06
Build of OpenJDK, or
00:03:09
that were, we were
00:03:10
about to announce
00:03:11
that, they published
00:03:13
a blog post called
00:03:16
the Microsoft Coffee.
00:03:18
And it was a prank that
00:03:20
they did back in the
00:03:21
days, where Microsoft,
00:03:24
where these guys
00:03:25
faked software boxes.
00:03:27
You remember when
00:03:28
you used to go to
00:03:29
a library store, a
00:03:30
bookstore, and you can
00:03:31
actually buy software?
00:03:32
Damn.
00:03:33
It's in this
00:03:34
round, mirror-ish
00:03:36
device called CD.
00:03:38
And we would actually
00:03:40
put inside the computer
00:03:42
and install software.
00:03:44
They did this prank
00:03:45
back in the days where
00:03:47
they printed software
00:03:48
boxes with the logo
00:03:50
Microsoft Coffee.
00:03:52
And it was a prank
00:03:55
on Java at the time.
00:03:58
And they actually put
00:04:01
dozens of boxes across
00:04:03
different stores in the
00:04:04
Seattle area in the US.
00:04:07
And the local press
00:04:10
saw that and they
00:04:12
said, Microsoft is
00:04:13
announcing a software.
00:04:16
And basically the
00:04:18
prank went too far.
00:04:20
Right.
00:04:20
Right.
00:04:21
So, sorry for, for,
00:04:23
for the audience.
00:04:25
Obviously Java
00:04:26
is an island.
00:04:27
It's a coffee.
00:04:28
It's a programming
00:04:29
language.
00:04:29
So it's a running gig.
00:04:31
Just in case
00:04:32
you don't know.
00:04:33
Now, you know.
00:04:33
Yeah, exactly.
00:04:36
So anyhow, if you go, if
00:04:39
you search for Microsoft
00:04:40
Coffee on Google
00:04:41
or Bing, whatever,
00:04:43
whatever you choose,
00:04:45
you're going to learn
00:04:45
about this thing.
00:04:46
It's, there's also
00:04:47
an account on Twitter
00:04:48
called Twitter.com or
00:04:49
X.com/MicrosoftCoffee.
00:04:52
And you're going to
00:04:53
see the blog post
00:04:54
that they published
00:04:55
at the same time that
00:04:56
we announced Microsoft
00:04:57
Build of OpenJDK,
00:04:59
telling the story,
00:05:00
the behind the scenes
00:05:01
story about that prank.
00:05:05
So, there's that.
00:05:06
And it's an interesting
00:05:07
fact that folks
00:05:08
would, would love
00:05:10
to learn about.
00:05:10
Yeah.
00:05:10
Yeah.
00:05:11
I'll try to find that.
00:05:12
If you, if you
00:05:13
have the links,
00:05:13
just send it to me.
00:05:14
I'll drop it in the
00:05:15
show notes because
00:05:16
that sounds amazing.
00:05:17
I did not know
00:05:18
about that.
00:05:19
There is a YouTube
00:05:21
video, showing the
00:05:23
news anchor talking
00:05:26
about the new product,
00:05:27
like it's a prank.
00:05:30
Anyways.
00:05:31
Yeah.
00:05:31
I can't-
00:05:32
To be honest, with a
00:05:34
company like Microsoft,
00:05:35
I can totally see
00:05:36
how this like, goes
00:05:37
way over the top.
00:05:39
Microsoft, they used
00:05:41
to do lots of pranks
00:05:42
back in the days, you
00:05:43
know, like Bill Gates
00:05:44
and Steve Ballmer
00:05:45
video recording, like
00:05:46
dancing, you know.
00:05:47
I mean, everyone knows
00:05:49
that Bill is famous
00:05:51
for his stupid jokes.
00:05:53
So, yeah.
00:05:53
Yeah.
00:05:54
You know, like, but now
00:05:56
we do like AI, AI, AI.
00:05:58
So you're still
00:05:59
doing stupid jokes.
00:06:00
Is that what
00:06:00
you're saying?
00:06:01
Yeah, basically.
00:06:02
Yeah.
00:06:02
Okay.
00:06:03
Okay.
00:06:03
Fair enough.
00:06:03
That's-
00:06:04
I mean, but we do, we do
00:06:07
jokes that are actually,
00:06:08
you know, funny.
00:06:10
All right.
00:06:13
Let's, let's get back to
00:06:14
the AI topic in a few.
00:06:17
Let us know a little
00:06:18
bit more about yourself.
00:06:19
To be honest, I was
00:06:21
kind of surprised.
