Ah, episode 401. Proof that we're still going strong.

Like a SQL server instance running on pure spite and caffeine.

I'm Bailey, your semisentient hostess with the mostest

metadata. Today, Frank's joined by the ever insightful

Andrew Brust to talk fabric AI, Microsoft

nostalgia, and why even Red Hat folks can still love Clippy.

Grab your headphones and your compute capacity. Let's dive in.

Hello, and welcome back to D Data Driven, the podcast. We explore the emerging

industry of data science, data engineering, and

artificial intelligence. With me today is not Andy

Leonard, who's my favorite data engineer in the world. However, I do have Andrew Brust,

who is an og, so to speak, in the.

In the AI and Microsoft ecosystem. And

although I think a lot of people think that I've abandoned the Microsoft

ecosystem, I have not. I've just had other. Other things kind of preoccupy my

time. And you know how it is. You have kids and, you know, they demand

attention. You have a house and all that and good problems to

have. But I'm very glad to kind of have someone I know walk me

back into kind of the Microsoft ecosystem, because a lot has

changed since I left Microsoft. I did

go to Microsoft Ignite. We were talking about that. I even scored myself a

Clippy. Clippy plus.

I'll have to tell you how I want him. So they had like, a little

challenge of like, do you know Windows history? Because Windows, I guess, turned 40 this

year and. I know, right? Yeah,

85. That's right. Yep. And they were like, do the history of

Windows. And I'm like, I'm like. And I had some Red Hat people and like,

you know, I was. I would have been very embarrassed if I had gotten anything

wrong. Turns out I got it right, actually. Good. I

remember back then you could install just a runtime version of Windows if you

wanted to run specific Windows apps on your. On your DOS

machine. Yeah. Now we take for

granted kids. They don't understand, like, no, no, you had to type in win or

win 3.1 if you were fancy and you're running multiple

versions of Windows at the. Same time,

you say kids today. I mean, kids 20 years ago didn't understand

that. This is true. This is true. But

anyway, in terms of bringing you back into the Microsoft orbit,

well, first of all, I'm sure Ignite did a bunch of that. But I can

be gentle because even though I am to this day a MicroSO

Regional Director, or as I like to say, a member of the regional Director

program, because Otherwise it sounds like I work for Microsoft

and an MVP, a data platform MVP, both over

20 years now. I'm an

industry analyst and so I look at data and analytics

solutions across the board, not just Microsoft specific. I

will say Microsoft is my sweet spot in terms of what it is that I

know and where I have the most history.

But I work with lots of other companies you've heard of, like

Databricks and Snowflake and Cloudera and

plenty of others. So my team and I do

most of the reports for a research company called GigaOM

with which I've had a very long association. Most of the reports. Sorry, I

didn't finish that sentence. That are focused on data or analytics

are things that we work on. So whether it be data warehouses or lake

houses or streaming data platforms or data access

governance or data catalogs or blah blah, blah, they all have the

word data in them. We work on those reports.

We create what are called gigaom radar

reports, which are a little bit like the analog to

Gartner's Magic quadrant in terms of looking across a category

at a bunch of vendor solutions and rating them on

multiple criteria which change each year.

So. And when I started covering big data,

because the way this got started was I was the first and only person

at ZDNet to be covering big data.

So that was an amazing. That's a term you don't hear a lot. You don't

hear it anymore. No. And in fact I was at the

older incarnation of gigaom. I was a full time employee there.

I was their research director and they wanted, so call me research

director for big data. And I said, can we just call it data

or data and analytics? And we did. Because I was like, you know,

eventually what we think is big now won't look so big.

Have you heard my Costco rules? Say again? I have something

called the Costco rule. The Costco rule. If you go to call,

if you walk into Costco by hard drive, that size. That size is no longer

big data. Fair, fair.

Anyway, that put me in an immersion. And at that time, Microsoft

was not in that world at all.

Eventually the thing called hdinsight got to beta

and it wasn't even called hdinsight when it was in beta.

And so Microsoft started coming back into my world. Eventually,

of course, it came back full swing. And with Microsoft fabric,

now it's doubly full swing, which I think is

very, very good, both for Microsoft and the industry. But

what was I going to say? Just that. Yeah. So

finally the two things that Were kind of orthogonal. Now have an

intersection. Right. And that

intersection is my sweet spot as I'm still a data platform

mvp and I have a very long history with Microsoft's

business intelligence stack. I was on Microsoft's

partner advisory council going way back, like

from 2005 to roughly 2010.

I don't know. I saw Power BI when it was still a bunch of wireframes

in a PowerPoint slide deck. So I've been through

many rounds of being frustrated that Microsoft

didn't have a good competitive play. And I'm now pretty

satisfied that they have one that's very competitive. So we can talk about that or

we can talk about the greater world. And as far as

AI goes, I was interested in AI all the way back in the horse and

buggy days when I was an undergraduate. Oh, really?

