Building Reliable AI Agents With Knowledge Gardens and MongoDB

[00:00:00] Do you need AI agents that you can trust? Well, with an AI data layer providing real-time connection within your data platforms, you can trust your agents to provide accurate solutions. So, scale your business by trusting your agentic AI accurately getting the work done for you. Trust its capabilities with Denodo. And you can do that by simply visiting denodo.com to learn more.

[00:00:26] What if your biggest AI challenge has nothing to do with models, prompts or agents, and everything to do with how your organisation understands its own knowledge? Well, today you'll hear why the future of enterprise AI might depend less on the intelligence of the technology and more about the quality of information feeding it.

[00:00:54] And my guest today is my friend Boris from MongoDB, officially friend of the show now. He's going to be rejoining me to introduce a memorable concept that he calls the Knowledge Garden. And he'll explain why businesses drowning in data can still struggle to generate meaningful outcomes. And along the way, we'll separate data from information, information from knowledge, and uncover why so many AI initiatives stall before they even reach production.

[00:01:24] And I hope by the end of today's episode, you'll have a much clearer view of what it takes to move beyond AI experiments, and start building systems that people can trust. And as it's Boris, you can expect practical insights, relatable examples, and a fresh perspective on why the organisations that organise knowledge effectively, how they could be the ones that get the most value from AI in the years ahead.

[00:01:49] So if you've ever wondered whether your AI strategy is built on solid foundations or quicksand, this conversation might give you a very different way of looking at the challenge. But enough for me, let me reintroduce you to my friend Boris right now. So thank you for joining me on the podcast once again. Always a pleasure to speak with you. But for anyone that missed our last chat, which was nearly a year ago,

[00:02:17] could you remind them a little about who you are and what you do? Sure, Neil. This is Boris Belek. I'm the Vice President of Industries and Global Field City at MongoDB. Funny, long title. But busy. I'm the geek who built industry solutions with our clients from an industry use case perspective. And that's what I do by now for 40 years. No hairs left. Sorry for that one, guys. But that's me. Love it.

[00:02:42] And when we last spoke around 12 months ago, I think we discussed helping enterprises simplify AI development. And now, fast forward a year, many organizations are moving from AI pilots to agentic systems now. But I'm curious, what have you seen change most in the enterprise AI conversation over the last year? Because it can be difficult keeping up. It's moving that fast right now. This is a great question. And I think my version is it became more professional.

[00:03:10] So, when we looked a year ago, it was a lot of experimentation. People are so happy about their first chatbot, their first track solutions. And then people started the term agentic script in. And people tried to figure out what does this actually mean. By now, it's getting more clear. And overall, by now, people become more moving more to production, have the usual production questions. Oh, my God, what do I do now? I am liable for what I'm doing. It's not an experiment anymore. It's not something in the backwoods.

[00:03:40] And this is changing the tone of the whole work, what's happening. And that brings, interestingly enough, more business people to the discussions of AI. Before, it was very developer-centric. Now, the business guys get into the rooms. The bean counters, too. And so, there's a lot of things happening in that space. Yeah, I think every tech conference I've been to this year, it's all been about agentic AI. And Google themselves admitted to me in a recent interview that last year's conference that they ran didn't mention agentic once.

[00:04:09] And, of course, this year it's mentioned hundreds of times. It's all anyone's talking about this year, isn't it? Oh, yeah, absolutely. And my big word, what I'm after is obviously data. Now, you can say, Boris, you work for Mongo. Of course, data is the core. But when you take a look, it's not only agentic AI, but agentic AI in combination with all the things, what happens in vibing code, vibing applications. So, agentic itself is one part of the big change.

[00:04:37] The code changes and how people pursue applications is a second part. And that drives the part on the data side, which is naturally, obviously, for MongoDB a boom, fair enough. But we see this happening that data in the original form doesn't work anymore. And I'm glad you brought this up because a lot of conferences, another phrase I hear a lot is no data, no AI. But you've argued that many organizations don't actually have a data problem. They have a knowledge problem.

[00:05:07] So, tell me the difference between data, information, and knowledge, and why that distinction matters when building AI systems now. Oh, this is. So, now you're hitting like the Boris button. But pretty much, I compare this like this. Data are these tidbits of things. Seeing Boris is a banking client, I have an address which is stored in five tables. I have a balance of 500 bucks. Cheat dude. And that's pretty much it. And then they have other systems which may have information about my mortgage.