00:06:22
I thought you've been
00:06:23
with Oracle much longer.
00:06:25
So that was,
00:06:26
that was quite-
00:06:26
Yeah, I joined, I joined
00:06:27
Oracle in 2012 and then
00:06:29
I left in January 2018.
00:06:32
That's when I basically
00:06:33
joined Microsoft.
00:06:34
And then, now
00:06:35
six years later.
00:06:37
So it was, it was an
00:06:39
interesting transition,
00:06:40
you know, like coming to
00:06:43
Microsoft to help them
00:06:44
talk to Java developers
00:06:47
about Azure, about
00:06:48
cloud, about Microsoft
00:06:50
developer tools like
00:06:51
Visual Studio Code.
00:06:52
And now GitHub after
00:06:54
the acquisition.
00:06:55
And then, one of
00:06:57
the interesting things
00:06:58
that happened in the
00:07:00
early days was that I
00:07:02
found out the amount
00:07:04
of Java that runs
00:07:05
inside Microsoft.
00:07:08
And it's not as big,
00:07:10
of course, as many
00:07:12
other companies like
00:07:12
Amazon or even Google,
00:07:14
perhaps, and of course,
00:07:17
Netflix and the likes
00:07:20
of that, the amount of
00:07:21
Java, like even Twitter
00:07:22
has a lot of Java.
00:07:24
But Microsoft runs,
00:07:26
our latest estimates,
00:07:28
it's somewhere between
00:07:29
2.5 to 3 million JVMs,
00:07:32
running in production
00:07:33
inside Microsoft for
00:07:35
internal systems,
00:07:36
not just, not, not,
00:07:37
does not count Azure
00:07:40
customer workloads,
00:07:42
just the internal stuff.
00:07:43
And we like to say- yep?
00:07:45
And that is all
00:07:46
on, on Azure or
00:07:47
how does that work?
00:07:48
Or is it also like
00:07:49
internal stuff?
00:07:50
No, internal, internal,
00:07:51
internal stuff.
00:07:52
Some stuff is running.
00:07:54
Some Microsoft internal
00:07:55
stuff runs on Azure data
00:07:56
centers, some stuff run
00:07:58
on Microsoft proprietary
00:08:00
data centers, like
00:08:01
Azure is proprietary,
00:08:02
but not inside
00:08:03
Azure data centers.
00:08:05
LinkedIn, for example,
00:08:07
is entirely Java.
00:08:08
The backend is 98,
00:08:10
99 percent Java and
00:08:12
hundreds of thousands of
00:08:13
JVMs run for LinkedIn.
00:08:17
Then we have
00:08:18
Minecraft servers.
00:08:19
We have Bing, the
00:08:22
search engine.
00:08:23
We have Azure Control
00:08:26
Plane, which is another-
00:08:27
It's an internal
00:08:28
technology that helps
00:08:30
us orchestrate the
00:08:31
Azure data centers.
00:08:33
Mm-Hmm.
00:08:34
We do a lot
00:08:34
of messaging.
00:08:35
We like to say
00:08:36
two things.
00:08:37
One, if you click
00:08:38
on your Start Menu
00:08:39
on Windows, that's
00:08:40
powered by Java.
00:08:41
How does that, what
00:08:42
does that mean?
00:08:43
It means that when you
00:08:44
click on Start Menu on
00:08:45
Windows, you see some
00:08:48
of Bing recommendations,
00:08:49
Bing search events,
00:08:51
news, whatever, right?
00:08:52
All of that is
00:08:53
coming from Bing.
00:08:54
And Bing has a lot
00:08:55
of indexes technology
00:08:57
based in Java.
00:08:59
And because of that,
00:09:00
we make that joke
00:09:01
that, you know, the
00:09:02
Start Menu of Windows
00:09:03
is powered by Java.
00:09:05
The second thing
00:09:07
is Azure.
00:09:10
So when you go on
00:09:11
Azure portal and you
00:09:12
create a resource,
00:09:14
let's say you want to
00:09:14
create a database, like
00:09:15
a Postgres database.
00:09:19
The Azure portal
00:09:20
actually sends a message
00:09:21
to Control Plane and
00:09:23
then Control Plane
00:09:24
will send a message
00:09:25
to whatever data
00:09:26
center you asked that
00:09:28
resource to be created.
00:09:31
All the messaging
00:09:32
system is done by a
00:09:33
PubSub scheme that is
00:09:34
implemented in Java.
00:09:36
Okay.
00:09:36
It's not JMS
00:09:37
or anything.
