S4 | E24 | AI Sprawl, Token Costs & Enterprise AI with Nikhil @ Cribl

Alex Hutchings: 00:00:03

Welcome to the Think Data podcast, brought to you in partnership with Dataworks. If you want to stay up to date with the latest breakthroughs and trends in the world of data and artificial intelligence, and if you're curious about some of the strategies that companies and founders use to launch data and AI products, then you're in the right place. Our aim is to bring together a diverse lineup of fantastic guests from the founders through to accomplished leaders and product owners at some of the most fascinating data and AI companies worldwide. They will each offer you their own unique insight into what it takes to launch and scale a great data business. Thanks for tuning in, and I hope you enjoy the episode. AI is supposed to make businesses more productive, but what happens when it also starts to become one of the biggest expenses? With seventy-seven percent of workers now using multiple AI tools every week, companies are facing a real dilemma, and it's called AI sprawl. There's multiple tools. They're all doing the same job. They've got duplicated prompts, and costs really do spiral. So the question everyone has asked is: how do businesses get the benefits of AI without losing control of the bill? Today, I'm really excited to be joined by Nikhil Mungil. He is the head of AI, R&D at Cribl, and he is here to talk about the true cost of AI, why companies start to give employee token budgets, whether workloads will move away from the cloud, and why AI spend could be managed in much the same way as payroll. Nikhil, really, really good to have you on the Think Data podcast today. You've got a really interesting background. You've actually got kind of one of those, uh, fantastic kind of pedigree and company names that people who will be watching and seeing your profile on LinkedIn after this will, will certainly relate and, and kind of recognize who they are. But you've worked for the likes of Substack, Splunk, ThoughtWorks, and obviously from technical engineering capacities. And then you joined Cribl about three and a half, four years back. I was, I was really interested to kinda touch on what led you to join Cribl, because obviously the domain that you're sitting in, and obviously ultimately your role within that kind of AI R&D piece in this space is super hot. But obviously three and a half years ago, um, well, it was still hot, but it's probably been become even hotter now. So yeah, what, what, what made you join Cribl? And, and if you look at your background, kind of what, uh, what put you in the best position to land this gig?

Nikhil Mungil: 00:02:42

Hundred percent, Alex. So thanks for, thanks for having me on. Yeah, I've been fortunate, you know, fortunate to have worked across a pretty good set of, set of companies so far. And one of the things I've really focused on in the last ten to fifteen years of my career has just been in the observability and security industry, right, where we deal with a large amount of data, machine data, logs, telemetry, you know, every- everything else that is generated by computers and by networking equipment. And when I joined Cribl three and a half years, four years ago, the opportunity was just so huge when it came to applying advances in AI into, to machine data, right? Because we are, we are in, in the hot path where we get to work with so much data, and ultimately, when we, when we think about a software like Cribl or any, any software that c- is capable of, you know, working across large volumes of data, it really becomes an information compression problem, right? You have a lot of signal that is being generated by computers, by networking equipment, and by everything else, and AI is just such a perfect technology. You know, both traditional ML models as well as, you know, modern LLMs in order to really just compress that petabytes upon petabytes of, of, of noise and signal all jumbled together into a single insight. Like, "Hey, your... This node is under attack," or, "This firewall is about to hit its end of life," or whatever.

Alex Hutchings: 00:04:08

Mm-hmm.

Nikhil Mungil: 00:04:08

You know, whatever really crisp insight that, you know, we as human beings are looking for is, uh, something that AI helps us compress this, you know, vast amount of information into and just pull that insight out. And that was, those were like the original pieces if I had to, you know, put it in a single sentence that really drew me to this opportunity. And, and at Cribl, I've been fortunate to have established our AI research and development organization, both from an engineering as well as product engineering standpoint, like product management as well as engineering management. And I, I have a, an incredible team that's capable of executing really fast. We train our own models that can work on telemetry data, and then we also build agentic products that enable our users to take advantage of those models and then also, you know, get, get their work done.

Alex Hutchings: 00:04:53

Yeah, interesting. And for the more non-technical listeners of this episode, how would you describe Cribl and the problem that it's solving?

