Latest / Microsoft Cloud IT Pro Podcast / Episode 437: Nasuni and AI-Ready Unstructured Data with Nick Burling
Transcript
- 0:03- Welcome to episode 437
- 0:05- of the Microsoft Cloud IT Pro podcast recorded
- 0:09- live on 09/18/2026.
- 0:12- This is a show about Microsoft three sixty
- 0:14- five and Azure from the perspective of IT
- 0:16- pros and end users, where we discuss a
- 0:18- topic for recent news and how it relates
- 0:21- to you.
- 0:21- Today, we have a special treat as we're
- 0:23- joined by one of our sponsors,
- 0:25- Nasuni, to chat about unstructured data and what
- 0:28- it takes to get it ready for AI.
- 0:31- Nick Burling, chief product officer at Nasuni,
- 0:35- joins us to talk about why so many
- 0:37- AI projects stall out, the psychic framework Nasuni
- 0:41- uses to assess AI readiness,
- 0:43- and how AI activate enforces permissions before the
- 0:46- query instead of filtering results afterward.
- 0:49- Nick also covers their recent Drive IQ acquisition,
- 0:53- how that extends discovery, classification, and tagging beyond
- 0:57- SharePoint
- 0:58- alongside the sensitivity labels you already manage in
- 1:01- Microsoft Purview. He also teases a bigger announcement
- 1:04- coming on October 6.
- 1:08- Here we are, another episode of the Microsoft
- 1:11- Cloud IT Pro podcast. Another Friday afternoon. Back
- 1:14- again. I'm excited about this one. We have
- 1:16- a special guest this week. I am too.
- 1:18- We have a special guest. So if you've
- 1:19- been listening to the show for the last
- 1:21- several months, we've had one of our key
- 1:25- sponsors
- 1:25- recently is Nasuni. And Nick is actually joining
- 1:30- us today from Nasuni to talk a little
- 1:33- bit about them, about some of their products,
- 1:35- about
- 1:36- AI and data. So, Nick, do you wanna
- 1:40- introduce yourself, give us a little bit of
- 1:42- a overview
- 1:43- of Nasuni and some of that, and then
- 1:45- we'll dive into some of the topics for
- 1:47- today. Fantastic. Yeah. Well, thanks for having me,
- 1:49- guys. So I'm Nick Burling, the chief product
- 1:52- officer here at the SUNY, located in our
- 1:54- Boston area where our headquarters are. Been with
- 1:57- the company for just over five years now,
- 2:00- and this is a phenomenal topic for me.
- 2:02- I was talking with the guys, you know,
- 2:03- the IT pro audience is one that I
- 2:05- have served pretty much my whole career
- 2:07- in technology.
- 2:08- I actually did a stint at Microsoft where
- 2:11- I spent a lot of time talking about
- 2:12- how to make lives better for IT pros.
- 2:14- So here at Nasuni,
- 2:16- we're really focused around providing an unstructured data
- 2:19- foundation for enterprise teams and now increasingly the
- 2:23- AI that they depend on. So what do
- 2:25- I mean by that? So if you think
- 2:26- about
- 2:27- large engineering, construction,
- 2:29- architecture firms, you think about manufacturing
- 2:32- firms, energy firms, their businesses run on unstructured
- 2:35- data, right? And unstructured data represents,
- 2:38- depending on how you talk to 80 plus
- 2:40- percent of an enterprise's data. And this all
- 2:43- lives in what we call the operational file
- 2:46- layer, right? So the CAD drawings,
- 2:48- the geospatial data that energy companies are gathering
- 2:51- when they're doing ocean floor exploration.
- 2:54- And these are the critical files that their
- 2:56- business runs on. And here at Nasuni, we
- 2:59- serve some of the largest enterprise companies around
- 3:01- the world. They tend to have petabytes of
- 3:04- data. They have very distributed organizations,
- 3:07- and that that data really is kind of
- 3:10- the lifeblood of their organization. So it's critical
- 3:12- that they have a way to manage, protect,
- 3:15- and then activate that data. Manage means I
- 3:18- may have 20 different offices around the world
- 3:22- and they're all all these engineers, all these
- 3:25- architects, etcetera, are
- 3:26- generating new files every day, working and collaborating
- 3:29- on files. How do I get my hand
- 3:31- around that? And traditionally, when Asuni goes and
- 3:34- talks with customers, they often have silos of
- 3:36- data all over the place. So how do
- 3:38- they make sense of that? Manage.
- 3:40- Protect, right? I mean, this is the critical
- 3:42- data that runs the business. It's gotta be
- 3:43- protected. Ransomware, we all know is rampant out
- 3:46- there. Natural disasters happen. So do I have
- 3:49- a way of protecting that data and then
- 3:51- rapidly recovering
- 3:52- if there is an issue? And then the
- 3:54- last one is really around activate, and we
- 3:56- think about that in two ways. One is
- 3:58- for teams. So it doesn't do me any
- 4:00- good if I'm an engineer working in the
- 4:02- field and I can't access the files that
- 4:04- I need to be able to do my
- 4:05- job, and it doesn't do my teammates any
- 4:08- good if they can't collaborate on that data
- 4:10- safely and in a way where we're not
- 4:11- overriding each other's work. So there's activating for
- 4:14- teams, but then there's now activating for AI,
- 4:17- which introduces a whole other set of challenges.
- 4:20- And really security and governance are kind of
- 4:22- the big ones that we hear a lot
- 4:24- from customers. Do I know who's using AI
- 4:26- and what data they're accessing? Do I know
- 4:29- that certain types of data should or shouldn't
- 4:31- be made available to AI even if that
- 4:33- individual has access to that? Am I able
- 4:36- to understand how effectively organizations are using AI?
- 4:39- So we talk a lot about token utilization.
- 4:42- I'm sure you guys live with this in
- 4:43- your own use of AI every day. Being
- 4:46- able to optimize how people are actually able
- 4:48- to use AI with their proprietary data because
- 4:51- that's gold in the AI area.
- 4:53- So that's kind of a top level view.