[00:05:36] They have other information which has things about whatever else I'm doing. But that's tidbits of things. The interesting part is the information is Boris is a client of the bank and has a total engagement with the bank of $100 trillion. So, I'm musking myself right now. And that is obviously a complete different view than the 500 bucks on my checking account. So, at that point, we talk information. But what is Boris's desire?

[00:06:04] What is Boris's skimming over our mortgage offerings right now? Boris moves from Paris to Barcelona. Ooh, maybe he needs to buy another apartment somewhere. These are suddenly where all these tidbits of data becoming information, becoming knowledge. Understanding of your client is not data points anymore. The classical enterprise data warehouse, right? Boris may churn as a client in the next 12 months because he's unhappy with his interest rate.

[00:06:33] That is yesterday. Today is. Boris is right now on our website looking for mortgages. And we have about 20 seconds to react to this one. There was an agent pop up. Hey, this is Billy, your agent for your mortgage. Hi, Boris. How are you? I can help you. You want to talk to me? That's a 20-second window. Real time. But I need to understand what is my desire? And that's knowledge. And that's where this knowledge garden factor comes in.

[00:06:58] And sorry if I'm bragging a little bit, but when you think about historically data were very well structured. Now, they are unstructured, semi-structured things. And obviously, a garden is bigger. A garden has parts like Versailles, full structured. And then you have my backyard, which is the opposite. And you need pieces of bows to make it really, really nice and that the bees are humming, aka the revenue flows.

[00:07:21] So that's where the knowledge garden concept brings structured, unstructured, semi-structured data, signals and context together, which is way more than just structuring your data. Banks have the info that I'm a client and have 500 bucks on my checking account, but they don't know what my desire is. And I love this concept that you've brought up here, the knowledge garden, incredibly memorable phrase. I especially love how you've described it there, bringing the bees in, etc.

[00:07:50] But what does a knowledge garden actually look like inside a modern enterprise? And why is it becoming so important for achieving that AI success? Oh, this is when you take a look, AI mostly started out in many areas from the analytics side. The analytics, the enterprise that are warehouse guys, moved up to machine learning, which was already fresh. And then they got the AI charter automatically. So that's very backward looking, analyzing data and looking at large volumes of data.

[00:08:18] Now in a knowledge garden, you need real-time data for AI. What is Boris? And what is Boris' state in the whole picture in a holistic view? And you're interacting with me. You need to have the agents being able to pull data in literally zero time. Real-time becomes really real-time like a breaking system. It's not anymore, oh, please hold the line. I'm checking your database. After the third time, you hang up.

[00:08:46] You don't want to spend your time looking at this little, you remember the Microsoft thingy? Yeah, you don't want to look at the ball of fun or whatever we call it in those days. No, you need rhythm. Hey, Boris, how are you? Hey, I'm looking for a mortgage. Yeah, I see that. I have your records here. Do-do-do-do-do. If you do this kind of interactions, you need a different access pass. So I need my data in a structure that I see the holistic Boris in this knowledge garden

[00:09:10] and the historic star schemas, the historic concept of ontologies don't work anymore. I need still understanding that Boris is a client, has a client. These are data points. They still exist. Versailles. But then there's the interactive piece. Boris was yesterday here and had a chat with the system, and he wants to maybe continue that one. He just hung up on the chat because he was sidetracked and his wife called for lunch. So how do we do this one?

[00:09:40] That's not data. That's real-time connectivity context. And that's where the knowledge garden acts differently. And we call those systems of action versus a classical system of record. And that drives a lot of different ways to deal with data and on structures as well, right? Suddenly, I become the consumer view with thousands of elements to myself. Some of them are maybe video parts, interviews, which I gave like this one. And other things are hard factual data that I pay my credit card.

[00:10:10] I actually do. And so those kind of things play into this picture of who's Boris. And the structure of this one is a knowledge garden. There's a second angle to it. The business connectivity, right? You want to know the mortgage department has a complete different view of Boris, the client, than my retail person who tries to sell me some investment plans. And they both have the same Boris, but they have different viewpoints on it. And that is where the code filing comes in.

[00:10:39] And now we're closing back to the agentic. Because if I have now agents who want to work with me as a client, as a consumer, or even worse, with other agents talking about Boris. And oh, by the way, Boris, we need to insure your house new because the value changed and we checked this one out. We can do this for you right here as well. Then my banking agent talks to my insurance agent, pun intended, and they sort this one out.

