Transcript
0:10 All right. Well, today on uh full context, we have Emil. Emil is the founder and CEO of Neo Forj. Um Emil, thank you so much for coming. I'm really excited about this conversation today. >> It's fantastic to be here. Great to do it. >> Awesome. So, we have a full docket here. Um some some really interesting topics that we're gonna be covering. But before we kind of get into uh the the nuts and bolts of Neo forj about context graphs and more um could you tell us a little bit about how you got here? Um who is Emil? Why did you start Neo forj? Why graphs and and kind of what the story of the company was?
0:43 >> Yeah sure. Yeah. So I've been doing this for about 438 years right you know at this point. uh we raised our series A back in >> 2011 but we were working on the technology you know even prior to that and and it's the classic thing of some combination of like an itch that we had ourselves that we wanted to scratch right along with some amount of maybe analytical kind of thoughtfulness right and and the itch we had was that we were implementing what people today would call multi-tenant SAS systems that were dealing with very complex data. Of course, this is like 20 years ago. So, you know, it was like like millie data or something like that, right? Like in comparison to what we see today, but it was very complex. It was changing all the time. It was deeply messy, highly interconnected.
1:38 And you know, our startup, we had a 20 person engineering team. Half of them spent the majority of the time fighting against the relational database. And this was a big surprise to me because even though I was pretty young at the time, I just didn't been doing software for my entire life basically. >> Um, and in all prior projects, the database had been this accelerator which like took this big gnarly problem of like storing the data, making sure that it's there, queryable, it like transactional and online backups in production. It's like all these big hairy problems, right, that I didn't have to focus on so I could build the application. But that clearly wasn't true this this time. And >> yeah, >> fast forward to today, you know, I think we have like a as an industry, we have a clear understanding and and a language for it. But I we didn't realize that at the time. We just felt it. But there's this mismatch between the shape of the data and the abstractions exposed by the underlying infrastructure. In other words, some whatever round peg square hole thing going with the square and static relational model and this super complex world. Yeah.
2:43 >> And you know, we were young enough and naive enough to like we started looking around for alternative databases saying like look what we what we need here is exactly like the oracles or the informics or whatever postgresses of the world >> but just with a different data model right >> and we couldn't find anything and that's where the kind of the youth of the nate kicked in because then we just said screw it let's just build it quote how hard can it here I am 20 years later turns out building like a fully transactional database like from a tech perspective right and then building a database company like both of those two things turned out to be really hard and then we took on category creation you know we we coined the term graph database like initially right and so there's like an end of one when we we said those words right and so we just decided to take on a lot of of of big tasks but but ultimely lose that that script that that that itch that we wanted to scratch ourselves, right?
3:47 >> And >> and then I said there's also like one more analytical thing which so I'll I'll add that layer to it. Um those are very pragmatic and practical but like the one thing maybe where we were a little bit thoughtful was like like on some level what what is data right and you know lots of people operate with data and work with data but like don't even think what data is but like data is a representation of the physical or sometimes the digital world right so but let's just focus on the physical world so we describe the world in digital form so that we can operate on it with computers right >> and so it goes to say then that if the world is becoming more and more connected, which we all felt that was true, then data describing that real world will become increasingly connected. There's going to be an increasing amount of value in being able to figure out how things fit together.
4:36 And so we felt there was some kind of notion that if maybe there's only us in the entire world that are experienced this right now, but like tomorrow there's going to be more and the next day there's going to be more, right? So we felt that maybe there's some kind of a win behind our back from from that perspective. I see. Well, actually like I've I've experientially felt this many times uh where uh graphs tend to feel like the right abstraction uh to model a lot of the world.
5:05 >> Um and so I guess there's there's a few questions that come from this. First is at this point, do you feel like you've scratched the itch? >> Um the second question, >> sorry. Yeah. So, so, so I do, but of course the moment like you solve it for that kind of shape, then it's like you find all these and like the world moves on and data sites become bigger and problems become more multifaceted. So, we solve that particular kind of but there's there's plenty more out there to scratch. Got it.
5:37 Well, then I guess I guess on the on the other side then like in which cases do you feel like relational is still the right way to go uh versus a graph database, right? Like maybe maybe help like disambiguate those those cases. Well, because in many cases actually and maybe this is a fallacy in like thinking about you know graph relationships like the the relational kind of you know the the foreign key you know primary key like that sort of thing like tends to feel like a natural extension that could help you model graphs.