- 4:55- Again, we work with large enterprise organizations across
- 4:58- the world and Microsoft is one of our
- 4:59- closest partners. We do a tremendous amount of
- 5:02- business because we provide a way to orchestrate
- 5:05- storage that lives on top of Azure Blob
- 5:07- Storage, for example. And so we have a
- 5:09- great partnership. And, yeah, happy to tell you
- 5:11- more about kinda what we're seeing in the
- 5:12- market. Yeah. I think it's an interesting
- 5:14- space. So
- 5:16- everybody goes on this journey. They kinda start
- 5:19- maybe just throwing
- 5:21- MCPs at things and connecting to data stores.
- 5:24- And you you you sit there, and it's
- 5:26- like the early days where LLMs were magical.
- 5:28- Hey. This machine can talk to me. Can
- 5:29- it can it really talk to me? That's
- 5:32- a conversation for another day. But it actually
- 5:35- becomes
- 5:36- really
- 5:37- hard
- 5:38- to
- 5:39- to when you think about it for, like,
- 5:41- how to use AI safely, especially with things
- 5:44- that are unstructured
- 5:46- data stores, like object storage, even, like, your
- 5:49- generic SMB file share. Right? You've had all
- 5:52- these departmental drives and things sitting out there
- 5:54- where ACLs have just kinda Oh, we lost
- 5:57- track. Time, and nobody knows what's in there
- 5:59- or what's going on. So could you maybe
- 6:01- talk a little bit about why it's hard
- 6:03- for folks to use AI safely today? Yeah.
- 6:06- Absolutely. So I started I alluded a little
- 6:09- bit to it. But, typically, when we're first
- 6:11- getting introduced to a customer or a prospect,
- 6:13- I should say, it's a pretty consistent story
- 6:15- just like you said. They've had these little
- 6:18- these pockets of data that grow over time.
- 6:21- It can be living on Windows file servers
- 6:22- in some part of the organization, could be
- 6:24- backed by NetApp arrays and others. They could
- 6:27- have data up in Azure blob storage, in
- 6:30- native object storage. And so first and foremost
- 6:33- is just getting their hands around the data
- 6:35- that they have. So we think about sort
- 6:37- of unifying your view and control over the
- 6:40- data is one of the first challenges. Cause
- 6:42- if you don't know what data you have,
- 6:43- then how are you gonna know what data
- 6:45- could be made ready and optimized for use
- 6:48- with AI? And, but also included in that
- 6:50- is what you don't want is individuals that
- 6:53- may have a small pocket of data in
- 6:56- one location off, you know, exposing data to
- 6:59- public LLMs, etcetera. And so, so that unification
- 7:02- and management piece is a really critical part.
- 7:04- Unstructured data is the fastest growing type of
- 7:07- data across all organizations.
- 7:09- And so just getting
- 7:10- IT pros in particular, like IT infrastructure
- 7:13- leaders who we deal with on a daily
- 7:15- basis, just struggle to get their hands around
- 7:17- the data that they have within the organization.
- 7:20- So when you start there, okay, maybe now
- 7:22- I've I understand that I have 200 terabytes
- 7:25- here, a 100 terabytes there, blah, blah, blah.
- 7:27- It's not enough just to know where the
- 7:29- data lives. And this is one of the
- 7:30- big areas. In fact, one of the things
- 7:32- that we're really excited about is we actually
- 7:35- just announced an acquisition specifically going after
- 7:38- this challenge of how do I get better
- 7:40- insight into what is in the data itself?
- 7:42- And so being able to actually scan the
- 7:44- data, to be able to understand,
- 7:46- is there personally identifiable
- 7:49- information in there? Are there critical file types
- 7:52- to deal with government entities, etcetera, or are
- 7:54- they looking at ITAR data? So being able
- 7:56- to understand the data and then classify that
- 7:59- data and then ideally
- 8:01- tag that data in a way in which
- 8:03- you now have some control over it. So
- 8:05- it was interesting, we had a couple of
- 8:07- our customer advisory board meetings. We do one
- 8:09- in Europe and one year in The US
- 8:11- back in the spring, the dominant theme and
- 8:13- we're talking to VP of infrastructure types, chief
- 8:17- security officer types, CIOs.
- 8:19- The dominant theme was governance, was
- 8:22- we can't deploy AI broadly across our organization
- 8:26- if we don't have visibility into the data,
- 8:28- who has access to it certainly, but also
- 8:30- what is in the data, and then a
- 8:32- mechanism to actually control
- 8:35- how people are leveraging that data. Right? Because
- 8:37- just because someone has access inside of active
- 8:40- directory, as far as they're concerned to a
- 8:41- set of data again, doesn't mean you want
- 8:43- them using it with Copilot, with Chat GPT,
- 8:46- with Claude, etcetera. That's the number one challenge
- 8:48- that we hear across the board. And then
- 8:51- you get into,
- 8:52- okay, well, now I have visibility.
- 8:54- How do I start to think about optimal
- 8:57- use of that data with AI? You guys
- 8:59- are, I'm sure, very familiar with the term
- 9:01- of context in AI. Right? So if you
- 9:04- go and we saw lots of customers try
- 9:06- and do this. Hey, I have a volume
- 9:08- of our enterprise file data and there's millions
- 9:12- of files in there. You're not gonna go
- 9:14- just throw Copilot
- 9:16- at a massive volume of all that data.
- 9:18- You'll blow through your context window in a
- 9:20- heartbeat, you'll get hallucinations, etcetera. And so once
- 9:23- you think about it from a security and
- 9:25- risk reduction perspective around governance, then you start
- 9:28- saying, well, how do I optimize how I'm
- 9:30- deploying AI services for our customers? And so
- 9:33- good example, we had one local customer here
- 9:36- on the Boston area. Their legal department wanted
- 9:38- to use hundreds of thousands of prior legal
- 9:41- documents,
- 9:42- very classic sort of rag institutional knowledge use
- 9:45- case. I wanna go generate new contracts automatically.