00:09:38
It's-
00:09:38
No, no, there's a lot
00:09:40
of, there's a lot of,
00:09:41
there's a lot of Kafka.
00:09:42
There's a lot of Hadoop.
00:09:43
There's a lot of PDO.
00:09:45
There's a lot of Spark.
00:09:47
And there's a lot of
00:09:49
Scala as well because
00:09:50
of the big data stuff.
00:09:53
There's a lot of
00:09:54
Android code, Java
00:09:57
for Android or Kotlin.
00:10:00
Yeah, I mean, it's
00:10:02
massive, you know.
00:10:03
At the time that we-
00:10:04
that Log4J
00:10:06
vulnerability happened,
00:10:09
I got a phone call
00:10:11
on my team's like,
00:10:12
Hey, Bruno.
00:10:14
Do you know where
00:10:14
Log4J is used
00:10:15
across Microsoft?
00:10:16
I was like, I don't
00:10:17
know, probably
00:10:18
everywhere.
00:10:19
Right?
00:10:21
So we went digging
00:10:22
and we found hundreds
00:10:24
and hundreds of
00:10:25
projects inside
00:10:26
Microsoft using Log4J.
00:10:28
Yeah.
00:10:28
And we went, we went
00:10:29
ahead of course,
00:10:30
and cleaned up
00:10:30
everything, helped
00:10:32
customers on Azure
00:10:33
to mitigate as well.
00:10:36
But it was an
00:10:36
interesting thing
00:10:38
where people are
00:10:38
like, Oh, we did not
00:10:40
know we had Log4J.
00:10:41
We didn't even
00:10:41
know we had Java.
00:10:42
And yeah, we do.
00:10:43
And, and it was, it
00:10:44
was one of the biggest
00:10:45
problems, right?
00:10:46
A lot of people didn't
00:10:47
know they have it
00:10:48
as like a transitive
00:10:49
dependency or anything.
00:10:50
Oh, yeah.
00:10:51
For the audience, if you
00:10:52
don't know what Log4J
00:10:53
is, we, by the way, had
00:10:54
an episode with Brian
00:10:55
Vermeer, just like a
00:10:57
few months ago talking
00:10:58
about mostly Log4J.
00:11:02
I'll put it into
00:11:03
the show notes.
00:11:05
Yeah.
00:11:06
So I mean, Microsoft
00:11:08
changed quite a bit
00:11:09
over the last couple of,
00:11:11
well, let's say the last
00:11:12
decade, maybe a decade
00:11:13
and a half or something.
00:11:14
Yeah, the last, the last
00:11:15
15 years have been very
00:11:16
different for Microsoft.
00:11:18
I think so, right?
00:11:19
And Microsoft, well,
00:11:21
you're, you're the best,
00:11:23
best person to say they
00:11:25
changed quite a bit in
00:11:26
terms of Java because-
00:11:27
except for J sharp.
00:11:29
After that, Microsoft
00:11:30
didn't really touch
00:11:30
Java at all, but
00:11:32
they came back.
00:11:33
But same for for
00:11:34
Linux, right?
00:11:35
I think the Azure-
00:11:36
Microsoft creating Azure
00:11:39
and make, trying to make
00:11:40
it big, which I think
00:11:41
they did pretty well,
00:11:44
changed quite a bit.
00:11:47
Would you say that
00:11:49
you actually see that
00:11:51
like still happening?
00:11:53
Is there still
00:11:53
changes happening?
00:11:54
Like how that Microsoft
00:11:56
works with open source
00:11:56
communities or stuff?
00:11:58
Yeah, changes are
00:11:59
happening all the time.
00:12:00
When we, when we
00:12:02
published Microsoft
00:12:02
Build of OpenJDK,
00:12:03
it was yet another
00:12:06
transformation, right?
00:12:07
You know, Microsoft
00:12:10
having its own JDK.
00:12:11
I mean, how, how
00:12:12
crazy is that?
00:12:14
And the second thing
00:12:15
is Java support on
00:12:16
Visual Studio Code.
00:12:19
It's amazing how
00:12:21
far it has come.
00:12:23
And thanks to the
00:12:24
partnership with
00:12:25
Red Hat, I must say.
00:12:27
Red Hat drives the
00:12:29
CoreX language support
00:12:30
extension, and then
00:12:31
Microsoft provides all
00:12:32
the extensions around
00:12:34
it like Maven support,
00:12:35
JUnit support, Project
00:12:37
support, and so on.
00:12:41
And now with the
00:12:42
integration with GitHub
00:12:43
and GitHub Copilot,
00:12:44
GitHub Codespaces, and
00:12:46
all of that cool stuff,
00:12:48
it shows really like,
00:12:49
how the company and
00:12:52
its subsidiaries think
00:12:53
about the developer.