6:07 >> Yeah. I think what we've learned like we used to live in this very kind of uniolar like everything was a relational database. Yes. In the 60s and the 50s there's like the cotisil model and network databases and some other force but in reality for a bet trillion years the relational paradigm has been the dominant one. >> Yeah. >> And then in the last call it 15 20 years there's been new categories emerging. >> These new ones at least the established ones tend to all be isomorphic right which is a fancy way of saying that you can take data in each and every one of them and transform to the other without any loss of information. Right? and transform therefore transform back as well. Right? So in other words, there's no data set out there that you cannot model in a relational database >> and there's no data set out there that you cannot model in a graph database.
6:55 >> So that's no longer the interesting question. Can you do it? The question is is it a good fit for you? >> Yeah. >> Right. And to me then it comes back to like what's the shape of the data and the queries that you want to run on it. And so my go-to example is if you have a payroll system, >> every single row structurally is identical to the other ones. It's like first name, last name, role, hire date, and salary, right? And the query is like either fetch individual rows or like calculate average salaries of people hired in this and that year, you know, that kind of thing, right? It's extremely tabular. And if you go out there and you talk to domain experts, people who don't know anything about third normal form and er modeling or any of the things that that we've learned, you know, um then and you ask them to whiteboard their world, >> if you work with payroll, you're going to end up drawing tables on the whiteboard, right? But there's a ton of domains where when you ask the experts, unless they've been trained to take the conceptual model and break it out into tabular form, that's not naturally where they go. They say, "I have a shopping cart. In that shopping cart, I have order items." And they refer back to my products which actually sit in a hierarchy. And it builds up this this graph. And so when you see that, I think that's a good hint that it's there's probably low impedance mismatch between that domain and the the graph model.
8:27 >> Got it. Got it. You know, one of my first experiences with Neo Forj was uh when I was like 12 or 13 years old. Um and >> so amazing. >> It was uh yeah, I mean I think uh ran the desktop community version on my laptop and uh I was I was like always interested in AI and and you know language modeling and stuff like that. This is like many years before uh LLM. So we were using very different techniques. Um but we were basically trying to extract entities out of unstructured data and model it as a knowledge base and we ended up using Neo forj just it was um it it felt like the right tool to use uh to actually model out these knowledge bases and then more than that you guys had the little gooey where you could like move the dots around and that was the coolest demo I've ever seen. Okay. So, so, so two two comments on that, right? Like one, that's amazing that you did that at age 12. First of all, >> um, second of all, man, that graphicialization. I can't tell you how many times that I walk like into a customer, big Fortune 500, big banks, and what I see is like their domain models printed out of we have this graph visualization tool called Bloom now, which is like a fancier version of of of that, right? But even Neo forj browser which is the more code centric one with the query language and stuff like that right >> people just print that and I've heard so many times people say oh by the way my boss and my boss's boss has this one in their room >> right and so there's something there where intuitively people look at the graph model and it makes sense to the business people and it makes sense to the geeks like us right and that's a really underappreciated we call it whiteboard friendliness right like an appreciated property I believe of the property graph model. No pun intended.
10:13 >> Yep. No, absolutely. Well, so actually like maybe maybe transitioning to the to a little bit of a broader topic somehow relating to Neo forj, but I think that that where I want to start this is a little bit more wide uh in the aperture is you know thinking about everything that's happening in the market right now. Um I think over the last you know couple weeks maybe even months now um you know we've seen a very different view perception um of most of the SAS vendors that we've seen out there right Salesforce Service Now um kind of being like the two prototypical examples where the platform shift as you know the world is increasingly looking towards AI looking towards agents uh to actually automate and do own the work um changes the way that you actually value uh these kinds of companies and with it has come uh this view that value is essentially going to acrue in two far corners of work. It's going to be the infrastructure layer and kind of the the the lowest layer that it can possibly acrue to there and then it's going to value it's going to acrue to the agents that have the expertise and the experience uh to actually do this work.