Nikhil Mungil: 00:05:03

Yeah. So when you think about large corporations, so, you know, think about Fortune 100s or large banks, government agencies, contractors, you know, what have you, they have large systems. They have data centers upon data centers. They have racks. They have servers. They have endpoints like computers and phones and everything else. And all of this infrastructure is producing telemetry data. So whenever a packet enters a network or when a networking packet or an information packet leaves a network, a log line is produced. You know, when an application, a server-side application is running, it's constantly producing logs and telemetry data. And it's the job of IT practitioners and security practitioners and admins to collect all that data, to sanitize it, to keep it around for forensics, for incident response, for all kinds of use cases. And Cribl produces software that enables our users to wrangle this data. So it's, it's this engine for, you know, your IT data and your security data and allows you to turn that data into actionable insights.

Alex Hutchings: 00:06:10

Interesting. In terms of your, your role there over the last kind of three and a half years, what, what have you kind of seen as the biggest wholesale change in maybe how companies are approaching kind of their, their, their broader AI strategy, for example?

Nikhil Mungil: 00:06:24

Yeah. So I get to work a lot with, you know, CIOs and CSOs of large enterprises in just as a part of my job. And one of the things that we have seen is almost like a seesaw, you know, between extremely bullish expenditure when it comes to AI, and then taking a little bit more measured approach and, you know, trying to measure your return on investment and trying to, trying to measure, you know, where productivity gains are being focused. And then again, just overspending and just making space for innovations and new things to come out. Because like the way I think about it, AI is really accelerating two things when it comes to enterprises, right? One is concrete measurable productivity gains, right? It could be a customer support team being able to close a, you know, more number of tickets per employee, or it might enable an SRE engineering team to reduce the time between the meantime that it takes for them to resolve incidents and those sort of things. So that's, that's in your measurable set of things. And in the last three years, there's been a lot of focus on getting tight about how do you measure those things-

Alex Hutchings: 00:07:30

Mm

Nikhil Mungil: 00:07:30

... and how do you optimize those things. But then there is something else as well, which is not, which, which doesn't get discussed nearly as, as often as it should be discussed, which is AI also makes it possible for things that were previously not possible. And you're not gonna be able to, you know, lay down a framework to measure the ROI of that quite as easily as you would with the first thing that we just talked about.

Alex Hutchings: 00:07:54

Yeah.

Nikhil Mungil: 00:07:54

So just, just being-- just, you know, having this technology at your fingertips and the way, the way to think about it is, you know, having this extreme levels of intelligence on tap that you can just point to any problem that you want is gonna, you know, be magical in letting you uncover things that you could not in the past. You know, it's just y- so when a CEO is trying to make a large investment in their, for AI across their enterprise, they're also looking for a, a place where a thousand flowers can bloom and, you know, something novel can come out of it as well.

Alex Hutchings: 00:08:27

Mm.

Nikhil Mungil: 00:08:28

And that I feel like is just prone to overspending quite a bit, unlike the first case that we talked about.

Alex Hutchings: 00:08:34

Yeah, it's interesting, and we're gonna probably dig on that overspend piece sh- soon, but what is interesting, w- taking a look at your website, looking at some of the cases, I think it's on the first page, the way, uh, the, on the new app product that you've launched. I think when you-- a lot of companies' workflows have been built around these kind of SaaS solutions, but actually you're flipping that on its head and almost saying, "Well, actually, let, let's, uh, let, let's build, let, let's build around these workflows," as opposed to kind of the workflows having to adjust based on the apps. How are you approaching that? Because obviously a lot of companies who are going through AI strategy kind of transformation, and they're investing a huge amount of money. They, they are getting all the latest, shiniest tools, and they're kind of sticking it all together, and it's becoming a bit-- it's not becoming as productive. So, so if companies are approaching this from an AI strategy standpoint, what are they getting wrong from the outset, do you think?

Nikhil Mungil: 00:09:25

Yeah. I think, I think there's a, there's a balance between using SaaS vendorized products that, you know, help you accomplish a certain problem or, you know, solve a specific problem across your enterprise.

Alex Hutchings: 00:09:35

Mm.