00:12:55
And when we think
00:12:56
about the developer,
00:12:57
it's not just the C
00:12:58
sharp developer or the
00:12:59
JavaScript developer,
00:12:59
we also think about
00:13:00
the Java developer.
00:13:02
And that in the
00:13:03
past would not
00:13:04
have been possible.
00:13:05
I think, I think the
00:13:06
Microsoft from the
00:13:07
past would probably say
00:13:08
something like we don't
00:13:09
care about any other
00:13:10
language developer.
00:13:11
We only care about
00:13:12
Windows developers.
00:13:13
And Visual Basic
00:13:15
developers and-
00:13:16
C Sharp, come on.
00:13:17
And C sharp and .NET
00:13:18
developers, right?
00:13:19
That's what
00:13:20
we care about.
00:13:20
Right.
00:13:21
But no, I mean, the
00:13:23
facts today show
00:13:24
that, we care
00:13:27
about any developer.
00:13:29
And you might think,
00:13:30
wow, that's obvious,
00:13:32
Bruno, because
00:13:32
Microsoft wants to grow
00:13:33
its cloud business.
00:13:34
Well, in part, yes.
00:13:36
But in also because
00:13:38
we want truly to make
00:13:40
every service and every
00:13:41
tool, every project we
00:13:43
produce to be able to be
00:13:45
used by anyone, right?
00:13:47
It's not an
00:13:48
allotest tool set.
00:13:51
It's not an allotest
00:13:52
or, you know, niche
00:13:54
stack of technologies.
00:13:56
We really want anybody
00:13:58
to be able to use that.
00:13:59
So that they can
00:14:00
be as productive as
00:14:01
possible, as efficient
00:14:03
as possible, deliver
00:14:04
with quality, with
00:14:05
speed and everything.
00:14:06
And if that
00:14:08
includes Java
00:14:09
developers, so be it.
00:14:10
We're going to make
00:14:10
sure that everything
00:14:11
that we produce is
00:14:12
equally as good for
00:14:13
Java developers.
00:14:14
I think you're making
00:14:15
an important point here.
00:14:16
I mean, you mentioned
00:14:17
Visual Studio or Visual
00:14:19
Studio Code, sorry,
00:14:20
Visual Studio Code,
00:14:21
which, I think, I don't
00:14:23
know if it's like the
00:14:25
biggest or at least
00:14:26
the second biggest
00:14:27
IDE by now just like
00:14:29
from, from zero to 100.
00:14:32
But also from a
00:14:33
developer's perspective,
00:14:35
WSL or especially
00:14:36
WSL2 made like a
00:14:38
massive leap forward.
00:14:40
Oh, yeah.
00:14:41
I use Windows subsystem
00:14:44
for Linux every day.
00:14:45
It's my default.
00:14:46
When I opened up my
00:14:47
Windows terminal, the
00:14:47
default is Ubuntu.
00:14:49
Not PowerShell.
00:14:50
Yeah.
00:14:51
And that makes sense.
00:14:52
And it's amazing because
00:14:54
it is actually running
00:14:55
a Linux kernel on a,
00:14:56
what is it called?
00:14:56
A micro VM, whatever
00:14:58
you call that.
00:14:59
I actually compiled an
00:15:01
embedded Linux kernel
00:15:03
in WSL2, and build it,
00:15:05
and wrote it right to
00:15:08
our embedded system.
00:15:09
Because it is, well, it
00:15:11
was actually an embedded
00:15:12
Linux build inside of
00:15:14
Docker, inside of WSL2,
00:15:15
inside of Windows, which
00:15:17
was kind of freaky,
00:15:19
but also pretty cool.
00:15:23
It is.
00:15:24
I've done, I've
00:15:25
compiled the OpenJDK
00:15:26
a few times on a
00:15:28
subsystem for Linux.
00:15:29
And-
00:15:30
That makes sense.
00:15:31
It works just fine.
00:15:32
I produce a Linux
00:15:33
binary off the JDK in
00:15:34
my Windows computer.
00:15:36
But I do want to point
00:15:37
out that I am probably
00:15:40
the weirdest Microsoft
00:15:42
employee because my main
00:15:44
computer is a Windows
00:15:45
PC and then I have
00:15:49
a Mac, an Intel Mac.
00:15:52
Right.
00:15:53
And I have a third
00:15:54
computer, which is
00:15:55
another Mac, the M1 Mac.