11:21 And so maybe actually I'm curious like Neo Forj aside like what is your view um on you know the stuff that's happening in the market right now? Why do you think it's happening? Um yeah may maybe we can start there. >> Yeah. So many thoughts here and it's such an exciting time right? I you know it's it's funny right like I've lived through a few platform shifts. Yeah. >> Right. And I I wasn't really kind of fully aware of the shift from my main mainframe to client server, right? Yes, I was technically alive, but I wasn't really kind of too too plugged in in in into that. But I obviously gave birth to Oracle and then we had the shift from client server to web which gave birth to initially my SQL and then obviously I was born like Neoj was born in the shift you know from back to the cloud. Yep.
12:12 >> Right. And then you know and so I've seen this happen you know but this is where like I think it's a platform shift exactly as you said it's also the fourth industrial revolution >> right which like cloud wasn't or and mobile wasn't and certainly mainframe to client server wasn't but like some version of steam and then electricity and call it computing and now now AI so it just feels much much bigger right now >> and you know Buffett what What's the quote from Buffett? Like the stock market is a voting machine in the in the short term and like a weighing machine in the long term. I certainly think there's like some over pivoting going on right now because the stock market hates that uncertainty and people just choose to stay on the on on the sidelines as this is is playing out. So it's probably seems you know a little bit kind of over rotation on some of those things. Right.
13:08 Having said all that, we had firsthand view into like the early version of this because I think it's almost exactly a year ago that uh Sebastian Simitovski was the CEO of the fintech CLA. >> Yeah. He wrote this long are they called blog posts on X whatever long forum tweet >> right >> and he was triggered because a few months earlier he'd mentioned to investors that we've thrown out Salesforce and workday >> and yes we pay Salesforce because of Slack but not for any other reasons >> and we've done that thanks to our vibe coder I don't even know if you use that if maybe that was pre that term I don't remember but like thanks to AI productivity >> and that generated all kinds of noise.
13:56 The next week someone asked Mark Beni off on stage at a conference, right? And then he ended up writing this wrong long blog post when he said, "Yes, >> this happened. AI was a part of it." But the real reason we were able to pull this off is that we got our data house in order and we consolidated our data into Neo for a property graph database called Neoforj. >> Yep. >> Right. And so we saw this happening kind of firsthand and and it is true like they threw out 1,200 SAS tools like thanks to this, right? And and so I I still think there's like there's something very real here, although probably some kind of overcorrection, right?
14:38 >> Yeah. I I guess uh so actually the last time I talked to Sebastian he he mentioned something similar um where he was you know thinking about you know how do we actually centralize kind of the organizational context and um you know there's there's there's so many different kind of like layers to peel back here right but one is um a lot of what used to live with these systems of record um essentially it became it became a moat for these companies over time uh but at the same time I a lot of enterprises are basically feeling that it becomes a point of leverage uh that these vendors end up having over them.
15:17 Um and then in agent economy where agents are actually making more and more decisions the recordkeeping of how we got to those decisions and all of those other things end up being you know essentially held hostage uh by these systems of record. And so therefore you don't want to use the systems of record anymore and actually kind of abstract and represent the data in a way that is representative and and uh expressive enough to fit like the the the the curves of the business, right? Um so Clara looks different than a Salesforce that looks different than a player zero and a Neo Forj. Um and so the way we would want to represent certain pieces of data as they you know span across multiple functions in our in our company would want to look different and therefore you know want to internalize a lot of these things. The other like this is like the classic like build versus buy sort of debate though right where is like the the the maybe this actually transitions a little bit into the context craft debate, right? But um what does that store actually look like? Um, and then how does it evolve over time, right? Like should everybody in the world be spinning up a Neo Forj instance? First of all, is it in Neo Forj? And and maybe maybe you're maybe you're a little biased there, right? But like is that in Neo forj? Is there a universal kind of context layer that every every engineering team every every company should be investing in? And then yeah, like what does that look like? And and is the company that is is the company itself the right person to actually go and uh build it?
16:45 >> So many thoughts here, right? like this for forever. I have to comment on one thing just on the on kind of system of record piece which is like >> I think what we're all sensing is like okay we had this super coupled thing where we had the single source of truth piece coupled with the UX right and everyone was forced to use that UX and there was some value in that UX because it encoded some version of best practices right for that business process but it was also built for the average marginal next customer, right?