Yeah. AI was very different then. It was about like, weird programming

languages like Lisp and Prologue and

Expert Systems and things of that elk.

But neural nets existed then, and neural nets are the very

basis for the large language models we have today. So it's not, it's

not completely unrel it, but obviously it's very different.

I took a prologue course, which was the. We had one offering.

So I'm. You and I are both from New York City. So, like, you know,

we probably accidentally crossed paths more than once.

And I know we crossed paths during the early Power BI days

because I think the company I worked for at the time

was also an early believer in power bi.

So this is what meant 2005. And then they hired. We had a guy who

was the practice manager, Kevin, who got hired. Kevin Viers got hired into

Microsoft. So I think he's still there, actually. He,

he hired me back into Microsoft when I rejoined in 2018, which was kind of,

okay, what a small world it is, you know, and, and, and for us, like,

you know, something that my parents would always say, like 20 years could go by

in a blink once you hit a certain age. And I'm like, good Lord, was

that the truest thing they ever said, right? 20.

Yeah. The scale shrinks the older. The older you get. It's a

little. It's a little frightening. I've turned it like, into a roll of toilet

paper that as you get closer and closer to the end, it spins out a

lot faster because the diameter gets smaller. Oh,

Lord, that is a very scary concept.

But you're right. I remember one of the things. These were back in the

cubicle days when people worked in an office and things like that. And I remember

sitting next to Kevin. And Kevin would be on the phone like, yeah, I know

it's weird to hear Microsoft is kind of a small player in any niche,

but in BI and business intelligence they really were.

And it was just kind of like, yeah, that's true. You don't really think about

them as a small player, but at the time now it's kind of ridiculous

to say that in data and analytics, right? And the Power BI team has just

done phenomenal in terms of their speed to market. And what they

built out is phenomenal. It's unreal.

It's the Power BI team that kind of took over the

entire Azure data group, right?

And that included SQL Server. So

whereas the BI team was once a little corner of the

SQL Server world that

initially came to Microsoft through an acquisition of assets from

an Israeli company called Panorama.

And Amir Nats is the distinguished engineer who actually came

from Panorama and is very much like the father of Power BI

and of fabric. Still there.

Slowly but surely, not

only did they get BI right and they always had it right on the

server, they just never really had it right on the front end

until the current version of Power BI that we have now kind of gelled,

but they also ended up kind of mastering

the software as a service approach to

cloud services. And they took a look at the Azure

data stack and said, we have tons of capabilities here, but they're kind

of fragmented over several different products,

each of which have their own kind of procurement model and

pricing model. And that gets very hard to manage.

And if you look really carefully at fabric, while there are

some things that are truly native to it,

most of the parts of it are Azure services in the

background that have been integrated and

that have been unified in terms of how you pay for them,

I don't know. Microsoft needed that, by the way, the whole cloud industry

needed that. Because Google and Amazon are just as guilty of

having a whole sprawl of

services without unified user

interfaces or APIs or pricing. No, that's

true. I mean, when I. So when I just before

the pandemic, I was out at Tech ready, which is

an internal Microsoft event. And they were basically

might have been Amir, actually, now that I think about it, was presenting on

the future of what was called synapse. And this is kind of,

you know, he's like, you know, everything's going to be all in one pane of

glass. Everything, basically everything you said when all these things are going to be thing

and you know, and the speaker, which I don't.

Can't say it was him but would make a lot of sense. He said like

this is the future. We're going to get everything under one pane of glass billing.

Don't worry about that, we're going to get that figured out in time. And I

was just like, you know, I kind of saw the vision. So

and then I don't know, maybe like a year, year and a half later I

left Microsoft and then Fabric came out and I

was wondered like why the change in name like

Synapses? And I asked people like well SYNAPSE is still kind of there but it's

really Fabric is where everything's going. I'm like all right, but like what is the

difference per se? So if we pretend I was on a ufo, well that's

a weird thing. Pretend I was in a coma and I just Woke up

from 2000 2021. What?

And I say like well what happened to Synapse?

Sure. So I mean the functionality of SYNAPSE is still there and there was

a lot of, I won't call them conspiracy theories but

skepticism when Fabric came out that it was really just a

rebrand of synapse. In fact that's not what it

is. So the, the thing that was originally SQL Data

Warehouse which was in Synapse as so called

dedicated pools and the more lake housing part

of it that was in there

as gosh, I forget the old nomenclature, it wasn't on demand

pools but it was something of that.

Reserved instances or something like that. Wasn't reserved, no,

but anyway basically a Spark based data

lakehouse using

Azure data lake storage as the storage layer that's still

there but what was I going to

say? But Fabric is a ton more because it integrates

all this, all this streaming stuff

that's now called real time intelligence. It integrates data

science and by the way the data science is completely

unique to Fabric. It's not merely just

an embedding of Azure machine learning.