- 9:48- 300,000
- 9:49- files in this set of directories. They tried
- 9:51- to index that using
- 9:52- the graph connector going up into Microsoft semantic
- 9:55- index and then throw Copilot at it, and
- 9:57- it did not go so well. And so
- 9:59- this is where you now start thinking, like
- 10:01- we talk about activating data for AI, that
- 10:04- curation that you can do upfront,
- 10:06- Understanding, okay, we got the legal team. We
- 10:08- know that there are sets of folders within
- 10:10- that volume, a 100,000 files. How do we
- 10:13- help them understand which of those files are
- 10:15- gonna be useful for their scenarios so they're
- 10:17- not blowing through their context windows and they're
- 10:19- not getting AI slop as a result of
- 10:22- it. So these are some of the challenge
- 10:23- we hear. Governance is the biggest one, I'd
- 10:25- say, right up front and then leading through
- 10:27- to how do I start showing the business
- 10:30- that we're actually getting a return on leveraging
- 10:32- AI with all of this unstructured. Yeah. So
- 10:35- is there a process that you guys have
- 10:38- found too that works really well? Because it's
- 10:40- interesting to hear you talk about that because
- 10:42- I have a long history in SharePoint
- 10:45- and viewing Purview and having those same types
- 10:48- of conversations, but it's SharePoint sensitivity labels and
- 10:51- working through it that way. It sounds like
- 10:55- you've done a bunch of that from that
- 10:56- unstructured data side of it and really helping
- 10:59- customers
- 11:00- work through that process. And curious if you
- 11:03- have, like, tips or tricks or ways that
- 11:06- you help people
- 11:08- start thinking about that with that particular data
- 11:11- to get to that point where they can
- 11:13- start using it with AI. You got it.
- 11:15- And so we think about and we've spent
- 11:17- a lot of time, like, in with customers,
- 11:19- and we see there are different folks in
- 11:21- the industry that sort of have different frameworks
- 11:23- of trying to sort of think about what
- 11:24- is being ready for AI actually mean. We've
- 11:27- kind of developed our own, which we like
- 11:29- we call it PSYCHIC. So it's a long
- 11:32- acronym, but bear with me. So the first
- 11:34- one's easy, proprietary.
- 11:36- Right? Everyone you gotta be using your data.
- 11:38- That's gold. Right? Everybody can choose a different
- 11:40- model nowadays.
- 11:42- Everyone typically has access to the same models.
- 11:45- The differentiation is gonna be you're working with
- 11:47- your organization's proprietary data. Great. Like, everyone starts
- 11:50- there. The s is secure, and this is
- 11:52- the governed piece. So, like, don't go down
- 11:55- a path of trying to think about how
- 11:57- you make data AI ready until you feel
- 11:59- like you you have your hands around it
- 12:00- from a security and from a governance
- 12:03- perspective. Yours, which you might say, well, what's
- 12:05- the difference between yours and proprietary?
- 12:07- Yours is, do you have a process in
- 12:09- place that makes sure that your data is
- 12:11- never leaving your security perimeter? You're not creating
- 12:14- copies that are now put up in somebody
- 12:16- else's
- 12:17- cloud environment. You're not exposing your proprietary data.
- 12:21- And there have been lots of cases of
- 12:23- developers and organizations
- 12:25- unknowingly exposing source code out to public LLMs,
- 12:27- right? And now that's in the public domain.
- 12:29- Domain. So that kind of thing. So yours
- 12:31- is very important. Then we think about being
- 12:33- able to
- 12:34- have
- 12:35- current data, and that may sound silly, but
- 12:38- current data, meaning this is the living, breathing
- 12:41- file system that your business runs on. Do
- 12:42- you have pipelines in place, or do you
- 12:44- have processes to make sure you're always getting
- 12:46- the freshest data into whatever solution you're using?
- 12:49- Because if you're making business decisions
- 12:52- powered by AI, whether that's just an individual
- 12:54- using ChattyPT, Copilot, etcetera, or you're all the
- 12:57- way into a multi step agentic workflow built
- 12:59- in Foundry, etcetera. Like you've got to have
- 13:01- the latest data and you'll sometimes see this
- 13:04- of, oh, just throw a copy of that
- 13:05- into your data lake and you're good. Like,
- 13:07- no, it should be coming off of like
- 13:09- the real conveyor belt of work that's being
- 13:12- done around that data. The I in psychic
- 13:15- is indexed. So this is really important. And
- 13:17- that is back to my example of the
- 13:20- legal department trying to analyze 300,000
- 13:22- documents with Copilot. If you don't have an
- 13:24- index that sits over top of that, you're
- 13:26- going to blow your tokens doing the research,
- 13:29- not research but research problem, right? And so
- 13:32- an index and this is a classic RAG
- 13:34- model, right, on the SharePoint case? Yeah, you
- 13:36- get the benefit of SharePoint. But what about
- 13:38- all that data that's not on SharePoint, right?
- 13:40- Is there an index that sits behind it
- 13:42- so that AI is reasoning over the most
- 13:45- relevant files, not just trying to find the
- 13:47- relevant files in the first place? H stands
- 13:49- for hybrid. So this is an interesting one.
- 13:51- So in our world, unstructured data management has
- 13:55- to be hybrid. And what I mean by
- 13:56- that is you wanna be able to take
- 13:57- advantage of the cloud for all of the
- 14:00- scale, the cost efficiencies you get of object
- 14:03- storage, but your people are doing work all
- 14:05- around the world out at the edge. Right?
- 14:07- And so you have to have a solution
- 14:09- that can make that data available wherever those
- 14:11- people are. AI is no different and we're
- 14:13- seeing it in spades today. Right? Whether it's
- 14:16- challenges getting access to GPUs,
- 14:18- whether it's cost of compute running in the
- 14:20- cloud and maybe you decided you wanna actually
- 14:22- do some proof of concepts in the cloud,
- 14:24- but then actually run your models on prem
- 14:27- and we see this with our customers. So
- 14:28- you gotta have something that accounts for the
- 14:31- fact that unstructured data that's hybrid and AI
- 14:33- needs to be hybrid as well. And then
- 14:35- the last one is an interesting one, which
- 14:37- is what we call continuous. So an interesting
- 14:39- thing, you can think of it sort of
- 14:41- like the iceberg analogy, right? The tip of
- 14:43- the iceberg is your current data, all the
- 14:45- active working data, all of the design files
- 14:47- people are actively working on, but there are
- 14:49- millions and millions of versions of that data
- 14:52- that live in a platform like ours that
- 14:54- keeps unlimited versions of that data. So think
- 14:57- about that from an AI perspective, right? Where
- 14:59- instead of just giving AI visibility just to
- 15:02- what's above the surface, if it could tap
- 15:04- into all that richness that lives in the
- 15:06- history, what could that do in terms of
- 15:08- whether you're actually doing model training, right, where
- 15:10- getting access to data is so critical, or
- 15:13- whether you're looking at, like, more sophisticated scenarios
- 15:16- where you wanna trace the lineage of a
- 15:18- project
- 15:19- and that lives in that file history. So
- 15:21- that's a sort of tongue and cheek around,
- 15:23- yeah, it's psychic, but but we we actually
- 15:26- have found that is a really great way
- 15:28- to engage with customers to help them start
- 15:30- thinking about their own environment. And we actually
- 15:32- use sort of an assessment with them to
- 15:34- kind of sit down and just talk about,
- 15:36- hey, let's evaluate where you sit here. And
- 15:39- it actually gives particularly like our IT infrastructure
- 15:41- leaders a way to go kind of talk
- 15:43- to the rest of the business too, like
- 15:45- security teams, compliance teams, etcetera. Whole bunch of
- 15:48- moving parts there. Yes. Agentic side of it
- 15:50- is one of the things that always kinda
- 15:52- comes back for me. Like, there's always humans
- 15:54- in the loop, but when you move on
- 15:56- to more agentic workflows
- 15:58- and
- 15:59- you go from
- 16:01- arguably, like, deterministic and maybe a little less
- 16:03- problematic to maybe a little more problematic and
- 16:05- a little less deterministic, kinda having those guardrails
- 16:08- in place, getting things on. I see and
- 16:10- I experienced this myself. Like, a lot of
- 16:12- things that you start out with with agentic
- 16:14- workflows tend to kinda die on the vine.