Nikhil Mungil: 00:09:36

And then what, what AI makes it possible is it-- the software can change, and it can mold to your users' workflows as opposed to your users having to, you know, modify their workflows to fit with enterprise software.

Alex Hutchings: 00:09:48

Yeah.

Nikhil Mungil: 00:09:49

Right? And that's been the huge-- That's been a huge paradigm shift the way we have seen it play out and, you know, where we have seen the interest coming from in the market, which is sophisticated IT shops or sophisticated enterprises have 40, 50 enterprise IT vendors in their tool belt, right in their-

Alex Hutchings: 00:10:05

Mm

Nikhil Mungil: 00:10:06

... ecosystem. So they have everything under the hood, and a lot of times you're, you're, as a practitioner, you're stuck, you know, moving from screen to screen, copy pasting objects from one, one tool into a different tool, and that's what I like to call, you know, your, your, your human workforce is, has become inefficient because of the various edges and surfaces that are being exposed by enterprise IT applications.

Alex Hutchings: 00:10:31

Mm.

Nikhil Mungil: 00:10:31

Well, guess what, you know, or what has happened in the last 12 months is most of these, uh, sophisticated enterprises or IT teams at, you know, large enterprises have started to build their own software that can manage a bunch of different tools.

Alex Hutchings: 00:10:44

Yeah.

Nikhil Mungil: 00:10:45

And app platform, you can actually do that. You can model your existing workflows into a, an app or a surface or, you know, a set of dashboards, a set of scripts that need to be run, a set of actions that can, uh, you know, be automatically in- put into motion, and you can create that entire end-to-end workflow using an app.

Alex Hutchings: 00:11:04

Yeah.

Nikhil Mungil: 00:11:04

And that's the future, right? Because ev- there, there is so much difference and there is so much nuance in how individual teams go about accomplishing their work, and there is no reason why they should be using the same standardized tool.

Alex Hutchings: 00:11:17

No.

Nikhil Mungil: 00:11:17

There's a way to like make it easier and faster, so we're gonna see a lot more of that. We're gonna see a lot more instances where individuals and small groups of people, you know, four to five people usually will work on a single task in a certain way, and they will prefer to have their own software that enables them to accomplish their work faster.

Alex Hutchings: 00:11:37

Yeah. And the thing I also like as well on that is there is an al- almost like a vibe coding element to it, isn't it? So it's actually-- you've, you've opened up the accessibility for that function, so it's a non-technical function or it's a-- yeah, as you say, someone who's not a technically proficient coder, you actually, uh, open that up, which, uh, i- in your, in your honest opinion, do you think that's the way we're gonna be going with kind of enterprise-wide AI adoption, where it, it needs to be more accessible for people as opposed to just the highly technical folk?

Nikhil Mungil: 00:12:06

I think so, yeah. I, I, I think we're also moving away from, you know, vibe coding as a standard to more intention-based or like, you know, intent coded So when, uh, you know, a platform like Cribl allows you to build applications and, you know, there's also other platforms in the industry that allows you to-- that allow you to bring applications and build applications. It's-- you're really trying to encode the intent of your users and creating a, a, a workflow or an abstraction based on those intents, right?

Alex Hutchings: 00:12:35

Mm-hmm.

Nikhil Mungil: 00:12:36

Often get expressed in natural language or as a prompt or something else, maybe even as a screenshot or, you know, just some sort of a freeform description, which then gets turned into an application. And what's really cool about this is these applications can be as complex or as on rails as you want them to be. And when these applications are running on rails, the underlying platform really exposes meaningful building blocks that these applications come together and stitch together, right? So you may have a few dashboards, you may have an alerting engine, you may have a conversational interface. These are all building blocks that are available to this, uh, to this framework when it's creating these apps, and they don't have to be reinstantiated by a generative, you know, agentic coding harnesses. So if you, if you want a conversational element in your application, you're not gonna have a vastly different experience than somebody else who is looking for the same thing.

Alex Hutchings: 00:13:29

Yeah.

Nikhil Mungil: 00:13:29

Which will, you know, makes these applications quite robust and distinct from classically white-coded applications, you know, which, which may have like a lot of variance when it comes to quality and usability.

Alex Hutchings: 00:13:41

Yeah.