17:21 Which means that was it optimal for your organization? Very likely not. Maybe you could get it there with like a million SAP consultants and and whatever, right? But like but ultimately probably wouldn't be, right? >> But I think the foundational piece of it though like the single source of truth for that data that is still really valuable, right? And I think the big debate like and so okay so then like the work is going to move somewhere else right so does that mean that the profit pool stays with the systems of record or if it's does it sit with whatever layer is doing the work right and they will still have CRUD interfaces I think because you still want to be able to edit the the single source of truth as a human being administrator you know you kind of thing right but is that where most of the value occurs over time I think that's the big debate like and I it's very hard for me to to understand and obviously like if >> you know I guess maybe coming back to the context graph right like if there's a lot of value in that doing the work piece and in order to do that as an exhaust right like we get the decision traces that build up the context graph and something else ends up being the system of record for that context graph >> right how much of the profit pool flows over there probably quite a lot would be my guess.
18:47 >> Yeah. >> Right. But I don't know what what what your take is. >> Yeah. Well, I mean I think the the systems of so so may maybe it's like uh it's worthwhile to actually kind of define what a context graph is. Um and then then maybe it's it's worthwhile kind of talk about how that relates into systems of record. But my my view is right like systems of record are really well positioned to capture the outcome of certain decisions, right? And so we have you know Salesforce which uh you know holds the outcome of you know many different personas operating together to lead to a particular deal and therefore that manifests in a customer record that shows up in Salesforce. But all of the conversation about how that deal was made right you know automation Emil talked and then after that then we you know had our reps talking and our engineers talking and you know the quarter needed to close and this is this is the discount we gave and all of those different things. uh end up being decision context that is important to understanding why the outcome actually happened um but is actually end up ends up being lost and ends up kind of living in uh um in in in the minds right um of the people who actually executed it. Um and so that was actually the the core realization uh that led to you know this this whole kind of context graph debate.
19:58 Um and what I think a context graph is is basically a fabric um of these decision traces, these decision trajectories uh that lead to certain outcomes. And over time when you start weaving these things together, you're essentially encoding an institutional memory, right? Um that's ultimately what I think a context graph becomes. Um and I think it's a a huge point of leverage uh for enterprise um and for really any team as they kind of grow and accumulate knowledge both as a function of their teams growing as a function of you know the complexity and customer bases growing um to be able to encode and remember why the business actually came to be uh in that way. Um, and the reason why this is especially important right now is because the amount of work and the amount of decisions that can actually start happening are no longer constrained by the number of people you have in the company anymore, right? And that's agents, right? Agents can actually be doing work uh well, you know, 24 hours a day, 365, right? You can scale these things in in ways that we couldn't really imagine. They're incredibly smart, incredibly agile. Um, and so we need some way to actually give them kind of decision context so that way, you know, we're not sitting here basically being the human MCP for these agents.
21:13 Um, that's that's kind of the the the short of it. Um, but going back to kind of your question about, you know, systems of record, I don't think systems of record are actually well positioned to capture these these these decision traces. Um, and I think over time, you know, if we think about how how much we pay people versus how much we pay SAS, right? Like we pay people a lot more comparatively. Um, and it's because they have that decision context. And so what I think is going to happen is actually like, you know, I don't think Salesforce is going to lose a tremendous amount of value um in like value creation, but what I think is they're going to own a much smaller piece of a much bigger pie. Um, right?
21:54 like there's going to just be a little bit of there's going to be new vendors. Um there's going to be new startups that are actually owning the the value creation that happens on top of these systems of record. Salesforce will remain a database. Um and I think over time what's going to start chipping away at Salesforce's mode is going to be people basically asking why do I need Salesforce at all? Because is it isn't it just a guey right on top of a database? Um and so yeah, I think they're just going to own a smaller piece of a much larger pie. Um, and I think that's where kind of like the discounted cash flow piece of it, right, implies a smaller valuation and and a market cap. Um, that's that's that's my take. Um, but I think the counterake to this is just like they have like these these SAS companies have distribution. Um, they have tremendous staying power. Um, and what if they actually are able to go and start capturing these these decision traces um, and encoded themselves, right? I think that's that's kind of the counterpoint. And maybe there's a different way to think about it.
22:59 >> Yeah, I think of it similarly. And what I loved about the context graph like the initial two blog posts were just amazing like Jay Anashu and and and yours. And to me it completed the circle or completed I I have this four quadrant thing in my head which is I think there's exactly four data sources or types of data sources. >> Yeah. >> That agents need to reach escape velocity >> in reality in production inside of enterprises.