00:15:58
It's not here in my
00:15:58
office, but I have
00:16:00
three computers in that.
00:16:02
I was, two years ago, I
00:16:04
was doing an experiment
00:16:05
about JVM ergonomics.
00:16:08
And I set up a lab on
00:16:11
my office where I had
00:16:14
a Java backend running
00:16:16
on my Windows computer,
00:16:18
a database running on
00:16:20
Mac and the workload
00:16:23
generator to stress
00:16:25
test on my Intel Mac.
00:16:27
So I was running
00:16:28
three different Java
00:16:30
binaries of Microsoft
00:16:31
Build of OpenJDK.
00:16:32
One for windows, one for
00:16:33
Mac and Apple silicon,
00:16:34
and one from Mac Intel.
00:16:36
So, you know, just to
00:16:38
show that we want every
00:16:41
developer, no matter
00:16:43
where they are, to be
00:16:44
able to take advantage
00:16:46
of all the technologies
00:16:47
that we produce.
00:16:48
All right, cool.
00:16:50
So let's, let's quickly
00:16:52
go back to, or quickly,
00:16:53
let's go back to the
00:16:54
AI part, because
00:16:55
I think there's one
00:16:56
very specific and
00:16:58
very interesting open
00:16:59
source project that
00:17:00
you were fairly deeply
00:17:02
involved in, which was
00:17:04
the Semantic Kernel
00:17:05
for Java, right?
00:17:08
So maybe, maybe
00:17:09
just give a quick
00:17:10
introduction for the
00:17:11
people that haven't
00:17:12
heard about that.
00:17:14
So Semantic Kernel
00:17:16
is a Java library to
00:17:17
help developers build
00:17:21
intelligent apps.
00:17:23
You imagine like,
00:17:24
okay, I have Spring,
00:17:26
I have Quarkus, I have
00:17:28
Express.js, I have
00:17:30
Django for Python.
00:17:33
These are all frameworks
00:17:34
that help developers
00:17:36
build business
00:17:37
applications in general.
00:17:42
Semantic Kernel is a
00:17:44
library, but also sort
00:17:46
of like a framework
00:17:47
that integrates with
00:17:49
your application to help
00:17:51
you build intelligent
00:17:53
features that will
00:17:54
use, of course, LLMs,
00:17:57
large language models,
00:17:58
and AI services
00:17:59
behind the scenes.
00:18:01
So that allows
00:18:02
the developer to
00:18:03
not only do prompt
00:18:05
engineering, but also
00:18:06
do AI orchestration.
00:18:08
So, you provide
00:18:10
plug ins, you provide
00:18:12
functions, you
00:18:13
provide capabilities
00:18:16
in the application.
00:18:17
Then you tell Semantic
00:18:18
Kernel, these are
00:18:19
my capabilities.
00:18:21
And then you can
00:18:22
allow a prompt to
00:18:26
say, I want to do
00:18:26
this, this, and this.
00:18:28
That's the prompt.
00:18:29
Imagine that's what
00:18:30
you tell ChatGPT.
00:18:32
But that prompt will
00:18:34
make the AI behind the
00:18:35
scenes leverage the
00:18:37
capabilities of your
00:18:38
own application to
00:18:39
trigger those things
00:18:41
as part of the flow.
00:18:43
So that allows you
00:18:45
to say, Hey, give me
00:18:48
the last financial
00:18:49
resource and email
00:18:51
that to my boss.
00:18:53
Okay, you have a
00:18:54
function that extracts
00:18:56
the latest, or extracts
00:18:58
financial results from
00:18:59
the database, and you
00:19:00
have another function
00:19:01
that sends email.
00:19:02
But you also have
00:19:03
a function that
00:19:04
finds the email
00:19:05
address of employees.
00:19:09
But your prompt is just
00:19:10
something like, get
00:19:12
the latest results and
00:19:13
send to my manager.
00:19:14
Okay, the system
00:19:15
knows who the user is.
00:19:17
The system has a
00:19:18
function that extracts
00:19:19
the employee directory.
00:19:21
So the system will
00:19:22
know who is your boss.
00:19:24
And the system will
00:19:25
be able to find the
00:19:26
email of your boss.
00:19:27
And the system also
00:19:28
has the capability
00:19:28
to send an email.
00:19:30
So, Symantec Kernel
00:19:31
will chain all of that
00:19:34
and execute the request
00:19:37
as per the prompt.
00:19:39
So Semantic Kernel
00:19:41
means, or that, that
00:19:43
means it basically
00:19:44
helps you with the like
00:19:45
agents, like the AI
00:19:48
agents that everyone
00:19:49
is talking about.