Nikhil Mungil: 00:13:42

You know, creating applications using a platform like Cribl's app platform or any of the other tools out there, you're creating something that's far more robust, that's far more composable, and that can work in a way you expect it to work repeatedly.

Alex Hutchings: 00:13:55

Yeah, it's a really interesting inflection point because I, I was reading one of the stats I think for which I s-sent before this was like seventy, seventy of workers are kinda using multiple AI tools every week, and I think there's a problem which has been defined as kind of AI sprawl. What do you put that down to? Do you, do you think as a-- naturally as a race, we're, we're kind of used to, you know, seeing a new shiny tool and we think, "Oh, that's, that's gonna really help us." And then we sign up, we use it, but then the next new thing comes along or the, the original vendor gives us some additional features and functionalities that actually makes that subsequent vendor kind of null and void. And do you feel that's kinda where we are now, and where do you see this going? Because I've got my own thoughts on it and kind of I think consolidation's gonna happen pretty soon in terms of, you know, the bigger guys or, or the very quick growing, super powerful tools are gonna start to eat up the smaller ones. But in your own opinion, what are you seeing and what do you think is gonna happen next?

Nikhil Mungil: 00:14:55

Yeah. I think, I think, I think the way you framed it is pretty, pretty on point where, you know, there is a, there is a, there's a lot of fragmentation in which, you know, how the industry and, you know, vendors and as well as enterprises are approaching this entire problem. And if you had to take a step back, the way I would best describe it is you have intelligence as your raw resource that's being-- that's, that's, that, that's now come into existence, you know, thanks to OpenAI, Anthropic, and a handful of other labs. They've created frontier intelligence, which is just a, like one way to think about it is it's a, it's a, it's a fountain of tokens. You know, like it's a spout of tokens that are-- that represent intelligence. And there is a separate problem, which is how do you harness that, right? Like, so you have this, this raw flow of intelligence that's coming in, but it's useless unless you have harnessed it in a meaningful way. And the way to harness that, you know, if it comes to a coding harness, Cursor has done a fantastic job of, you know, harnessing that intelligence and applying it to agentic coding and, you know-

Alex Hutchings: 00:15:56

Hence why they got acquired, I guess.

Nikhil Mungil: 00:15:58

Right. Yeah.

Alex Hutchings: 00:15:59

Yeah.

Nikhil Mungil: 00:15:59

And, you know, so has CloudCode and Codex-

Alex Hutchings: 00:16:01

Yeah

Nikhil Mungil: 00:16:02

... and several other tools out there have figured out a way to harness this intelligence and make it genuinely useful for, you know, developers and software engineers. And I feel like I don't think we have seen that sort of unification of interface when it comes to harnessing intelligence for other job functions. I don't think there's quite a winner for, I don't know, harnessing intelligence for writing marketing copy, for example. Or maybe, maybe it's just ChatGPT, right? Maybe that's-

Alex Hutchings: 00:16:27

Mm

Nikhil Mungil: 00:16:27

... that harness. But I feel like there's a lot of-- one of, one of the reasons why we have so much fragmentation is because every, every, every vendor, every, uh, new tool out there brings its own opinion on how do you harness intelligence into something useful. And what we're gonna see is that the most useful of these interfaces are gonna organically rise up to the top and be adopted by all vendors, right? So, like once you figure out a way to like harness intelligence, it's pretty repeatable and, you know, you'll see lots of standardization and consolidation, as you put it.

Alex Hutchings: 00:16:57

Yeah.

Nikhil Mungil: 00:16:57

When it comes to enterprises, I feel like one of the things that enterprises should be doing is to own and control as much of the harness as possible, right? Which is almost antithetical to vendors supplying you the harness. Because the harness codifies what you deem valuable. You know, like what-- when, when, when an enterprise is like, "Well, I have this one workflow where, I don't know, I'm like approving receipts or I'm selling new product or, you know, like taking charge of my inventory." That's a, that's a workflow that requires some cognition along the way, and ultimately, the enterprise is responsible and knows how to best gauge whether or not that workflow was successfully completed. And, you know, what are the various edge modes and edge cases and failure modes of that, of that, of that. And the best way to represent that is a harness that harnesses your intelligence and then evals that allows you to measure evals and benchmarks and stuff like that, that allows you to measure how intelligence is accelerating that workflow. And then once you have that, then you can swap out different intelligence providers. You can, you know, take, take the right call between, you know, putting tokens towards advancing a particular workflow, or you wanna still have human, uh, a human workforce advancing that workflow. You can, you can be much more measured when it comes to putting it into motion.