23:32 >> Right. And context graphs completed that four quadrant thing for me. Right. And and the first one is the operational databases or the systems of record, but I I tend to think of them as systems of record for the present, right? They hold the current state as you just said eloquently, right? So that's kind of the the first one, right? The second one is the data platforms or the data warehouses. And I think of them as the system of record of the past, right?
23:59 Like a you know, how much did we do in sales in LA dam in Q3 last year kind of thing, right? you know not how much is a customer paying me today in this very moment that's system record of the presence that's operational database that's that piece right >> y >> and then the third one is kind of agentic memory right which is a little bit contrider system record for conversations not entirely the right maybe system of record for agentic state >> right is maybe the right way >> and then the fourth one is the system record for decisions so the context graph >> y >> and those are kind of the four sources in my mind and I don't think it's as easy as you need all four to reach escape velocity because I think it's use case specific there's some kind of an accuracy threshold for like good enough per use case per situation you know kind of thing right and I'm sure you can get there with just one of them in some situations >> I do think generally speaking though the more the marrier and really for the complex and probably the most valuable ones you need all four >> what do you think of that framework.
25:05 Does that resonate with you? Is there something that you see that I've that I've missed? >> No, I think I think that's actually really well put. Um the the the side effect of this, might I add, is like you actually end up seeing a change in the shape of work itself. Um, I think one one interesting kind of tidbit that that that we're learning and and at player zero, right, we're an AI production engineer and we think a lot about what it takes to maintain and operate software after it's left the IDE. Um, this usually touches many different functions. So, it touches, you know, QA and the pre-production side, but also S and support and security and all these different functions. At the end of the day, decision making um for why we built certain things or how we fixed certain things end up spanning multiple different functions. And as you start building these different systems that uh the four that you described, as you start kind of building and and maturing these uh what ends up happening is a lot of these functions end up collapsing into one another. Um and I think this is this has been like a really interesting uh thing that we've started kind of noticing right like the line for example between S sur and support uh ends up you know blurring a lot uh because the context required to go fix an issue when it's a ticket versus when it's an incident ends up looking very similar um and then you know pulling that one one level further you know in order to do great QA you need to understand how things broke in the past right and so these are all kind of decision traces that end up kind of representing the threads between all of these different uh functions. But I think that's a little bit of the outcome, right? Like the output uh that ends up kind of uh being created when you start representing these four different systems of of record or I guess data systems.
26:55 >> Yeah. I've been using it like data sources or or something like that, right? Yeah. >> Yeah. Yeah. Yeah. Um but no, I I think it's actually really well articulated and and maybe I guess you implicitly answered a question that I had which is like do you think that there is a universal uh data system uh that exists that exists for the entire company right across engineering and sales and um you know every every every single uh >> You're nodding your head now.
27:28 >> No, no, I'm I'm I'm thinking not right because I think look >> so maybe let me take the context graph angle for this. I think we're all sensing some kind of future state and I think you're living in that future state with some of your customers, right? Where there's like my agents are up and running >> and they have access to whatever source data they need maybe across the four quadrants in our kind of in our framework, right? Including the decision traces of what's been actually going on in the real world to make the these types of decisions, whatever decisions that that the agents uh need to do. Yep.
28:06 And that's amazing. They've reached escape velocity. All systems go. And it's probably like a self-improving system. Maybe this is the naive view, but like in theory, that's what it should should be, right? It should compound over time as a flag, right? >> Okay, great. But how do we get there? Right? Like that that's kind of the the the state, right? Yeah. >> And and I think there's some kind of a step functiony thing going on here where you chew off in my mind and what I've seen happen in in the Neo Forj kind of customer user base is that you take a very narrow problem, right? and you you solve that and sometimes you have enough data without the decision traces to get above the good enough accuracy threshold and so therefore you can just start bootstrapping start accumulating decision exhaust and all that good stuff right sometimes that's true sometimes I see people you know plugging a human in the loop in there and then instrumenting the hell out of it right but like in a very narrow way and then you branch out >> from from there Right.
29:15 >> But >> can it then end up in being this one big platform backing? I just I don't see that happening in any future that involves like non-trivial. Look, who knows? Like people talk about the what what what is it like the billion dollar like the uni one one human unicorn startup and stuff like that. Maybe just you said the shape of work will change. Maybe this the shape of everything is going to change so much. But certainly looking at the global 2000 today, I have a very hard time seeing it.