[00:11:11] This one, this is a complete new one. And that's knowledge. And you need knowledge graphs, knowledge structures, context, context memory. The classical relational system is simply dead. And that's where the preaching part stops. Sorry. And we will have many people listening from organizations that have spent years accumulating applications, databases, and so many integrations and APIs that were built to solve individual business problems. But we hear about technical debt a lot.

[00:11:40] But tell me about how you think this has created what you describe as integration debt. And why is that becoming such a barrier to AI adoption? Because a few years ago, everyone was talking about the age of the API, et cetera. But what has this led to from where we are now? Yeah, this is the integration debt is still there and it's getting deeper and deeper because as we talked about the knowledge garden, which has a holistic Boris view and where the mortgage

[00:12:07] department and whoever else needs to have access to data processing back office, choose your poison, whatever you want. And I love banks because we all have banking accounts and we all love our banks, right? So when we take a look on that one, the way how we have data pulled into these systems, I had a solution with one client at Bank and they needed 18 seconds to get the holistic picture together. 18 seconds. At that point, I hang up. Yeah.

[00:12:36] And this is where the integration problem starts to become really nasty. Now imagine you have an average 40 data sources. That's reality today. That's the main ones. And these 40 need now to be integrated to come up with. Boris is actually a client, is on the website looking at the mortgage people. The mortgage database need to be requested. Boris is a client database. You run nuts on this one. And even if you put sub agents who pull the stuff in parallel, you orchestrate it together.

[00:13:06] Oh, and then you want to be compliant. At that point, the challenge of the integration becomes minusculeous. And I saw people spending so much time building just agents who try to coordinate the access to data instead of focusing on what is actually the business problem. And remember how I started the discussion. Business people sitting in the room. Now they said, I don't care how you do this at the bottom. I want to sell more mortgages. I need to have five points more. That is really, how do I get there?

[00:13:35] That is a business question. And we at T people sometimes get stuck into my API call is more beautiful than your API call, which probably is true. But when we take a look how this works out, these access patterns become really a problem and the complexities. If I can buy a new mortgage onboarding solution and I get stuck on the fact that I can't get to the data of my mortgages, that's embarrassing. But reality happened last week. Very, very exciting situation.

[00:14:03] So these kind of things you need to hit. And it's very, very hard to have this wide agent and data triangle, which somewhere got disconnected if the data are sitting behind five integration layers, two caching layers. Oh, and then I built the meta cache. It's the latest terminal. The meta cache. The cache of all caches above. And I said, why don't you just build a new data layer, which is fitting to your modern

[00:14:30] data, which recognize the knowledge components you need to have in, which recognize the business domains need to be in, and which builds one structure. It does not mean you have everything in one single bulk. Back to your question, how does it look like? A garden has the orange trees, has the strawberries, and maybe have some beautiful oaks in the corner with a table under it. That's a garden. So the oak tree would be the example of your core banking data, the strawberries in checking account, and the orange tree, maybe a mortgage.

[00:15:00] So when you look at this one, there's different parts, but together it builds a garden. But suddenly you have components which belong to each other. And these components can be managed together. And that's why I love this picture of the garden. I know this sounds very basic, but when you think about it, there's a lot of truth in it, isn't it? It really is. And I also think one of the biggest challenges with all things agentic AI, and something we don't talk about enough, is that word trust.

[00:15:24] Because before a business can allow an AI agent to take action across workflows, et cetera, how should they evaluate whether that system is trustworthy enough for production use? It almost feels like we're only weeks, months away from hearing a few bad examples out there from people that don't evaluate correctly. But what should they be doing here to establish that trust? Neil, this is literally at the end, the core, right? Now we are establishing the data. We know how we build. We can vibe it.

[00:15:53] We can agentic it. We have the orchestration. And now, can we trust it? And then every look in the room like, uh, who's responsible for this one? Like, and everybody points at everybody. So we came up with two metrics, which is kind of funny. And this is maybe not sounding really exciting, but we call it the agent confidence score and the business risk score, the ACS and the BRS. And we came out with this one with something very basic.