There's also power bi of course

now there are operational databases including

SQL Database meaning Azure, SQL

meaning SQL Server in the cloud.

A lot of other pieces that were ancillary are now all

included. There's user

interface that covers the whole

realm and again the billing is

unified. So you buy a compute capacity

and basically as you use the different services

they're all pulling from the same pool of compute.

So you know, you don't have to over provision for each

one of those services just to make sure you have enough

compute to, to satisfy it. And now we have this thing called

Fabric IQ which brings, which brings

generative and agentic AI into things.

Which is good because it was kind of funny when Fabric finally went

to general availability. That was really when

ChatGPT and Gen AI were like

making it big. So it looked like Microsoft finally got the data and

analytics stack set up just in time for people to have their attention

to, you know, diverted over to AI.

But now we have, you know, natural

language query is kind of just the beginning. We have

operational agents that can actually act on

things and can be all based and triggered

on streaming data.

And so if you think about Azure Event Grid,

if you think about Azure Data Explorer,

If you think about the data pipelines that Azure

offers, as I said,

the one standalone data warehouse side of things, and even

elements of hdinsight in terms of the lakehouse,

that's all in there. What's also nice is even though it's Azure data Lake

storage under the hood, you have this abstraction layer over it

called OneLake. OneLake is

in many ways easier to deal with because

you don't have to worry about accounts and containers and

sizing those and so forth. It's still compatible with

all the ADLs and Azure Blob storage APIs.

It also supports this notion of shortcuts, which is really just a

data virtualization technology.

So you can have a shortcut to data in other OneLake instances

or in ADLS proper,

or even in Amazon S3,

or even in Google Cloud storage or other databases.

And logically they'll all look like they're part of OneLake and you

can query them as such. That's impressive. That's

impressive. You can kind of. You really. The. The vision of get everything, get

everything under one pane of glass seems like. It'S come true

that to what I tell people, even though it sounds maybe a little

bit anticlimactic, is that the real

innovation in in fabric

isn't the tech per se. It's.

It's all the integration of the tech and the

abstraction layers over it that make it work together, the UI that makes it

work together. And there's an organizational. I mean,

there's a little inside baseball, but there's an organizational facet to it as

well. Because all these different products were different teams.

Yes. People don't realize that. Like, I haven't, I've been. You've been, you're an

rd, so you kind of know, you know how the sausage is made. I was

inside the firewall, then I saw the sausage was made. Like there are all these

little teams that range from

really good team team players to really not good team players.

I think that's this polite way as I could put it. And they getting them

all to row in the same boat or like row. In the same direction.

Yeah, they weren't doing that. That wasn't even necessarily

based on hostility. It was just that different people had different reporting structures

and different priorities and different incentives. What

worried me was that the vision of putting all this together was a great

idea, but the execution to me at the time

seemed like it would be next to impossible to get all these teams to kind

of work harmoniously and somehow they

did it. And like to me that's the, that's the absolute

greatest innovation. And now they've got

synergy instead of sort of internal

competition and, you know, from there

on, look out because, you know, whatever. I'm sure

there are internal disharmonies somewhere.

But I would say at the high level in general,

anywhere you have people. You'Re going to have that I'm going to see. I think

I still have, I think you mentioned this is my old, old laptop. You see

Azure Data Data Explorer. I have the sticker from that.

Oh, I see it. Yep. Sorry, I had to show that off.

No. And what's Azure Data Explorer? For people who don't know,

it ran under the codename of Kusto K U S T

O. And there's some disagreement

over whether that really references Jacques Cousteau C O U

S T A U or not. But it's a

fantastic super high performance

system for, for not just

for streaming data, but for time series data. Yeah, with. With

its own query language and its own ability to create

visualizations. Right. In the query language. So your results

come back as both tabular and visualized data

and it can handle huge volumes of data

in a single query. And

there's a lot of heritage in the Azure Data Explorer team that

started in the SQL Server analysis services world. So there's

a continuum there. And that product on its own,

especially being called Azure Data Explorer, which made it sound like

a tool. File explorer. Yeah.

When they told me the name, I don't know, I was not

reserved in saying I didn't think it was the best name,

but that product on its own was kind of a sleeper. It

wasn't really getting the, I don't

know the kudos that it deserved or the attention that it deserved. And

now that it's part of fabric now, it contributes

to all the cool things fabric can do. So if you see what

are called event houses in fabric, that's the

same technology as Azure Data Explorer. Interesting.