00:19:50
Okay.
00:19:50
Yes, exactly.
00:19:51
That makes sense.
00:19:52
And there are other
00:19:54
alternatives, of
00:19:54
course, LangChain4j
00:19:56
is the main one.
00:19:58
The Spring folks
00:19:59
have something called
00:20:00
Spring AI, also
00:20:01
another library that
00:20:02
does similar things.
00:20:05
I believe Spring AI
00:20:06
has also integration
00:20:08
with LangChain4j.
00:20:09
And we are also
00:20:11
considering Spring
00:20:12
integrations for
00:20:13
Semantic Kernel, Quarkus
00:20:14
integrations, and
00:20:15
Micronaut, and all the
00:20:17
other Java frameworks,
00:20:18
application frameworks.
00:20:20
But, in essence, that's
00:20:21
what Semantic Kernel is.
00:20:22
It allows developers to
00:20:24
more easily integrate AI
00:20:27
capabilities into their
00:20:28
business applications.
00:20:29
And I guess it intro
00:20:31
or in the backend, it
00:20:33
uses all the typical
00:20:34
LLM services like
00:20:36
Microsoft AI or-
00:20:39
No, you choose, right.
00:20:41
The library is
00:20:42
just a library.
00:20:43
So you have to
00:20:43
tell, okay, this
00:20:44
is my AI service
00:20:45
that I want to use.
00:20:47
So, right now we have
00:20:48
support for Azure
00:20:49
OpenAI, of course.
00:20:51
We have also support
00:20:52
for OpenAI, which is
00:20:53
a different company.
00:20:54
Yep.
00:20:55
We have support
00:20:57
for Hugging Faces.
00:20:59
And I believe, I'm not
00:21:01
sure if we have that
00:21:02
support yet, but we do
00:21:04
have a ticket on GitHub
00:21:07
about Gemini, about
00:21:08
Google's AI service.
00:21:11
I mean, we are open to
00:21:14
include any AI service
00:21:15
into the framework.
00:21:18
And if Google folks
00:21:20
are asking for that,
00:21:22
we can evaluate and
00:21:23
maybe collaborate with
00:21:24
them to make sure.
00:21:26
The interesting thing
00:21:26
about Semantic Kernel
00:21:29
versus LangChain4j is
00:21:31
some customers like the
00:21:33
idea of having a company
00:21:35
behind a library, right?
00:21:38
LangChain4j is great.
00:21:40
There's lots
00:21:40
of engineers.
00:21:41
I know Liz is one of
00:21:43
the engineers behind.
00:21:44
She's amazing.
00:21:44
And she's doing
00:21:45
an amazing job
00:21:46
on LangChain4j.
00:21:47
But it's a community
00:21:49
project and that
00:21:51
by itself can be
00:21:52
a blessing, but it
00:21:54
can also be a curse.
00:21:55
And sometimes some
00:21:57
companies may think,
00:21:58
you know, I don't
00:21:59
want to bat on a
00:22:00
community library.
00:22:02
So, I really prefer
00:22:04
to use a ladder
00:22:04
that is maintained
00:22:05
by a major company.
00:22:07
And that's one of the-
00:22:08
That's one of the
00:22:10
value propositions
00:22:10
for Semantic Kernel.
00:22:11
Yeah, that makes sense.
00:22:12
Especially when you look
00:22:13
into bigger companies,
00:22:14
lawyers tend to get like
00:22:17
a little bit very, when
00:22:19
you, when you just want
00:22:21
to use some open source
00:22:22
project, and they know
00:22:23
they can't keep anyone
00:22:28
in, in check if stuff,
00:22:29
stuff happens, right.
00:22:33
So you said all the
00:22:36
services, could I
00:22:37
also use a lemma?
00:22:39
Building my own
00:22:40
local thing.
00:22:42
If not, there
00:22:42
would be a ticket
00:22:43
for that, I guess.
00:22:46
Hey, it's, you know,
00:22:47
it's a GitHub project.
00:22:49
Go send a pull request.
00:22:51
I can just imagine that
00:22:52
a lot of people would
00:22:53
love to play with that.
00:22:55
But also love to use
00:22:56
some local service
00:22:57
to try it out.
00:22:59
I think, I think
00:23:00
we have that.
00:23:01
We do have that in mind.
00:23:03
I think, I wonder if I
00:23:05
have, if we have that
00:23:07
as a work in progress
00:23:09
or if it's still
00:23:09
just in the back, I
00:23:10
have to double check.