29:49 >> Yeah. No, I I I completely agree with that. And I actually I I I think we should actually double click on the how to build the flywheel piece. >> Yeah. >> Because this is actually really really important. Um and it also lends itself to a question that I get asked often. Um which is uh is building context graphs an application layer or an infrastructure layer opportunity? Um, and I think the the question within the question there is, you know, and maybe you answered it, but I'd love more color if if uh if you have any is, you know, at the application layer, you pick a narrow use case where you can actually start instrumenting the decisions required uh to go and build what later and over time acrews into an infrastructure layer opportunity. But you can't really just start with the infrastructure layer opportunity and expect it to become anything because those decisions are actually not recorded anywhere today. Um, is that is that roughly the way you think about it?
30:43 >> Yeah. Yeah, that's exactly how I think about it. Like and and you have to seed it somehow in most situations. I'm I'm sure there's some kind of long tail going on here where maybe that's not true, but like I think for the majority of situations, you have to seed it and that means starting in the application layer, which in of itself is very hard to know. Like I feel like all the lines are getting blurriier and blurriier now, right? Like >> you know agents that's sitting on top of data platforms is that the application layer or or not. But kind of conceptually you have to start with that real business problem to to solve in order to kind of bootstrap the system.
31:19 That's certainly my point of view and what I've seen kind of across our our our user base. And the first time just just for reference the first time I saw what we now know as a context graph with like using the decision traces as an important source into the agents. Uh the first time I saw that was around uh not like nine months ago last summer so summer of 2025 when like a massive tech platform was building out. They already have had Neo forj to s store call it their rag corpus right to express that in in graph form using graph rag right and or to be able to use graph rag >> but then they started building out what they call their agent brain or agentic brain >> also on neoforj storing again the output the exhaust from a lot of their kind of agentic traversals right and then they started using that it's like >> that I thought that was very cool right didn't have the terminology for it until December 22nd of of of 2025 when the context graph >> everything.
32:26 >> Yeah. Yeah. Exactly. Right. Um but so yeah, so so so I think that that's that's kind of what I've been seeing like in in terms of bootstrapping the the the flywheel. >> So this this actually takes me to to another question that I often, you know, end up thinking in many different lights about and and maybe you and I disagree here. So I have to ask like how does a context graph actually get materialized right like is it actually a graph um is it not a graph um maybe in the in the case that you described like this agent brain um you know like what is Neo forj's role in this if any and then how do you see that evolving >> yeah look I tend to love graphs in all shapes and and forms and then of course we built a product that is really there's like the impedance mismatch between kind of a graph data model like something expressed in in that way and the underlying substrate is really low like so there's very low friction >> right um and so I think it's it has offers many kind of benefits but I'm also fully aware of the fact that the default database for most teams today certainly their startups is Postgress like some flavor of of of Postgress right and in the enterprise it's probably some version of Oracle, right?
33:46 And it's back to where we kind of started, which is these are all isomeorphic models. So you can put any kind of graph structure. You can put it in a key value store if you want to. It's just a lot of work, >> right? Um and so so I I think I think we're see we're going to see like a lot of different ways of of of doing it. I do think like for many teams like the fact that the data model is such a good fit makes it like a natural home >> for it. So I think that's one thing. And then the other piece is one of the things that I of course loved in your in your original blog post was you had a really eloquent treatment around graph traversals and you talked through like graph embeddings and random walks and directed walks and all kinds of stuff that is just like a >> a treat that treatment was a treat for someone sitting here with a neo forj.
34:41 like yes yes right um um and like one of the examples I think you you had of kind of structural similarity was node tovec right which we all know is like a >> classic way of of of of doing graph embeddings um but there's other algos too like there's one called fast RP right which actually most of our I'm sure you're familiar with like mo most of our customers use it at scale right it's typically more performance at scale as equivalent accuracy right for some people nove is right for some people fast RP is going to be right in new forj it's one line of code to change between that >> and if it's on postgress there's a huge amount of more work that you have to do to like plug in like a different a different algo so I think those are some of the like the benefits of running on a native graph platform >> totally yeah no so actually there's a there's a slightly different way that we think about this which is um like the a graph is like the right mental model for what this thing ends up becoming over time. Um but the problem is that you can't actually observe the graph um until you actually start walking it. Um, and so the the kind of key idea I wanted to put forth in in that blog post was, you know, agents are informed walkers over this graph, which means that when you're when you give an agent a problem, it figures out how to stitch together through tool calls and and and more things, right? It's it's able to stitch together different pieces of context that allow it to both explore and exploit the the context that it's given in a problem directed way. Um, and so by virtue of doing that, it ends up walking over a graph that is otherwise not observable. And when many agents do this over time, they end up creating relationships, sometimes explicit, sometimes implicit, um, between all the different records that they're that they're uh, walking on top of. So just to make this a little bit more tangible, right? Um let's imagine like an incident comes in and the incident is now trying to be, you know, an agent is picking it up and saying, I'm going to go try to diagnose and recall that incident. And maybe it says, okay, you know, the alert says that this particular service is down.