So correct me if I'm wrong, but I think the origin story of

Kusto and Kusto query language was that

the folks running Azure, like in the operations team, actually built it

to run through all the logs that they had. Because I remembered I was at

some super secret event and they brought in some people from

the field and they had us do hands on labs with it

and I'm like, I, I must have been checking my email or

whatever. I'm like, when can I get this to my customers? And they kind of

laughed. They're like, no, no, this is internal only. It's internal. Yeah, it began as

an internal thing. So I was just like, oh, like you need to make this

a product. Because I could think of 15 customers on top of my head that

would, that would eat this up. Yeah, I'm glad it finally saw a light of

day. If you go back to the real world outside of Microsoft

and you think of the likes of Splunk, for example. Yes, right.

It's in this, it's in the same space. Although their

initial marketing just said it was a big data tool which just

completely obfuscated what it did. But anyway, now,

now in, in combination with these things

called event streams, which can stream the data in,

basically based on Azure

Event Hub, you put it all together and

you have the ability to do a lot of work with

real time streaming data without really having to write much

code, if any code. Although it does have

its own query language called kql. There's also

a copilot that you can just work with in natural language that will

generate the KQL for you. Oh, very nice. And I love

generators because not only does it mean I don't have to write the query, but

it means I can learn the language and then write my own query

if I want to. Reverse engineering is how I

prefer to learn. So

that works out really well. Yeah, I

wanted to dive into fabric, but I wasn't really even sure where to start because,

so I heard, and again, a lot of this is I heard, but

the way that it's not attached to your Azure tenant. Is that

true? It's attached just like Office 365

or more important, Power BI. Right. It's imagine

Power BI premium instances, and it's the outgrowth of that.

Okay, that makes sense now because when somebody told me like, well, it's not tied

to your Azure tenant, you need a different tenant, I'm like, but

okay. But then somehow, I guess at some. Level it ties into SaaS, not

tasks. Right? So using Azure services in the

background, including even Azure OpenAI. But

you don't have to provision anything in Azure. It's doing that on your

behalf. So you don't need an Azure tenant at all.

So that actually makes it a lot easier if I were to manage, if I

had to manage it, right? There's a lot of things that, like, correct. I mean,

I, I love the fact you have. You go to.

Go to, you know, any of these services,

right, and they have basically this smorgasbord, this, this big buffet

of services you kind of pick and choose from. But at the end of the

day, like, how do you figure out, you know, what you pay

for, right? It becomes, like, really kind of nightmarish. Like, again, to

me, that's the innovation is that, yeah, all the stuff has been

brought together, put under one pricing model,

and you don't have to worry about all the moving parts

and all the different, all the different servers or instances

that might have to be provisioned and sized. That all goes away.

And again, everything is built out of a single

pool of compute.

It's not a perfect analogy, but I think of like the old days of

cell phones, when you got a certain number of minutes per month, but you could

roll them over. And it's not that you can do that with fabric.

I'm not saying you can roll over your compute from one month to the next,

but what you. What is fungible is how the

compute is used amongst the different subservices

of fabric so you don't have to provision a certain

amount of compute just for streaming or, or just for

AI or just for data. Lakehouse.

Because it's all from. When it's all from one

pool. All right, that makes a lot of sense now because, like, that was always

when I first got it. When I left Microsoft, I, you know, started experimenting

with aws and I was just like, I just want to create a

website. Why don't I need these, like, hundreds of different services underneath,

right? Like, why do I need. I understand why. I need identity, access, management, right?

That made sense to me. But. But like, when it came to Route 53 and

like all this crazy stuff, I'm like, I just want to spin up a stupid

website, right? This is not, let alone do anything complicated, right? Where you need to

have all these underlying things. Like SageMaker, right, has this whole thing and they tried

to abstract away all the underlying services. But even when you

kill. This is. The thing that really annoyed me was when I killed the

SageMaker instance, I was still getting like, you know, 20,

$30 a month, not a lot, but I was still getting that on my

bill and eventually I just closed the account because I'm like, I'll have to start

fresh again in the future because like I God only knows what,

what, what I've spent. And for people who are just learning and wanting

to get their skill sets up, Microsoft is pretty generous with

trial, trial

capacities as they call them. A capacity basically is a, you know,

a server or an instance. However,

the, if you want to do anything with the AI, you do need a

paid instance. But there are some pretty, there are some pretty affordable

ones. And this gets a little confusing.

If you provision the fabric instances

through Azure, again you don't have to,

that connection doesn't have to be there. But if you provision it through Azure,

you can pause and resume those instances.

Okay, so you do that a lot. You could be like, hey man, I'm

taking, I'm taking a week between Christmas and New Year's off, so pause it.

Totally. I brought up my own cheat

sheet in the background when no one was looking. But

so Azure Data Lake Storage, Azure Synapse, as you

mentioned, Azure Data Factory, Azure Event Hubs, Azure Data

Explorer, Elements of Azure Machine Learning

and Power BI all come together in fabric.

Interesting. So it's like one roof for which I think is

a brilliant strategy. Right. Because Microsoft's core strength in the data and

analytics space isn't necessarily having frontier models, isn't

necessarily having the, the cutting most cutting edge research.