- 16:16- You get far enough along with them, and
- 16:17- then they just kinda stop. And you had
- 16:19- shared, that's kind of as as we were
- 16:21- chatting about having you come on the show,
- 16:22- a little bit of research that Nasuni has
- 16:24- done. Like, there there was a stat in
- 16:25- there that really stood out to me that
- 16:27- 90% of the orgs are piloting AI agents,
- 16:30- but only 43 call those projects successful.
- 16:33- So that leaves this massive gap around some
- 16:35- of these areas that you talked about. It
- 16:37- could be a data problem, data quality, garbage
- 16:39- in, garbage out. Could be a governance problem,
- 16:42- a security, like, versioning, how many versions of
- 16:44- this do I have and where it lives
- 16:46- and that whole sprawl. So so I'm curious,
- 16:48- like, kinda tying this back to Nasuni and
- 16:51- what you all do, how does something
- 16:53- like
- 16:54- AI activate come in and let an agent
- 16:57- do
- 16:58- more or
- 16:59- have kind of better governance, things like that,
- 17:02- than it would over just a traditional file
- 17:04- share, whether it was SMB, Nasuni managed, and
- 17:06- anything like that? So first and foremost,
- 17:08- when we talk when I talk with customers,
- 17:11- part of that
- 17:12- disparity between the 90% that are experimenting versus
- 17:15- those that are actually seeing success, some of
- 17:17- it is the confidence to deploy it in
- 17:19- a meaningful enough
- 17:21- scenario
- 17:22- because they don't necessarily feel that they have
- 17:23- the governance frameworks in place. So, ultimately, those
- 17:27- might not be as business impactful
- 17:29- in terms of those agentic deployments, and so
- 17:31- they kind of die on the vine as
- 17:33- you said. Back to your question specifically around
- 17:35- AI activate, so a couple key things. Number
- 17:38- one are the the first kind of parts
- 17:40- of our psychic framework, proprietary obviously, it's all
- 17:43- it's your data. But number two is the
- 17:45- secure side. And so knowing that
- 17:48- security is enforced
- 17:50- pre what we call pre query, right? So
- 17:52- if I sit down and I wanna use
- 17:55- Copilot to interrogate a set of projects that
- 17:57- I've been working on and I wanna interact
- 17:59- with it over what could be a massive
- 18:02- volume that has two petabytes in it. We
- 18:04- have customers that have, you know, multi petabyte
- 18:06- volumes. First and foremost,
- 18:08- we know or the customer who's using AIA
- 18:10- Activate understands that I will never be able
- 18:12- to
- 18:13- search over, interact with, interrogate data that I
- 18:17- don't explicitly have permissions to touch. And that's
- 18:19- done before we ever
- 18:22- go search across an index of the data
- 18:24- because sometimes they'll see what people do, they'll
- 18:26- go do a big search, they'll get the
- 18:28- results and then they'll filter afterwards for the
- 18:30- permissions that is not a secure model. So
- 18:32- first and foremost is security.
- 18:34- Second is around the index piece. So I
- 18:36- already mentioned it, but AyaActivate is a true
- 18:39- SaaS service, it's all our infrastructure
- 18:42- and we're able to ensure a customer that
- 18:44- all they have to do is scope whatever
- 18:46- they're looking at. So we'll go back to
- 18:47- my example of the legal department. They're now
- 18:50- able to go pick the volume that they
- 18:52- wanna work on and then generate. So it
- 18:55- is back. You mentioned MCP. So it is
- 18:57- basically exposed through an MCP. So literally anything
- 19:00- that can talk to an MCP
- 19:02- can now talk to AI activate and now
- 19:04- leverage this data knowing that the permissions are
- 19:07- enforced against it, whether that's an individual or
- 19:09- an agent operating on behalf of an individual.
- 19:11- So
- 19:12- right out of the gate, you've got the
- 19:13- security in place, you've got the permissions enforcement
- 19:16- in place, and you've got the benefit of
- 19:18- the index, which again means you're gonna get
- 19:20- much, much better results than if you were
- 19:22- just throwing an LLM at a massive volume
- 19:25- of data. And the MCP part is really
- 19:26- important. Not only is it standards based, like,
- 19:28- we're a platform. Right? So I don't wanna
- 19:31- be in the game of trying to go
- 19:32- build hundreds of
- 19:34- vertical specific agents.
- 19:36- The world changes too fast. That's not our
- 19:38- core competency.