Alex Hutchings: 00:18:17

Yeah. It's such an interesting point. And also, uh, putting on your point about putting more tokens towards, it, it does beg the question of that kind of the spend that these enterprises are gonna incur, and equally where those budgets sit and, you know, whether it's, uh, obviously certain teams will be using more c- uh, tokens, will be spending more money. So what's your thoughts on that in terms of AI budgets and where they should ultimately sit? Because obviously y- you feel like it should be managed more like in a, you know, how they manage payroll, for example. Certain sectors and teams should be having their own spend. Is, is that something you, you truly believe?

Nikhil Mungil: 00:18:50

Yeah. I think, I t- I, I think I do believe in that pretty, pretty firmly because just because of the differences in different tasks that people do-

Alex Hutchings: 00:18:59

Mm-hmm

Nikhil Mungil: 00:18:59

... right? Across an enterprise, somebody who is in the finance FBNA team and what they do on a day-to-day basis is so different from what a customer support person might do. And it's really hard to consolidate all of this, you know, at the very top and say that every person is gonna get, you know, so many dollars in tokens, because some workloads are just so much more well-suited to be accelerated using AI, and other workloads not, you know? And if you're-- So there is a, there-- it's, it's somewhere in between, you know, the CEO deciding token spend for every employee and every employee getting to decide token spend for themselves, right? It's, it's probably somewhere in between that.

Alex Hutchings: 00:19:39

Yeah.

Nikhil Mungil: 00:19:40

Probably a team level or, you know, organization level, like a team of team, team of teams level. But I would, I would, I would expect like every 25 to 30 people it changes. You know, what are your use cases and, you know, how are you measuring success and what are you accomplishing? What are you trying to accomplish-

Alex Hutchings: 00:19:55

Mm

Nikhil Mungil: 00:19:55

... with tokens changes. And, you know, it's, it's, it's AI systems as well as human labor who are also both, both entities are capable of producing intelligence or producing tokens, if you will-

Alex Hutchings: 00:20:07

Mm

Nikhil Mungil: 00:20:08

... us people included. So just a question of, you know, using the right token generator for the right task, if you will.

Alex Hutchings: 00:20:16

Yeah. It's funny because we started to see things even creeping on to kind of, uh, um, from a hiring standpoint, candidate packages around their new candidates coming in and negotiating this from the outset. And says, "Well, how much, how much money am I gonna have to spend on, you know, tokens and kind of, you know, AI capability?" 'Cause they don't wanna join a company and then realize actually there's not actually the money there for them to spend, and they're just, they're being hamstrung on their development anyway. So we're certainly starting to see that shift on top candidates negotiating as part of their package.

Nikhil Mungil: 00:20:45

Yeah. Yeah. I, I think that's also like tied to a little bit of earning it as you go, right? Because you're, you know, if you're, if you're burning through a lot of tokens but you're generating a lot of business value, then-

Alex Hutchings: 00:20:58

Of course. Yeah

Nikhil Mungil: 00:20:59

... everyone's fine with it.

Alex Hutchings: 00:20:59

Who, who minds, yeah.

Nikhil Mungil: 00:21:01

Right. That's great. But then, you know, if you have Bob from some team trying to, you know, burn like a million dollars worth of tokens every month without producing anything, then, you know, there's gonna be questions. So again, I feel like it's just so, we are, we are still in the early days here of, you know, what, what, what, what solid policy looks like. And then once we have the policy or governing that policy becomes a much easier problem to solve with, you know, the endpoint solutions or MCP solutions that are token gateway solutions. So I feel like we have, as an industry, we have a lot of muscle around enforcing policy and, you know, governing policy. But I feel like the bigger question is what's that policy gonna be and, you know, how-- what makes for a fair policy that's concretely driving your business forward-

Alex Hutchings: 00:21:45

Mm

Nikhil Mungil: 00:21:45

... without, you know, undue taxes or undue costs. So keep, keep the bottom line low, keep the top line up, right? So it's, it's the same old thing really. It just needs to be codified in a, a concrete policy around AI tokens.