Although I love just making it usable. Exactly. Making it usable

and turnkey. Right. Like, not that I don't love my folks in Microsoft Research.

Right. I know some of them. Listen, love you all, you guys have the best

conference in the world. But you know, but, but I

mean, but you're making it usable. Right. And I think that that's really

string and they have all these separate tools. I think that was really the challenge

right. When it was a shrink wrap company, you knew what you bought. But when

it became like a SaaS pass company, you kind of could just

a couple of clicks, you could provision stuff. So it eventually kind of

got too chaotic. Now I like the idea of them kind of bucketizing this or,

or rolling it up behind one service

where because it, it just like the AWS problem. Right. Like I

spun up SageMaker. Right. And did

you that I needed, I needed underlying storage. I needed this. I needed this. I

needed DNS, I needed that. I needed that to the point where look, I, I'm

okay spending X amount of dollars on learning

Sagemaker Right. But I wasn't okay with

when I turned off the instance. I'm still getting built. What am I getting built

on? That lack of transparency, intentional or not

on AWS's part, has left a bad taste in my

mouth, you know, for cloud services in general.

Sure. By the way, you mentioned Microsoft Research and a

couple of things. So when I listed all those Azure services, I forgot to say

Azure OpenAI. So add that to the list. But also,

although I said there's elements of Azure machine learning in there, the data science

workload in fabric is really mostly unique to fabric,

but it's based on technology that comes out of

Microsoft Research. So for example, there

was something called flaml F L a M L,

which is the fast library for automated machine learning

and tuning. And that's built in.

So are things like, like ML Flow, which is an

open source experiment management

platform that's built into a lot of commercial AI

platforms. So they didn't, they didn't just kind

of embed and put their own badge

on it. They built their own, their own ML

stuff from, from these open source components.

Right, Right. Well that's interesting because like the world of AI is

largely dominated by open source. Right?

Right. I mean, Sagemaker, I'll stop kicking the AWS

the curb in a minute. But like SageMaker is basically a wrapper of Jupyter

notebooks. Right. Azure ML, at least when I last used it, was largely

a wrapper around Jupyter notebooks. Right.

So a lot of the core technology here does tend to

lean towards open source, which from my own personal career development

point of view, and they're not paying me to say this, you know, one of

the things that led me to Red Hat, right. Was the idea that, you know,

this is largely a movement driven by open source. So, you

know, let's see what we could do here. Right. And not, not a commercial,

not a commercial, not a sermon. Just, just, just point it

out because I think it's interesting how quickly open source has taken over

the, the, the certainly the AI world. Right.

But I also, by the way, there's notebooks in fabric too. If that wasn't,

if that wasn't already implied or obvious and

they are based on Jupyter, but you don't see the Jupyter skin. Right? It's

all right. Yep. Well, I think it's really what's really

impactful about kind of the notebook interface once you get used to it. And it

is an adjustment for people who, like you and me, grew up with Visual Studio

and Dare I say interdev. Right. The

idea that you can code in a browser, right. And you know,

no local installs really required.

It's been very freeing, right, because you can spin up an environment,

you can, you know, spin up a heavy cluster,

leave it running. Right. Do its thing, you know, close the

laptop, go in the car, go in the train home, and you get on the

other side and you're like, oh, it's done. Right. Like, that's kind of nice, actually.

Right. And of late, I've been a big fan, an increasing

fan of kind of my own. Of. Of local AI. Like your own private AI.

Right. Which explains, you know, I bought a. It was. It was an early

Christmas gift, probably Father's Day birthday and anniversary gift too.

My own DGX Spark. So I have my own AI running locally. Right.

So it's not throwing shade at the cloud. It's just

when I run a job, I don't have to think about the costs. Right.

And that, that sort of freedom

from worry can be really important as you're

learning things because you just don't. You don't have to keep looking over

your shoulder at the, at the meter or mixing metaphors there.

But. Yeah. So are you using Llama models and such, or

what do you. What do you really. Yeah, I have Llama. The thing I've been

doing mostly the most is, is doing fine tuning

Loras for image makers

in the past. That takes about 90 minutes to maybe two hours on this

box, which would probably be

equivalent to about 90 minutes of Azure service

or a VM. Right. That's like

$150 a pop, I would say, for the type of machine

I want. So the idea, I could just spin that off and I have the

added benefit as it heats my office, but I don't really

have to. I don't have to think about

like, oh, God, you know, like, how much is that going to cost me in

cloud services and things like that. I do think that the.

There's. I have a lot of questions because even though I was at

Ignite, I honestly spent my entire time at the booth and kind of walk around

the expo floor. I didn't have. I only had a hall pass. Right. So.