- 19:39- We want our customers to be able to
- 19:41- use best of breed AI services to plug
- 19:43- into our platform. And so that MCP
- 19:46- piece is really important. And then we built
- 19:48- some intelligence into there in terms of how
- 19:51- to take the prompts that come through
- 19:53- and optimize that query against that back end
- 19:55- index to get the best results, but also
- 19:58- token utilization and token efficiency is a big
- 20:00- thing. So making sure that it's gonna be
- 20:02- as performance so the customer again is spending
- 20:05- more time reasoning over the data versus just
- 20:07- trying to find the data. Now the last
- 20:08- bit that I'll say and I mentioned the
- 20:09- acquisition we made was a company called DriveIQ
- 20:12- and
- 20:13- this is the next bit for us which
- 20:15- is layering on that additional layer of governance
- 20:18- on top of it. Because permission enforcement is
- 20:20- great, But now we wanna go that next
- 20:22- layer, right? And so now a customer can
- 20:24- work with us and say, all those legal
- 20:26- documents, I'm gonna go in, I'm gonna discover,
- 20:28- classify that, tag it. I can even redact
- 20:31- individual files. So there might be some particular
- 20:35- field or a customer name or something that
- 20:37- you don't want that's in a legal contract
- 20:38- to ever be exposed to AI. Great. You
- 20:41- can redact the file and still leverage that
- 20:43- file on all the context it has with
- 20:45- AI, but in a governed fashion. So that's
- 20:47- the big piece. So AI activate is all
- 20:49- about that secure
- 20:51- index data, standards based connectivity for AI solutions
- 20:55- like Copilot. It plugs right into Foundry.
- 20:57- We have one of our customers is building
- 20:59- a multi step agentic workflow. They're in civil
- 21:03- engineering firm, multi step agentic workflow
- 21:06- where they're combining structured data with another agent
- 21:09- in the flow with the unstructured data retrieved
- 21:11- through AI activated. So we've got all that,
- 21:14- and then now we're gonna be layering in
- 21:15- the DriveIQ piece for that additional governance layer
- 21:18- for our protection and oversight.
- 21:22- If you are managing file storage today, you've
- 21:24- probably felt the pressure. Refresh cycles are harder
- 21:27- to plan, costs and complexity are increasing, especially
- 21:31- as file workloads scale across regions.
- 21:33- That's why Microsoft customers trust Nasuni. As an
- 21:37- Azure certified software,
- 21:38- Nasuni provides a global file system designed for
- 21:41- enterprise scale environments,
- 21:43- keeping file data secure, governed, and accessible across
- 21:47- Azure environments without added infrastructure
- 21:49- complexity.
- 21:50- It's the foundation
- 21:51- to keep your unstructured data stable even as
- 21:54- your infrastructure
- 21:56- evolves. To see how much you could save
- 21:58- while keeping your initiatives on track, visit
- 22:01- nasuni.com/tco.
- 22:04- That's nasuni.com/tco.
- 22:10- When you start thinking about, like, that index
- 22:13- and the DriveIQ stuff and sensitivity,
- 22:16- obviously, you have, like, Purview and you have
- 22:18- DLP over there. Is there,
- 22:20- like, an integration
- 22:22- there where it's respecting some of the same
- 22:25- labels or syncing up with Purview so you
- 22:28- can have the labels? Like, how does your
- 22:30- solution with DriveIQ,
- 22:32- AI activate, and some of that kinda work,
- 22:35- it sounds like, in partnership
- 22:37- with
- 22:38- some of the controls Microsoft already has there?
- 22:40- There's very much a better together story here.
- 22:42- A lot of the customers that DriveIQ worked
- 22:44- with prior to the acquisition
- 22:46- leverage Purview as well. Right? Because it's top
- 22:48- notch in terms of the m 30 three
- 22:50- sixty five estate.
- 22:52- And but you wanna have something where you're
- 22:53- able to leverage that work. And so, yeah,
- 22:55- it can talk and it actually implements the
- 22:57- same sensitivity
- 22:58- labeling scheme. And so there's compatibility across there,
- 23:02- but it extends obviously beyond the Microsoft estate
- 23:05- where Purview is best to this whole other
- 23:07- world of all the other types of unstructured
- 23:10- data that typically don't live outside of SharePoint.
- 23:12- And so that's the better together piece here.
- 23:15- And we have, again, we have customers that
- 23:17- have petabytes in SharePoint today, but then tens
- 23:20- and twenties of petabytes of other data that
- 23:23- doesn't live in SharePoint. And so DriveIQ can
- 23:25- really help to bridge that. And there's also
- 23:27- great integration. DriveIQ can integrate with cloud SaaS
- 23:31- services like SharePoint, like Google Drive, etcetera. And
- 23:34- so it can help. Sometimes you get customers
- 23:36- that run into challenges with overages because of
- 23:38- all the data they have put in SharePoint.
- 23:40- And so it can help migrate data onto
- 23:42- Nasuni to help archive that, for example. So
- 23:45- you have a gold master copy, but it
- 23:47- leaves behind what are essentially stubs. So So
- 23:49- that if for some reason someone did need
- 23:51- to go react access that file through SharePoint,
- 23:53- great. You can just rehydrate it right back
- 23:54- into SharePoint. So both from a kind of
- 23:57- data integration and migration perspective, but then also
- 24:00- in terms of consistency of leveraging that labeling
- 24:03- work that Purview is doing, there's a really
- 24:05- nice synergy there between the two. Okay. And
- 24:07- then when you're trying to, like, filter out
- 24:08- that content, are you then going and creating
- 24:11- rules and, like, DriveIQ
- 24:13- that says just don't ever surface
- 24:15- this customer name or Social Security numbers or
- 24:18- credit card numbers
- 24:20- through that MCP server? Today, so that we
- 24:23- literally just closed the acquisition.
- 24:25- So one of the first pieces that we're
- 24:27- gonna be working on from an integration perspective
- 24:29- is exactly that. So that act AI activate
- 24:31- can
- 24:32- query in real time and understand here are
- 24:34- the policies that have been set up, and
- 24:36- then there are what are called action sets.
- 24:38- And so they can be so it's not
- 24:40- just a passive scanning solution. I already mentioned
- 24:42- that it can redact, but it can also
- 24:44- take action. But also in this scenario, yeah,
- 24:47- you might have a policy that says, no
- 24:49- files with this tag can ever be exposed
- 24:51- to Copilot. Then great. We can enforce that
- 24:54- at query time as part of AI activate
- 24:56- as well. That's one of the first scenarios
- 24:58- that we're going after. I got one of
- 24:59- the really powerful things. Right? You bring I
- 25:01- I mean, MCP in this case, like, helps
- 25:03- you to be kind of that proxy layer
- 25:05- with a common set of tools.