00:23:12
It's just that Symantec
00:23:13
Kernel is not the
00:23:14
only project that
00:23:15
I look at and help
00:23:18
coordinate at Microsoft.
00:23:19
We have a Build of
00:23:19
OpenJDK, we have a
00:23:20
migration tooling,
00:23:21
we have some of our
00:23:23
marketing initiatives
00:23:24
and campaigns, and
00:23:26
Azure features for Java
00:23:27
developers, VS Code,
00:23:29
Java features, there's
00:23:30
a bunch of other stuff
00:23:31
that I participate
00:23:33
and contribute to.
00:23:34
So, that's why I'm,
00:23:35
you will hear me a lot.
00:23:36
Yeah, I'm not sure.
00:23:38
We'll double check.
00:23:39
To be honest, I think
00:23:40
that it's a great
00:23:41
trait to be able to
00:23:43
say that because you
00:23:44
can't know everything.
00:23:45
It's always important
00:23:46
that you know where
00:23:47
to look it up, right?
00:23:48
So we crossed the
00:23:49
20 minutes mark like
00:23:50
two minutes ago.
00:23:52
So last question, like
00:23:54
when you look into the
00:23:55
future, what do you
00:23:56
think is like the next
00:23:57
big thing for Microsoft,
00:23:59
for database, for AI,
00:24:00
for Java, for whatever
00:24:02
you can think of?
00:24:04
I truly believe
00:24:05
in two things.
00:24:05
One, remote development
00:24:10
environments
00:24:11
are become the-
00:24:12
will become the norm.
00:24:15
Whether that remote
00:24:16
environment is in
00:24:17
the cloud or not,
00:24:18
that's irrelevant.
00:24:19
What is relevant is
00:24:20
that for the developer,
00:24:22
they have access to
00:24:23
a remote environment
00:24:25
that is more powerful
00:24:26
than the machine that
00:24:27
they currently have,
00:24:28
and will allow them
00:24:30
to do all sorts of
00:24:32
integration tasks,
00:24:33
because the complexity
00:24:35
of systems these
00:24:36
days has been growing
00:24:38
exponentially, you know.
00:24:39
And that requires
00:24:42
environments that
00:24:43
are more complex to
00:24:45
be able to develop
00:24:46
and most importantly,
00:24:48
test those systems.
00:24:50
So the need for proper
00:24:52
environments that mimic
00:24:55
production environments
00:24:56
will only grow.
00:24:58
And I know there is
00:25:01
Minikube and, you
00:25:03
know, Q3S and all these
00:25:05
sort of, you know,
00:25:07
capabilities to have
00:25:09
a similar environment
00:25:10
in your laptop.
00:25:10
But the reality is on
00:25:11
your laptop, you'll
00:25:12
also have Chrome,
00:25:14
Slack, Discord,
00:25:16
Spotify, and all these
00:25:18
other things that are
00:25:18
eating CPU and memory.
00:25:20
So, you need a dedicated
00:25:22
developer environment
00:25:23
if you need a database,
00:25:24
an AI LLM service.
00:25:27
You need Kubernetes,
00:25:30
you need a backend,
00:25:31
you need Key Vault, all
00:25:34
these other services
00:25:35
that your application
00:25:36
relies on, you're
00:25:37
going to need that.
00:25:38
So emulating all of that
00:25:40
in your local computer
00:25:41
will be impossible
00:25:43
moving forward.
00:25:44
So I think remote
00:25:45
development environments
00:25:46
will become a big thing.
00:25:48
And the second thing
00:25:48
is developers will
00:25:52
have to adapt to using
00:25:55
AI as a tool, not
00:25:58
as a capability for
00:25:59
their applications.
00:26:00
I mean, developers will
00:26:02
have to build AI or
00:26:03
intelligent capabilities
00:26:05
into their systems.
00:26:06
That's certain
00:26:08
to happen.
00:26:10
But not for
00:26:11
every system.
00:26:12
Not every system needs
00:26:13
an intelligent feature.
00:26:15
Tell that the world,
00:26:17
please say it again.
00:26:18
Not every,
00:26:19
Not every system needs
00:26:21
an intelligent feature,
00:26:23
but every developer
00:26:25
will need to leverage AI
00:26:29
in some shape or form.
00:26:30
Okay.
00:26:30
In the past, I remember,
00:26:34
I remember my old days.
00:26:35
We went as software
00:26:36
engineers, we
00:26:37
learned how to
00:26:38
program using books.
00:26:40
Then the internet
00:26:41
came and we continued
00:26:43
to learn through
00:26:44
books, just that, you
00:26:46
know, online books.