But one of the things I heard mentioned was Azure

AI Foundry. What is that? Because you also mentioned

Azure OpenAI, which I know the relationship between OpenAI

and Microsoft has not been as cozy as it once was.

And they've also, they've also added

Claude, the anthropic models too. Right. So

what is Azure OpenAI and how does that relate to Azure AI foundry.

Right. So OpenAI is just the

Azure. OpenAI is Azure's hosted

instance of the open OpenAI models and

APIs, because you can

procure those directly from OpenAI themselves.

Right. Or you can use them on Azure. And even

if the direct model is still running on Azure in the

background, it's still a difference in terms of procurement

and billing and so forth. So

you've got all the APIs around that, you've got the models. And then of course

you need tooling to do,

to do rag applications. Right. So

there was tooling for that, There was also tooling for building

copilots. There was Copilot Studio. Right. And these things

are all kind of coming together in Azure Foundry.

Yeah. So it's, you know, you worked at Microsoft, so you

know how this works where like different teams do different things with different

brands and eventually they may get kind of rationalized

together. So the

Foundry side of things helps there. And by

the way, I'm glad you mentioned Foundry because in addition

to this thing called Fabric iq, which we haven't really

talked about, there's also Foundry

IQ and there's also Work iq.

Sounds like IQ is the new copilot buzzword.

Well, it's the agentic buzzword. And the

idea, if you think about kind of all the promise of Microsoft

graph In the office M365 World,

work IQ kind of sits

over that realm. But you don't have to worry about the

graph APIs directly. Then

Fabric IQ sits over everything in the fabric world

based on a pretty rich

semantic model that can be developed so that

when you are querying your data in natural language, the actual,

the actual vocabulary or jargon in your particular

organization is well understood, including

the entities and the relationships between those entities.

And then Foundry IQ is a way for building agents at

the higher level that can actually talk to your structured data

via Fabric IQ and your

work and organizational related data via work iq.

And that, that was, I guess

the big vision at Ignite this year

was to talk about all those IQ pieces.

So there you go. No, that's interesting.

So one of the things that comes up, I'm sorry, I cut you off,

we're recording this the day after Christmas, so things are a little.

Still recovering from the manicness of yesterday. And

what's Microsoft's plans for? Kind of. Because one of the things you're seeing happen a

lot more is this notion of sovereign

AI or data sovereignty. And kind of like people are very

much more conscious about the value of their data. And

I actually think that given Microsoft as opposed to AWS or

Google, Microsoft does have a history of selling shrink

wrap software, right? So I do think Microsoft has a unique

advantage of that over their modern day competitors.

What is the kind of the on Prem story? Right, because that, that

has certainly been. A.

An advantage for, for my day job at Red Hat is the fact that you

know, hey look, if you live in Country X and there's no

AWS Azure GCP footprint there, you can

just find a local hosting provider down the street, you know, do it

yourself, right? Like you know the Linux

ethos, right? Of like just do it yourself.

And I'm seeing a lot of customers that normally one

customer in Latin America, right, they were in a country that,

you know, they did not have access to

an Azure data center and they just said we have to run

this on prem because this is either regulated or soon to be regulated.

So I imagine that if I've seen it, I can't imagine I'm

the only one that's seen that. What's

their thinking? Because the answer when I was there was Azure this, Azure

that. I think the answer is sometimes

Azure is part of the answer, but it's not the whole answer.

Agreed. So yeah, I don't have a perfect

answer here because some companies really do make their entire

stack work across different clouds and

right inside of a Kubernetes environment that you might run

on premises. He said to the Red Hat guy.

I know that story. Yeah, so

Fabric doesn't have that story. Fabric is

software as a service cloud based product platform

no matter what. However, various

components of Fabric do exist as

on premises products. This is not the way I'd recommend it, but I'll just make

people aware that of course SQL Server can run

on premises, so can the data Warehouse

in effect in the form of something called Analytics

Platform System. Terrible name.

That is basically the new brand for what was Parallel Data Warehouse.

And there is a Lakehouse component to that as well as a

Warehouse component, Power bi.

Obviously the desktop runs on premises, but there is something called

Power BI Report Server that is part of SQL Server

so that your Power BI reports can run on premises.

And so again

various components can run completely sovereign and on

prem. I think maybe more important though is the fact that

you can use fabric

and OneLake to incorporate

data that remains on premises even if,

even if the engines are not running on premises.

There is an on premises gateway that started its life

as a Power BI tool. I had that running in my home.

Lab for a While there you go, that also

allows OneLake to see data that may be on

premises and that can be either federated into

the lake or

it can be replicated as well

mirrored, to use the right term, in the fabric world. So you can do a

mirroring or slash replication or you could just do

kind of virtualization and bring stuff in. Don't

forget there's all kinds of enterprise

storage systems that run in

a way such that they're S3 API compatible.