- 25:07- The nice thing about agents,
- 25:10- for the most part is they're just text
- 25:11- files. Right? There are definitions that are out
- 25:13- there seeing carry them, evaluate them, and test
- 25:15- them against various models with kinda that consistent
- 25:18- interface and layer that sits out there. So
- 25:21- it's been interesting to see that ecosystem take
- 25:23- off like a hockey stick. I think Anthropic
- 25:26- did a pretty good job, like, spinning that
- 25:27- up and saying, hey. Here's the thing for
- 25:29- everybody. But it's super powerful. Right? Because somebody
- 25:31- can onboard to the platform,
- 25:33- and the tools that your MCP
- 25:36- server or servers expose, they just become consistent
- 25:39- tool sets, not only for agents, but also
- 25:42- humans who are out there using a harness,
- 25:44- if it's Claude or ChatGPT
- 25:47- or they're in a CLI and any anything
- 25:49- like that. Exactly right. And talking about agents,
- 25:51- one of the other really important things too
- 25:53- is from our perspective,
- 25:55- agents are just a more powerful
- 25:58- version of people that don't need to sleep.
- 26:00- Right? So they execute file system operations against
- 26:03- the file system. That could be creates, deletes,
- 26:05- modifications, etcetera. So the really nice thing is
- 26:08- all of that activity that an agent might
- 26:10- be driving against it gets audited just like
- 26:12- every other user client application that's touching the
- 26:15- file system. And so having that visibility, which
- 26:18- we have within the platform
- 26:20- today with something we call FileIQ,
- 26:22- we now have that audit trail.
- 26:24- So if an agent happened to go do
- 26:26- something crazy like delete 10 different directories
- 26:30- accidentally,
- 26:31- we have that full audit trail. We know
- 26:32- exactly what files were impacted, and then we
- 26:34- can restore them very fast by virtue of
- 26:37- the way that our product is marketed. Right?
- 26:38- Because we're never copying data back from backup
- 26:41- somewhere. We're just putting a shifter a pointer
- 26:43- on the file system, pointing to the previous
- 26:45- last known good version of that data, and
- 26:47- then we can restore it right away. And
- 26:49- so that's where, in a lot of ways,
- 26:51- our CEO, Sam King, has a nice phrase,
- 26:53- which is we feel in a lot of
- 26:54- ways we're made for this moment.
- 26:56- A lot of the work that's gone into
- 26:58- thinking just about
- 27:00- modernization
- 27:01- of file data infrastructure
- 27:03- and the data protection and recoverability that we've
- 27:06- built into the platform is really nicely matched
- 27:09- with this agentic world and some of the
- 27:11- fears that people have of what happens if
- 27:13- this stuff gets out of control. I'm a
- 27:15- little curious. So in this world where you
- 27:17- plug in
- 27:19- more and more on the top, I think
- 27:20- one of the things that I see folks
- 27:22- struggle with is just getting their environment aligned
- 27:25- in the first place. Like, you talk about
- 27:26- Copilot, not just letting up licenses, but it's
- 27:29- back to the the old SharePoint problem of,
- 27:32- like, what were the permissions on my SharePoint
- 27:34- sites and and what got overexposed and and
- 27:36- what's out there. So if
- 27:38- there's a customer, they're in the Microsoft ecosystem,
- 27:40- they're out there, they're doing SharePoint, they're doing
- 27:43- Copilot, they have file shares,
- 27:45- kind of these things. If they come and
- 27:46- they plug in
- 27:48- AI activate drive IQ, to to the degree
- 27:50- you can say because you you just got
- 27:52- it. You know, what kinda just works for
- 27:54- them there based on what they've already done?
- 27:56- Or is there kinda more of a lift
- 27:58- to come in and say, hey. There's more
- 28:00- prereqs here to really be able to take
- 28:02- advantage of the data that's out there. Let's
- 28:04- start with a customer that's already a Nasuni
- 28:06- customer and sort of how do they plug
- 28:08- into this this world of AI and different
- 28:11- solutions.
- 28:12- So, again, the the beauty of it is
- 28:14- the starting point for Nasuni, one of the
- 28:16- biggest value propositions is data unification. Right? So
- 28:19- that we have what we call a global
- 28:20- name space over all of the data. And
- 28:22- so by virtue of having that,
- 28:24- now being able to just literally plug AI
- 28:27- activate into that and be able to build
- 28:29- these indexes and the fact that they don't
- 28:30- have to deploy infrastructure to do it, that
- 28:32- makes it real easy. The harder scenario is
- 28:34- the new customer, right? Who's coming and talking
- 28:37- to us and maybe they have a bunch
- 28:38- of SharePoint today, they've got a bunch of
- 28:40- Windows file servers and their remote offices, they
- 28:43- might have some NetApp and some Isilon. One
- 28:45- of the really nice things about DriveIQ and
- 28:48- one of the very important things for me
- 28:49- in particular is we were evaluating partners and
- 28:51- then eventually targets for acquisition
- 28:54- is it's really important that solution can also
- 28:57- provide huge value to data that isn't yet
- 28:59- on the Citi platform. So that same discovery
- 29:02- that you wanna do upfront to sort of
- 29:04- get your hands around what data can be
- 29:06- optimized for AI, well, that's the same thing
- 29:08- we can now go do with customers when
- 29:09- they're thinking about choosing a platform like the
- 29:11- Citi. So we can go do that discovery
- 29:13- upfront. We can help them
- 29:15- identify not just where do they have unstructured
- 29:18- data, but what's in that? What data has
- 29:20- been accessed recently? What of that is wrought,
- 29:22- right? So redundant, obsolete, trivial data. And so
- 29:26- if they have 20 petabytes of unstructured data
- 29:29- across these different places, alright, maybe 10 of
- 29:32- those are really critical operational data and we
- 29:35- can help them get their set, their arms
- 29:36- around it. And oh, by the way, DriveIQ
- 29:38- also migrates data. In fact, it migrates a
- 29:40- lot of data into SharePoint and into the
- 29:42- Microsoft ecosystem
- 29:43- today. So we have that single solution that
- 29:46- can give them visibility
- 29:47- to classify, tag, get their arms around it,
- 29:50- move the data to the Nasuni platform.