[00:16:21] When you start running your system, you will never be perfect. And that's the first part where clients get normally nervous. What do you mean it's perfect? I know we have hallucinations. Hold, hold it, hold it, hold it. When you have humans doing the work, they do errors as well. How do you do when you have humans? Oh, we have four-eye principle. We do same thing. If a case is pretty standard and you can validate with small, uh, small models, you have rule-based

[00:16:47] systems, you can validate the outcome of an agentic system and you can grade it. And that's the agent confidence score. You grade the agents on their behavior and you come up with a confidence level. Based on input data on results. Does it make sense? Is it in a context that it looks somewhat realistic? So if somebody has a windshield repair and it's a six time in one year, so there's one

[00:17:13] data point six times in one year, which maybe opens up, there's maybe some fraud involved. So, but the system says, yeah, send it up for approval. And the system says, yeah, my agent confidence score. They don't understand that part six times in a year. So let's push this out and confidence score goes low. This is a very basic example, but you can build with small models, very good judges. So this is who guards the guardian. And you build small judges, which much with judge the system.

[00:17:41] And then you say, what is my ACS? My confidence score. And you can say, well, if the confidence score is one, never hallucinates. It's perfect. Like every human being. No, it's not. So you go and say, where's my threshold on the confidence score? Now you look at it on the backside. You have all the experience from your typical business scenarios where you look in what is the decision factors where we say a decision needs to have a review. Where does it go to a next level audit? And that's your business risk score.

[00:18:10] And that is as well as something what the business people understand saying, okay, I understand if I have a business risk of 20%, I may be willing to take that. Normally, I'm comfy with 10%. That means 90% of the solutions go through. The 10%, they go to human review with agentic review level. Or maybe 5% go to level one with humans. And the second one gets a complete case review because there's a high fraud detection.

[00:18:39] So and with these two measurements, you will be surprised. You can make some very simple meters based on the historic system because you are replacing normally workflows. You know what the quality of your workflow was. Based on this one, you built on the agentic side, the agentic confidence and the business risk confidence. And that helps two people. This helps the techies on one side to understand how good is my agentic setup.

[00:19:05] And the other side, which is what the business people taking the risk. And these two things together are amazing, simple to implement. And what we have seen over the last months is I get a lot of this. Now I get it. Because this is what the business people at the end underwriting the decision that the system works. And hallucination is nothing else than a human error. You can argument, well, but it's a machine and the machine should be always. Otherwise, it's not. We know what models are by now.

[00:19:34] We don't need to get into that one. That's a model is a model and a model can make the wrong things, the wrong path somewhere and that happens. So deal with it. If we see something goes the wrong path, send it to the humans. And that's the same thing how you would work otherwise in a workflow. So this is how we drive this confidence. The second part of the table stakes. It's amazing how many people ask me, what do you mean? I need an audit trail on every agentic decision.

[00:19:59] I need to track every single decision by weight, by what, by extra, which brings us back to the knowledge garden and the context. I need to store all of this. I cannot ignore that because for the ones which I pass out where I say, this is not kosher, I need to have all the details which led to this one, the whole data lineage. And that's where the other part, the auditability and compliance, regulatory part in many cases,

[00:20:27] or even in a retailer to understand I have 80% return rate on these shoes. Something is bloody wrong here. And then comes out, oh yeah, we ship always to lefts. Brilliant. And I think also over the last, what, two, three years, hallucinations are something that have dominated just about every AI discussion. But you suggest the conversation should actually now be much broader than simply asking whether the answer is right or wrong.

[00:20:55] So again, for people listening, how should they be thinking about confidence, risk, governance, and human oversight when deploying AI agent? Again, a lot to go at here, but what should they be doing? Oh, this is pretty much the next level. When you look at it, you need to build the data lineage, what we just discussed shortly before, and to look at the holistic business process. Normally you don't build agentic systems just because we're agentic now. Oh, we're agentic. No, you built it.

[00:21:25] I have a business problem to solve. And that system is offloading my humans to do smarter stuff. And what does this mean? So in our case, we want to offload standardized insurance cases. We want to offload standardized mortgage requirements and requests from people. We want to help on the doctors to ensure that there's no double impact on any medication people give. These kind of things, these are supporting functions.

[00:21:51] And you want to make sure that for the part where the system is deployed, it behaves as you want to be. And to be honest, models will change. We see model change every three months. If somebody tells me my model is 0.5% better than the other model and say, great, I don't care about your 92% versus 92.5% because my cutoff is 80%. And that's the trust part. This is how do you go for that one?