Alex Hutchings: 00:21:59

Yeah. I couldn't agree more. And I think at the moment it's the real challenge for... I'm fortunate enough to have some really good execs, startup founders onto this, uh, show, and the, the hardest question to answer is how they're measuring the ROI of their solution. You know, that's 'cause at the moment there's-- it's very hard. Is it time? Is it, on your point earlier, if you've got Bob in accounts spending a million dollars a month and go, "Actually, is that a good investment of our time?" What, what's your view on that when you're talking to enterprises about looking at the ROI piece? Is, is that really as hard as it sounds?

Nikhil Mungil: 00:22:31

I think so. It's, it's hard if you try to do it at a global scale across your enterprise. You know, it's just because again, your, the return that you're trying to make or the progress you're trying to make is just so different based on a department, you know, a department success goals or team success goals-

Alex Hutchings: 00:22:47

Mm

Nikhil Mungil: 00:22:47

... and so forth, right? Especially and if you look at, you know, I'm here in San Francisco, so I also speak a lot with tech companies.

Alex Hutchings: 00:22:55

Yeah.

Nikhil Mungil: 00:22:55

And tech companies are so talent dense and so small compared to the impact that they have. So you might have like 200 people employed at a company which is having a vast, huge cultural and economic impact. And in that case, every group of like four to five people in that population of 200 or so employees is gonna be doing something very different from the next set of four to five employees, right? You know?

Alex Hutchings: 00:23:19

Yeah.

Nikhil Mungil: 00:23:20

They might be running community, somebody might be running sales, somebody might be, you know, creating the perfect Facebook ad or, you know, something else. And there is just no way a single policy is gonna be able to tell you their ROI on token spend when it comes to that sort of stuff, right? So I feel like it's, it, it needs to be, it needs to be defined and applied at, you know, a certain, at, at certain task level. Also, you know, a lot of, a lot of jobs or a lot of job roles are a combination of different tasks, right? So if, for example, if you, if you take a healthcare enterprise or, you know, a, a doctor for example, right? I mean, a doctor doesn't have-- Well, you know, a doctor probably has a lot of tasks on their plate, right? I mean, they're reviewing, they're reviewing files. They're, you know, keeping up with the latest research and they're, you know, using, using whatever telemetry they got from a, from a patient to, you know, formulate a hypothesis, and then they're delivering the message to the patient, right? And I feel like not being a doctor, we are typically exposed to just the final part here, which is delivering, you know, the news to the patient. But there is like several other task families.

Alex Hutchings: 00:24:26

Mm.

Nikhil Mungil: 00:24:27

Each one of them have a different token ROI if you had to- Accelerate some of them with tokens.

Alex Hutchings: 00:24:33

Yeah.

Nikhil Mungil: 00:24:33

But you're probably not gonna be able to accelerate the final part with, like, an audio-video model trying to, you know, deliver a message to a patient, right? I mean, like, the ROI is just not gonna be there for that.

Alex Hutchings: 00:24:44

No.

Nikhil Mungil: 00:24:44

So you have a human being do that.

Alex Hutchings: 00:24:48

Yeah, no, it, it's super interesting. I think there's so-- And what we're seeing from a staffing standpoint is, uh, you talk about those teams of four or five, we've seen those teams are becoming far smaller nowadays, but they are supported by those AI agents and those co-pilots, and that's become-- driving efficiencies. We're talking about engineering orgs have wholeheartedly been changed from, say, four or five years ago, uh, and we think that will continue. So you just kind of people obviously with their co-pilots, with their AI assisted development tooling, supporting them to work on bigger and more interesting things. Whereas actually before you had a lot of people working on maybe one project, you've probably got far less people now.

Nikhil Mungil: 00:25:25

Exactly. Exactly. And when you factor in things like recursive self-improvement or, you know, any agentic system or any intelligence power system that's automatically capable of improving its own abilities, then you really start these-- y-you really start seeing these things compound and, you know, accelerated at a pace that we have never honestly seen in the past.