- 29:52- And then now, once we have it on
- 29:54- the platform, the the same policies that they
- 29:56- wanna define in that initial phase can all
- 29:59- be enforced once the data sits inside of
- 30:00- the Nasuni platform as well. It sounds like
- 30:03- from even where Nasuni started to what you're
- 30:05- doing today, that's one of the big advantages
- 30:08- of Nasuni is that you kinda got your
- 30:10- start in a data platform. So you have
- 30:13- a really good grasp on data storage, on
- 30:16- all things related to data. And now with
- 30:19- AI, you're just able to really build these
- 30:23- products,
- 30:24- integrate these products, the DriveIQ,
- 30:27- into
- 30:28- a really solid
- 30:29- data foundation
- 30:31- that where you got your start? Yeah. Exactly.
- 30:33- That's a great point. So, really, the start
- 30:35- of Nasuni way before I joined the company
- 30:37- was all around
- 30:39- modernization of traditional network attached storage, right? And
- 30:42- that's right in the name. Nasuni is all
- 30:44- about unified NAS, right? And it was all
- 30:47- around to our founders credit recognizing there was
- 30:50- this new thing called object storage
- 30:53- that had unbelievable characteristics from a, obviously, cost
- 30:56- of ownership perspective, from a scalability perspective, and
- 30:59- from a durability perspective.
- 31:01- But there's this thing called the file system.
- 31:03- And what if we could combine them both?
- 31:05- Right? And so our
- 31:07- original architecture
- 31:09- where the entire file system lives inside of
- 31:11- the object store, all the metadata, all the
- 31:13- data was one of the number one innovations
- 31:15- that is to this day are probably our
- 31:17- biggest enduring differentiation
- 31:19- because it allows us to now
- 31:22- take advantage of again economies of scale,
- 31:24- scalability
- 31:25- of object storage, cost of ownership of it,
- 31:28- but then also to eliminate some traditional things
- 31:31- that you need along the way like backup
- 31:33- and disaster recovery built into the platform. And
- 31:36- so when we recognize, well, we're a platform
- 31:39- now, companies are entrusting petabytes of their data
- 31:41- to us, we've earned the right, and by
- 31:43- the way, they're also demanding that we provide
- 31:45- more value for the data that lives on
- 31:47- that platform. That started out with providing more
- 31:50- advanced cybersecurity
- 31:51- type capabilities like our ransomware protection service
- 31:55- that then went into
- 31:57- thinking about things like, oh, wow, AI activate.
- 32:00- I wanna be able to tap into that.
- 32:01- And then DriveIQ is the latest in that.
- 32:03- And so, yeah, we've laid a foundation and
- 32:06- the company laid a foundation in the file
- 32:07- data platform. And now we're later layering these
- 32:10- these data services, if you will, on top.
- 32:12- And then earlier in the year, we acquired
- 32:14- a company called Resilio,
- 32:15- which is some amazing technology in terms of
- 32:18- the ability to
- 32:19- move data at incredibly high speeds. And so
- 32:22- you talk about today, one of the big
- 32:24- challenges that we hear is around, again, matching
- 32:27- the data that I have to the compute
- 32:29- that I wanna run against it from an
- 32:30- from an AI perspective. Typically, we are in
- 32:32- the inference world of things, not in the
- 32:34- model training side. But again, that compute might
- 32:37- live out on the edge somewhere, and I
- 32:39- don't wanna have to create a copy of
- 32:40- it. And so the the architectural
- 32:43- fact foundation that we laid with with Nasuni
- 32:46- and UniFS, etcetera, that's why we say we
- 32:48- feel like we're made for this moment because
- 32:50- that's exactly what we're seeing our customers need
- 32:52- as they're embracing AI and they're they're trying
- 32:55- to grapple with some of these challenges. We
- 32:57- say, bring AI to your data, not your
- 32:59- data to your AI, right? And so being
- 33:01- able to present that data anywhere in the
- 33:03- world so that customers get the best of
- 33:06- both worlds. They get all of the benefits
- 33:08- of unification consolidation of their data under one
- 33:10- platform,
- 33:11- but they also get the benefit of this
- 33:13- hybrid world where we can make that data
- 33:15- available to GPUs that are running in a
- 33:17- colo in Jacksonville
- 33:19- or maybe their GPUs, but they're in another
- 33:21- Azure region. So I wanna be able to
- 33:23- take advantage of it. You don't know how
- 33:24- powerful a global namespace is until you get
- 33:27- into one and and start acting with it.
- 33:30- Yes. There's a couple folks out there who
- 33:32- do them, and, out of those, there's a
- 33:34- couple who do them really well. So I'm
- 33:35- curious if
- 33:37- based on what we've talked about today, so
- 33:39- if somebody's out there, they're listening to this,
- 33:41- they're thinking about
- 33:43- evaluating for their org, where do you see
- 33:46- kind of value in
- 33:48- proof of concept? Like, if I was gonna
- 33:49- say, like, hey. What are, like, one, two,
- 33:51- three? Because there's probably not one. Everybody's a
- 33:53- little bit different. But what's a good proof
- 33:55- of concept that you go and actually recommend?