[00:22:16] And to bring that kind of leverage down, this whole hallucination discussion was, I think, something as well where people expected AI systems like in the really bad movies, Hall, anybody remembers, right? Killed the humans in the spaceship because they figured out they did a mistake. We don't want to be in that situation. Let's be realistic. These are mathematical models. They are trained on something and they make errors. And the errors are called now hallucinations. So I'm very, very grounded in this one by now.

[00:22:45] I admit I was really excited at the beginnings as well, like free cars and things like that. But that is, I think, gone. This is by now, it's more practical infused. And again, for people listening, feeling somewhat inspired today, maybe they want to finally move beyond those impressive demos and proof of concept, et cetera. What are some practical steps that they could take to build AI systems that are ultimately more reliable,

[00:23:11] accountable and capable of delivering real measurable business value and improving business outcomes? Because that seems to be where the big focus is this year. But getting there, it can be a little bit more complicated. And then this is a nice part where obviously I'm very zen. The knowledge garden, the knowledge garden drives the capabilities for you and activates, enables you. You can't throw out the other stuff here. You can't.

[00:23:37] There are too many processes on it and this discussion about modernization, integration depth and so on. You need to solve that. But right now, when you look into AI, we can drive the data into a knowledge garden setup, which is pretty much in the old terms an ODL, an operational data layer, which brings the data forward. You have them integrated in a new way to fit for these agendic solutions.

[00:24:03] And that suddenly unlocks a key for the kingdom that you can vibe code, you can vibe agents, and you have suddenly systems interacting with each other. And that drives in the business discussion where the business says, we need X achieved. And you can say suddenly, I can build this for you in four weeks. These are the risk scores. These are the things. We do this every time by now. And we love to tell people, do it in workshops with your business people in the room. And most of the people, look, you don't understand. We have a consultant.

[00:24:33] We have a business consultant. We have our business people. We have our tech people. We can't get them all into one room. Just, you know, maybe get some consultants in front of the door first. Start with yourself. Build your knowledge garden, your understanding. It can be a prototype. Two, three sources. Start with this one and talk to the business about what is the outcome. Vibe something together. Is this what you see? It doesn't need to be perfect. We're not talking about taking white coat, which in two days into production, a bank. You can't do that.

[00:25:01] But we can start building the concept and see people, this is what we can achieve. Because to be honest, most cases, the business people heard about it. And all what they heard was hallucination, risk, and doesn't work. So now if you give them a realistic thing, this is how we deal with your data. This is a new world. This is how we vibe agents. This is how the agents adapt. This is your ACS, your BRS. And this is a business solution you're looking forward. Is this what you want?

[00:25:30] You get very pragmatic. And we love to be in the workshops like guides. In the old days, you called the scrum masters. We're a little bit like the agentic scrum masters kind of here. Helping the people to get over the first two or three. Then normally they're off to the races. And that gives you really reliable solutions. Specifically, if you have the checklist saying you have an ACS, you have a BRS, you have a compliance, you have the trust pass. You know how to validate the stuff.

[00:25:59] You will be surprised when you have this whole checklist and you go up to a senior exec saying, we've done our homework and we have those lists. We're happy to share them with our clients. If you share those, they say, wow, you guys have a plan. And then they're, uh-huh. And then it's becoming way better that we do agentic AI. Good, good. The board asks for agentic. You have a plan? Good, good, good. Don't implement. Don't take risks. Yeah. That is basically our day by day.

[00:26:28] And I always tell people, if you don't start with your knowledge garden, everything you wipe on top will be built on quicksand. And that is a really ugly term. And we saw that coming 18 months ago. And obviously, everybody tells me the whole time, Boris, Jason, yeah, I understand MongoDB, that's self-serving. It's not. Even Elon Musk posted yesterday about Jason. That was really funny. So, and invoking Elon is always dangerous, but he posted about Jason.

[00:26:59] And that maybe gives you an idea. And if we were to look ahead, as agentic AI will inevitably become more and more embedded into business processes, what do you think will separate those organizations that are able to follow your advice there, successfully modernize around knowledge and trust? For those that end up with just another generation of disconnected AI solutions that are, as you said, they're ultimately built on quicksand.

[00:27:26] I think the fast movers or the, let's say the intelligent fast movers right now, they will really drive a complete new business. And we see this already in some areas. Mortgages is one typical example. When you see how people automate this, simplify this, 1500 pages documentation becoming suddenly an online process, fully automatic, because the data sit somewhere anyway. They will gain market share. They will move faster and they will drive a much higher NPS.