Alex Hutchings: 00:25:46

Yeah, definitely. One final question for you, which I'm, which I'm interested to touch on given your background and where you've come from. If you're sitting down with a CEO now who is actually looking at investing millions into kind of AI over this kind of next twelve to eighteen months period, we've seen some of the limitations with, with having so many vendors and so many different tools, and these companies becoming hamstrung by their investment because they've almost paid for it, now they've got to use it. What bit of advice would you give them if they're entertaining or looking at their kind of wider AI strategy?

Nikhil Mungil: 00:26:16

Yeah. My advice would be to own as much of the harness as possible, right? Have-- Intelligence is something CEOs are gonna have to procure from intelligence providers, you know, be that a hyperscaler or be that an AI lab. But the harness is something that, you know, is, is truly unique to their enterprise, right? It's, it, it, it, it encapsulates, you know, their customer relationships, how they conduct business, you know, how they, how they move their business forward, you know, what sort of products they build.

Alex Hutchings: 00:26:44

Mm-hmm.

Nikhil Mungil: 00:26:45

And all of that is in a, you know, all of that nebulous sys-like s-system of systems perhaps, is what I like to call a harness, right? I mean, that's how you harness employee intelligence. So like, you know, when a new employee joins, you onboard them into that system, into that harness, right? So companies should, CEOs should insist on owning the rights and owning, you know, operating as much of the harness as they, they, they can, which includes putting that intelligence to work and then benchmarking and evaluating how well that intelligence did the job, right? So it comes in two parts, harnessing it and evaluating it. And that's where your investments should be focused because token costs keep falling, right? Tokens are, you know, there, there are lots of really smart people in this world who are working on creating GPUs, building data centers, you know, training models, putting the models in those data centers so we can count on intelligence tokens continuing to become available, uh, you know, costs going down, capabilities improving, increasing until we, you know, reach AGI. But how do we harness that-

Alex Hutchings: 00:27:49

Mm

Nikhil Mungil: 00:27:49

... is very much the thing that's going to prevent you from also provider, intelligence providers changing their terms, you know, changing the capacity of their tokens, changing-

Alex Hutchings: 00:27:59

Yeah

Nikhil Mungil: 00:28:00

... the capability of their tokens. It's just gonna safeguard, right? Because you'll be instantly able to tell when something degrades or something gets better. Like your, your benchmarks start improving or, you know, they start worsening and then the, the harness allows you to really be-- just insert any intelligence into the mix.

Alex Hutchings: 00:28:20

It's a fascinating period right now because I do think obviously post-ChatGPT, and I know that's kind of not when AI was created obviously. But I think for, for a consumer, that's kind of when they think AI actually started, whereas actually AI's been around for twenty plus years, isn't it? We look at back in machine learning in those days. And I think what, what's really fascinating now is these predicaments organizations are having to kind of, they've almost created a rod for their own back where this innovation has led to so much accessibility, so much data, so many fantastic tools. But on your point, and I think for people listening to Nikhil and Cribl is check their website, check what they're doing because the, the questions you pose to these CEOs is ultimately what they should be thinking about, isn't it? It's harnessing it and how do we deploy it, and ultimately how do we adopt it in a, in an effective manner? Because token costs are coming down, but that doesn't mean you need to go and get more packages and more tools.

Nikhil Mungil: 00:29:07

Hundred percent. Yeah. It's a, it's a, it's a balance between investing in the future, but then also insisting on returns with, you know, what was invested last quarter.

Alex Hutchings: 00:29:19

Yeah. Awesome. Nikhil, it's been really good to have you on this morning. So, uh, really interesting points here, and I think, uh, for anyone looking at Cribl, they've got a really intuitive front page which really explains exactly the problem that you guys are solving. So yeah, thanks for coming on this morning, especially after your well-earned PTO last week. It's good to, uh, good to get you on.

Nikhil Mungil: 00:29:37

Awesome. Thank you so much for having me, Alex. Appreciate it.

Alex Hutchings: 00:29:39

Thanks, Nikhil. Appreciate it.