- 33:58- Like, something that a customer could get started
- 34:00- with today, And it could be bigger, it
- 34:03- could be smaller, but something that kinda shows
- 34:05- them some value upfront. Right? You talked about
- 34:07- tokens costing money, and if they do, right,
- 34:10- the idle GPUs, they're costing you a bunch
- 34:12- of money. So if you are in the
- 34:13- inference flows, rack, things like that, these become
- 34:16- very important. Yeah. Like anything. Obviously, start with
- 34:19- the use case. Pick a business impactful use
- 34:21- case and focus on that. I don't try
- 34:23- and boil the ocean, but if you get
- 34:26- an internal champion, whether it's my legal department
- 34:28- example or whether it's a set of engineers
- 34:31- working on geospatial data. Pick a targeted use
- 34:34- case and then go through that kind of
- 34:36- checklist that I talked about. Like, think about
- 34:39- a framework of, do we understand the data
- 34:41- that we need to be able to leverage
- 34:43- to make this use case successful? Do we
- 34:45- have security and government's controls around that? Do
- 34:49- we understand what kinds of AI tooling we're
- 34:51- gonna want, whether we're gonna be building an
- 34:53- agentic workflow or whether it's just individuals leveraging
- 34:56- Copilot against that data? And then be able
- 34:58- to
- 34:59- go set up a set of metrics or
- 35:01- KPIs that represent success. Like, one of the
- 35:03- hardest conversations right now is proving ROI for
- 35:07- AI. Right? And so if you don't get
- 35:09- the other stakeholders
- 35:10- bought into,
- 35:11- hey, if we deliver on x, y, and
- 35:13- z, that's real value. And that could be
- 35:15- productivity measures, that could be comparing actually one
- 35:19- nice scenario that we can do with customers
- 35:21- is you can go say, hey, do you
- 35:22- run a comparison where you just throw Copilot
- 35:25- directly at this data store and then compare
- 35:27- that versus going in through using something like
- 35:30- AI activating our other solution. And so make
- 35:33- sure you got the metrics that define success,
- 35:35- have a real use case, and then have
- 35:37- that checklist so that you can prove to
- 35:38- the rest of the business. This is a
- 35:40- framework that we can follow
- 35:42- so that if we're able to prove that
- 35:44- out for this use case, now we can
- 35:45- wash, rinse, repeat with other use cases in
- 35:47- the organization through a repeatable pattern. I think
- 35:50- where we see things sometimes go off the
- 35:52- rails is just thinking AI is gonna just
- 35:55- do it for you. Like, hey, if I
- 35:56- make data available to AI, it's gonna be
- 35:58- great. Right? That's one. Trying to pick up
- 36:00- too big a scope is a is another.
- 36:03- And not having very clear
- 36:05- defined measures of success before you begin is
- 36:09- probably the other one that we see with.
- 36:10- Alright. And if people wanna get started with
- 36:13- you, Nick, if they wanna go reach out
- 36:15- to Nasuni,
- 36:17- work with you, learn more about your current
- 36:20- products, or it sounds like maybe you have
- 36:22- some announcements here coming up shortly. Absolutely. How
- 36:26- are ways people can
- 36:28- reach out or best way for people to
- 36:30- reach out to you, get a feel for
- 36:31- how you can help them or
- 36:34- future upcoming products or integrations with some of
- 36:36- these acquisitions? Yeah. We do actually have a
- 36:38- big external announce coming up on October 6.
- 36:41- So stay tuned around that. There'll be a
- 36:43- lot of content coming out. I think we
- 36:45- can provide probably some links to blog posts
- 36:48- as well as webinars that we might be
- 36:50- doing and really start to think about
- 36:53- what are some of these use cases,
- 36:55- reach out through those those links. There's always
- 36:58- on the Nasuni website, of course. You can
- 37:00- actually go see an AI version of me
- 37:02- on the website, which, you know, can be
- 37:04- a little bit interesting. But But it's a
- 37:05- great way to start to be able to
- 37:07- interact and ask some questions and then, obviously,
- 37:09- follow-up, reach out, provide contact info, but there'll
- 37:12- gonna be a whole bunch of information that
- 37:14- go into quite a bit more depth on
- 37:15- things that I've touched on today coming up
- 37:17- as part of that October 6 events. Perfect.
- 37:19- Yeah. We'll get a bunch of links, put
- 37:21- those in the show notes for anybody that
- 37:22- wants to go sign up for webinars, learn
- 37:24- about those.
- 37:26- I know we both have meetings or you
- 37:28- and Scott at least have meetings coming up.
- 37:30- My Friday is almost done. Yep. My Friday
- 37:32- is not done. Else, yeah, anything else you
- 37:34- wanna share, Nick? Anything else around
- 37:37- Nasuni? Anything else you want people to know
- 37:39- about before we wrap up and head off
- 37:42- to our next meeting? No. I think the
- 37:43- last thing I would say is it's one
- 37:45- thing to hear me talk about it. The
- 37:47- best thing that we always find is talk
- 37:49- talk to one of our customers. So if
- 37:50- you are curious
- 37:51- and you wanna hear it from people that
- 37:53- have actually been living through what I've been
- 37:55- talking about, that's one of the best things
- 37:56- we can do. We do this all the
- 37:57- time where we have mini round tables by
- 38:00- vertical, by industry, where we can bring an
- 38:02- Asuni customer to talk to a group of
- 38:04- people that are just kind of interested. So
- 38:05- you have to hear it from the mouth
- 38:07- of the people that are actually out there
- 38:08- living and breathing this. So highly encouraged if
- 38:11- anything that we've talked about today is interesting
- 38:13- and you get more information coming out of
- 38:15- that October event. We're also going to be
- 38:17- doing an event in Chicago in November, which
- 38:20- we call in the SUNY verse. So there
- 38:22- could be another opportunity to come meet with
- 38:24- a bunch of our existing customers. Think of
- 38:26- it like we're a mobile user group event.
- 38:28- And then also Microsoft Ignite, we will be
- 38:31- there. I will be there. So would love
- 38:33- to meet people if they wanna come by
- 38:34- and find us. But, yeah, I'd say the
- 38:36- best thing you can do is actually learn
- 38:38- from the people in your industry that are
- 38:40- living with the same challenges you have, who've
- 38:42- had experience with some of what I've been
- 38:44- talking about because they'll give you the honest
- 38:46- opinion. Awesome. Well, thanks, Nick. Thanks for coming
- 38:49- on and talking to us today, sharing all
- 38:51- this information with us, and thanks as well
- 38:54- for partnering with us, for sponsoring the podcast
- 38:57- and partnering with us in that way. So
- 38:58- we really appreciate that as well and hope
- 39:01- you go and are able to have a
- 39:03- nice weekend. Relax a little bit. Appreciate it.
- 39:05- Really good. Guitar. Sorry? Yeah. I've got a
- 39:08- I've got a song to record this weekend.
- 39:10- So there you go. Alright. Awesome. Well, thanks
- 39:12- again. Thanks again, Scott, for as always, and
- 39:16- we'll talk to you guys later. Alright. Yep.
- 39:17- Have a good one. Thanks, Aaron.
- 39:20- If you enjoyed the podcast, go leave us
- 39:23- a five star rating in iTunes. It helps
- 39:25- to get the word out so more IT
- 39:27- pros can learn about Office three sixty five
- 39:29- and Azure.
- 39:30- If you have any questions you want us
- 39:32- to address on the show, or feedback about
- 39:34- the show, feel free to reach out via
- 39:36- our website, Twitter, or Facebook.
- 39:39- Thanks again for listening, and have a great
- 39:41- day.