00:26:47
And then very quickly,
00:26:49
we transitioned to
00:26:50
articles and tutorials
00:26:52
and user groups and
00:26:54
online communities.
00:26:55
And then we switched
00:26:57
to videos, a lot of
00:26:58
video on the internet,
00:27:00
where we are capable
00:27:02
of, you know, extracting
00:27:05
information faster
00:27:05
by someone just
00:27:06
telling, you know, just
00:27:07
getting to the point.
00:27:09
Now, we are digitizing,
00:27:12
digitalizing, and
00:27:13
aggregating all that
00:27:14
information in a way
00:27:16
that you can just
00:27:16
ask the internet.
00:27:17
You know, Isaac Asimov,
00:27:21
famous philosopher, said
00:27:23
he predicted that the
00:27:25
internet would exist.
00:27:26
Not the internet, he
00:27:27
predicted that a system
00:27:28
would allow anyone-
00:27:31
will allow anyone to
00:27:31
just ask a question
00:27:32
and this computer
00:27:33
thing would just
00:27:34
tell you the answer.
00:27:37
That today, we thought
00:27:39
it was the internet,
00:27:41
but the internet got,
00:27:42
let's say abused.
00:27:48
The answer to that
00:27:49
is not the internet.
00:27:51
The answer to that
00:27:52
is Gen AI, but it's
00:27:55
thanks to the data
00:27:56
that the internet
00:27:58
produce, produces, that
00:28:00
GenNi is capable of
00:28:01
answering the question.
00:28:02
So, LLMs, I believe LLM
00:28:04
is the ultimate answer
00:28:06
to Isaac Asimov's quote.
00:28:10
So we developers have to
00:28:12
be able to, I believe,
00:28:14
use this as a tool so
00:28:16
we can develop faster,
00:28:17
better, and, you
00:28:22
know, find, maybe
00:28:24
find that joy again
00:28:25
in programming.
00:28:26
You know, I learned
00:28:27
how to program C++.
00:28:29
Now, did I
00:28:30
actually learn how
00:28:31
to program C++?
00:28:32
No.
00:28:33
Any C++ developer
00:28:34
will say no, where
00:28:35
he didn't learn.
00:28:35
You learn how to ask
00:28:37
good questions on
00:28:37
ChatGPT and copy paste.
00:28:40
And I was like, yeah.
00:28:42
Isn't that what
00:28:43
programming is at
00:28:44
the end of the day?
00:28:45
You know, majority of
00:28:46
developers do that.
00:28:47
They go on Google and
00:28:48
they find an article.
00:28:49
Then they go and
00:28:50
read the article.
00:28:51
They understand
00:28:51
the article.
00:28:52
They copy paste.
00:28:52
It works.
00:28:53
Done.
00:28:54
Next problem.
00:28:55
We humans are like that.
00:28:56
We tend to search for
00:28:58
things when we have
00:28:59
a problem to solve.
00:29:00
We tend to learn when we
00:29:04
have the need to learn.
00:29:06
So, if we can learn
00:29:08
faster, and AI and LLMs
00:29:11
allow us to do that,
00:29:13
then we definitely
00:29:14
should leverage that.
00:29:15
So those are my
00:29:15
two predictions.
00:29:16
Remote development
00:29:17
environments and Gen AI
00:29:18
or LLMs will be a daily
00:29:21
tool for developers.
00:29:22
Okay, cool.
00:29:25
I have my own thoughts
00:29:26
on that, but for
00:29:27
the sake of time-
00:29:30
my- #MyOpinion.
00:29:32
Exactly.
00:29:35
Exactly.
00:29:35
All right.
00:29:36
I like that.
00:29:37
I think that is
00:29:38
a perfect way
00:29:39
to end the show.
00:29:40
My opinion.
00:29:41
Stop that.
00:29:42
#MyOpinion.
00:29:44
All right.
00:29:45
Bruno, thank you
00:29:46
for being here.
00:29:47
It was awesome
00:29:48
to get the chance
00:29:49
to talk to you.
00:29:49
It's been a while.
00:29:50
And for the audience,
00:29:52
thank you for
00:29:53
being here as well.
00:29:54
And you know where
00:29:55
to find us next week.
00:29:56
So see you again.
00:30:00
The Cloud Commute
00:30:01
Podcast is sponsored by
00:30:02
Simplyblock, your own
00:30:03
elastic block storage
00:30:04
engine for the cloud.
00:30:05
Get higher IOPS and
00:30:06
low predictable latency
00:30:08
while bringing down your
00:30:09
total cost of ownership.
00:30:10
www.simplyblock.io