[00:27:55] That's really the part, the net promoter score of these clients and these, sorry, of these companies will be coming so much higher because their clients will appreciate the service level. And those who don't, yeah, I think there will be a change in that regards. And agentic is really about satisfying requirements. So agentic should not be self-purpose. I built an agentic solution because I built agentic solutions. The board needs agentic. Therefore we do agentic.

[00:28:23] And you invoke the word AI every second sentence. That is no value. That's, that's the, that's the old school. We sit on top of APIs. It doesn't work, but we have a solution. You want to change your business. You want to change your vision, how you interact with people internally and externally. And these are the companies which will embrace it in the right way, secure way, and will really change, to be honest, the way how the industries work. And I'm really looking forward to that one. My best example is right now automotive.

[00:28:54] When you see how cars changed from, I am having a key. I go in the car. I start the engine to, I talk to my cell phone, start the car in the basement. It's cold right now. And the system starts up, electric, fair enough, and starts ramping up the temperature. It's minus 10 outside. Get the battery warmed up, everything, and keeps the temperature exactly at my preferred level. Sees when I came downstairs, the doors are opening while I'm coming to the car.

[00:29:23] And the car says, good morning, Boris. Are we driving to work? Yes. And my Navi is set up. That is context AI. And this is what wins. And that is the same across all industries. For me, this is really the most fastest moving business part right now. And I think that is an inspiring and thought-provoking moment to end on. And I love listening to you today about why enterprises are turning to these knowledge gardens that you bring up.

[00:29:51] Great concept about structuring and cultivating knowledge just like a well-kept garden where information is planted deliberately, labeled clearly, and tended over time. It's a beautiful analogy there. And for people listening that are feeling somewhat inspired, wanting to talk with you, your team, and the work that you're doing at MongoDB, where should they go? Where would you like me to point, everyone? I'm in the lucky position that Boris Bialek, B-O-R-I-S-B-I-A-L-E-K, is unique.

[00:30:18] Even with 7 or 8 billion people on the planet, I haven't found a second one. So, if you look me up on LinkedIn, you find me very quickly. I'm reacting pretty fast and reply to things. We have as well, if you just Google MongoDB industry blog, you will find our blog where my team posts with all this stuff. But I'm happy to connect you as well with the people. But our industry blog is actually pretty cool by now that you find all these things, what we talked about, practical, described, and practical solutions. I always invite people to check outside of your industry.

[00:30:48] If you're in banking, check what manufacturing and retail are doing. You will see the concepts are applicable. And this is the best way to hit us and to find us. And yes, we're very happy to come in for workshops and go deeper into the various solutions. That's our life. And that's where the geeky part of me comes in. Love it. Well, I will have links to everything. So, if you're listening to this and you're interested in learning more about what knowledge guard and thinking looks like in practice, how it might look in your organization,

[00:31:16] or just talking about the difference between data, information, and knowledge layers for AI. So many big talking points there. Go over to techtalksnetwork.com. There will be a blog post associated with this episode where you can find all that information. But more than anything, Boris, big pleasure as always have you sit down with me, especially on an incredibly hot day. But thank you so much. Appreciate you, Tom. Thank you for having me.

[00:31:40] So, as we wrap up today, I want you to ask yourself, do you and your organization have a data strategy? And does it have a knowledge strategy? Well, Boris shared a powerful reminder today that successful AI projects begin long before the first model is deployed. They start with understanding context, trust, and governance, and how information flows across the entire business.

[00:32:08] And I think his knowledge garden analogy offers a simple way to think about a challenge that so many organizations make far more complicated than it needs to be. But over to you, if today's conversation sparked a few ideas, challenged a few assumptions, or helped connect a few dots, then we've achieved exactly what we wanted to do today.

[00:32:31] But as AI moves from experimentation into everyday operations, I really do think the companies that cultivate knowledge carefully could find themselves with a very different future from them still wrestling with disconnected systems. But over to you. Where does your organization sit right now? And what would your knowledge garden look like? TechTalksNetwork.com. You'll find 4,000 interviews there.

[00:32:59] You can leave me a recorded message or send me a DM. Whatever it is, connect with me on socials at Neil C. Hughes. And we'll keep this conversation going. But that's it for today. So thanks for listening as always. And I'll speak with you all again tomorrow. Bye for now. Bye.