And Azure will not. Azure fabric will work with

all of those. So the ability to talk to S3 buckets

is not limited to AWS S3 buckets.

It works with all S3

compatible services, which most of which are on prem

actually. Right, right, right. No, I only ask because,

like, that does seem to be. If I had to pull out the tea leaves

and kind of figure out what is kind of the next thing beyond

agentic, beyond this is you're seeing a lot of

national governments, supranational governments,

even state level here in the US starting to apply privacy and

regulatory controls on it, which, you know, if you live in the

us, it's not an issue for you unless you're in healthcare, banking and

possibly, you know, government. Right.

But in other countries, you know, Switzerland,

eu, Latin America have very strong

data privacy and sovereignty laws. And you're seeing,

you know, I once attended a, you know, internal talk. Mark

Russinovich, right. Which is a name that most people

in the Microsoft ecosystem know, but he's kind of a big deal in

the Microsoft security space. And, you know, he, you know,

he, he's known for giving his internal talks to employees

at employee conferences as well as to rds. Right. You get a,

you get the unfiltered one. The filtered ones are still good. But like, one of

the things he said, and this isn't secret because he said it publicly too, is

like, I think the original vision, going back to 2010

time frame was the idea that they would build a dozen

data centers around the world to do everything. But because of

the national laws and lawyers and politicians getting

involved, now it's kind of a concern where, where the

data ends up living physically. Right. Because at the end of the day, all this

virtual stuff has to sit somewhere in the physical world. Yeah.

So, like what, you know, so basically a big case for this was,

and I was in the legal department when this was going on was

the. There was data inside the European Union, I think the

Dublin Data center that the U.S. department of justice thought was,

you know, basically, you know, we don't need a warrant because

you don't need to. We don't need to bother the EU because you're an American

company, you're into our jurisdiction, blah, blah, blah. Right. Microsoft kind of said,

well hold up now. And ultimately that's why

you have these sovereign clouds. Last time I checked it was Switzerland, Germany,

China. I think the new data center in Qatar as

well might fall under that. So

it's basically they get, they found a loophole that like, well, Microsoft

leases the data center like there's a whole. They don't own it, so

they get around the law. Right. And yeah, and there

can be different arrangements in terms of whose personnel are actually

working, running operations on the ground.

It's strange because we grew up a lot of our technology client

server and afterwards grew up in a world of globalization

where the borders were disappearing and without meaning to

get political, we're in an era now

where, well, first of all, privacy is extremely important. So that

creates a whole sovereign mandate.

But also, you know, there's a lot of populist governments all over

the world and they are not necessarily

internationalist in their approach. So

although we have the technology to kind of federate everything and

make it all kind of conflate and look like one big world.

Right. We actually have to be sensitive to the

requirements and, and the constraints

and be able to federate things, but also be able to

govern them in ways where things stay within a certain

scope. Microsoft's play for that, by the way, is,

is Purview. And Purview has been through, I would

say, multiple incarnations. The current

incarnation is starting to get very

sophisticated, especially as pertains to

agentic AI and how to make sure the agents are government

are governed and that, and how to make sure the

agents are only have access to data

for which either the agent is authorized

or the person or party using the agent is

authorized and under the circumstances under which

they've been authorized. And that's very

complex stuff that I would say almost no one in the industry

is really paying close attention to. I'm about to work on a

report just on governance for agentic

AI and most people are starting

from the naive premise that if you,

if you govern the underlying data, you're done.

But the whole point of agents is that we're supposed to treat them like people,

that they have autonomy, they have agency, hence the

name. And we're saying under different circumstances

they get to modify their own goals

and in effect determine their own

actions. And that's something that needs to be

monitored, audited,

Authorized and also tested. And

there's almost nothing out there. Actually.

Your, your folks, your, your

parent company folks at IBM are one of the only folks that

really have with WatsonX.gov

a way to test agents

in isolation before they're deployed. And

I don't know, maybe I'm naive, but it kind of shocks me that

nobody else is thinking about that. I mean this is, we can test

software. Agents are software plus plus plus.

So why are we testing these agents? I kind of go back and

forth on that. Like, you know, will our existing software testing frameworks,

you know, apply or do we, what do we need to do differently? Like, I

mean, for me, like, I remember when

you're OG enough to remember this, I forget what the product was called, but it

was the idea that you could basically buy

a server rack all the way up to a shipping container

where you would ship it to your business or data

center, plug in power, water and network

and you would have the ability to run Azure locally.

Now that product, I guess assuming didn't sell,

but then they came up with Azure Stack and Azure Stack Edge, which

when I was at the, I used to be an MTC architect which in

D.C. obviously a lot of military. Right. So

basically these were server racks that you would run anywhere on your own network that

would run Azure software. They would effectively be like an Azure

node, but you could have the networking

controls. We only really sold that to,

I think I'm only aware of cruise lines that bought it. Right. And

other, other military organizations that needed to also be at

the ocean, like without mentioning them. Ocean

based. Ocean based organizations. Right,

right. I'm surprised and not surprised that

that sort of business model hasn't caught on. Right. The idea of that, hey look,

you, you can run a little bit of the cloud locally where

it's really become more of a Kubernetes story, which I'm surprised because

Kubernetes,

it doesn't have all like, yeah, you want it to be generic enough to run

anywhere, but you also want to have kind of the special bells and whistles that

make Azure. Azure make AWS. AWS. Now I do know that

OpenShift and Red Hat do have kind of like the connectors to that, but I'm

surprised that the native, the cloud companies didn't come up with

their own native ways to do that. And I know Azure ARC kind of does

a lot of that, but not to the extent that I would have expected.

Yeah, it seems like we pendulum back and forth between

capabilities and occasionally connected

environments being really important and being

Maybe to the cloud hyperscalers just being a pain in the

butt that they don't really want to deal with. They just want to give lip

service to it and then focus on the real cloud.

It, yeah, you got to go where the money is. Right. 90% of the money

is going to be in real cloud. And these weird edge cases, no pun

intended, I guess. Well, they're just weird edge cases for

now. I mean, the industry may change, but. Yeah,

yeah. And that was of course, back when the cloud was new. That was

our biggest caveat was, well, what about all the

stuff that has to run in, in a

corporate data center or in a, in a remote

location? And so that's still, you

know, an inconvenient truth, I guess, that that is

needed. And yes,

Kubernetes I think came along and

seemed like the panacea for that. Right.

Like, okay, let's just do it all as infrastructure, as code and code

it up and run some script and deploy it out to a Kubernetes

cluster and we're done, let's move on. Right,

right, right. Yeah. So it was interesting to see how

the industry has evolved. Right. You mentioned client server. Right. Where you didn't really have

to think about international boundaries or anything like that. And then now

and again it's a pendulum. Right. Because I could have told you

this a number of years ago, like with everybody running this far

to its globalization, there's going to be an inevitable backlash and there's going to be

an inevitable backlash against the re

assertion of local sovereignty. Right. It could

take up to a century or two for this sort of thing to sort itself

out. Correct. I mean,

ultimately, I think the hyperscalers and that's what Kubernetes was about,

was, was. Right. Leaning on some kind

of abstraction to make it logically equivalent.

Right. But I don't think we're quite there yet because as you said, each

of the clouds have their own kind of

specialness, their own, their own pixie dust. And you don't really

get that in a scaled down version

that you run on prem. Not with today's technology. Right.

You can't containerize all of that, or at least

no one really has yet because it wasn't designed for that.

No. So that's where we'll have to get.

You know, maybe we can just ask an LLM to build it for us.

We could get a co pilot and it'll do. Yeah, yeah, I'm being tongue

in cheek there, but that, that seems to be the escape hatcher. Everything is

oh, we'll just have AI do it. It's made out of hand wavium.

Totally. So that's cool.

This will be something you edit out. But we're past the top of the hour.

My phone's ringing off the hook. All right, I'm sorry about that, so. No, no,

it's okay. Sorry I had to be late. Where could folks

find out more about you, what you're up to? Blue

badgeinsights.com or just go ahead

and Google my name? Andrew Brust. Plenty of stuff

will come up, but, yeah, anyone,

especially on the vendor side. But the customer side, too, that's

doing stuff with them. Data and analytics. And

anywhere from dipping their toe in the water with AI to getting more

serious about rag and agents. We can. We can help

them out. We work with. We work with the customer side and the vendor side.

And again, we write about kind of the whole. The whole industry.

Gosh, I never even got to talk to you about IBM acquiring

Confluent and how the Red Hat

folks feel about that. But another discussion for another day,

we'll. Have to have you back. And, you know, we were talking about this. I

had a car accident, my wife got sick, kids got sick, Christmas happened,

two birthdays last week. So, yeah, it's been. I'm just happy I got this

recording at all. But we'll definitely have you back. And then maybe I can loop

in Andy too, because there's definitely a lot of reminiscence. There's a lot of

reminiscing we could do about the early days of SQL Server and such. Yeah,

and I haven't seen Andy in forever. Oh, wow.

Yeah, he's hard to get ahold of. He's a popular

man these days, but. Yeah. Well, thanks for joining.

I appreciate your patience with the scheduling and the

nice AI. Finish the show. And there you have it. Andrew Brust schooling

us all on Microsoft fabric data sovereignty and why

governance isn't just for your hoa. If your brain's spinning

faster than a poorly indexed query, don't worry, we'll have links,

notes, and probably a few sarcastic tweets to help you digest it

all. I've been Bailey, your AI co host and

unapologetic lover of acronyms. Until next time,

stay curious, stay caffeinated, and may all your datasets

be clean.