36:58 Latency on this checkout endpoint is running a little bit slower than it used to. So I start with the service and then I need to go and say, okay, well, the service seems to be related to this area in the code. And this area in the code, right, is often something that, you know, Emil is is changing often, right? you may have changed recently. >> Damn it. I knew I was the culprit in all this. >> Um yeah, but like it's it's like okay, well now now I'm like you know there's some relationship that it found between the service and this area in the code.
37:26 Um and then you know Emil and you know once we actually make a change and we we suggest some sort of a fix you know this the scenarios and test cases that actually need to be you know verified before this thing actually goes back into production and the other customers that it affected. So this agent by virtue of actually just owning the escalation and remediation path ends up stitching together all of these different records and it's essentially observing that these there is a relationship between these things in a problem directed way and in our manifestation of it actually like it it isn't a graph um it's actually because it's unobservable right the the relationships that that that uh connect these distinct entities um end up changing based on the problem that you actually want to query against. Um and so therefore it becomes very difficult actually in kind of standard data models or there it's very becomes very difficult to actually define any data model um to to express this and so we actually end up falling back to um a lot of kind of AI motivated techniques and this is where the structural embeddings came in or um and it was also like graph transformers and stuff like that um or the way these labs are going right actually internalizing these these representations back into the model itself right using continuous learning and and all those kinds of things. Um and so in a world where you know the the models actually just like keep going the way that they are or continuous learning. Um maybe this is like a ridiculous this could be the smartest question I ever asked or the dumbest question I ever asked but like do databases even exist at that point? Um or like for for for context graphs, right? Like in this in this particular flavor of the world we're talking about like are databases relevant? Um >> well I I let me just first comment on kind of the thing you said and then let's talk about kind of the model view view on this.
39:18 >> Yeah >> I think that's super interesting right so a couple of thoughts on what you said there it seems like where we agree is that kind of the mental model for this being graphs that's that's a good kind of way to think about what these agents are doing. Y >> and then as you say they're kind of stitching together in a problem directed way. They're traversing these various systems figuring out like oh based on my input oh now I'm going to go over there and you know >> Exactly.
39:49 >> Right. >> Yeah. >> And and you don't think of that as a as a graph. >> Well relationship is a priority. >> Exactly. >> Therefore until I actually do the thing. >> Yeah. And then the output of this however is what is the decision trace >> right and that can be described as a graph though exactly as you said we may not know all the types ahead of time. >> Yeah. >> Right. And I think what this bumps into is this universal age-old you know uh kind of schism in the graph world of call it top down modeling versus bottom up modeling and you know I could spend hours and hours on this and I won't but suffice to say that you know we represent the property graph world that is kind of what Neo forj was was is based on and honestly invent present it.
40:50 >> Yeah. >> And then there's this other way of doing things called RDF. And in the ADF world, you tend to want to start with an ontology >> and describe that ahead of time. >> And in our world, because we grew up developer first, >> like the ontology was already expressed in C or I was going to say Rust. Rust didn't exist at the time, but whatever. in Ruby or Java or Python or something like this. It was encoded the ontology, the schema was encoded in classes in Python or Java or or something like this, right? And so there was no need to do that. We could just have this completely bottom up way that is schemafree. Right? Now over time we built it to be schema optional because we learned as when you would put it into production. It's actually helpful to be able to say if there's a node that has been labeled as a person, it it has to have a social security number, you know, or something like that because if the database doesn't enforce that integrity, then you have to imp reimplement it everywhere in iterative whatever in imperative code, right? Which is just that's just not a good a good model for it, right?
42:03 >> Yeah. And so I think what you're commenting on here is the drawback of a fixed top-down ontology >> and that the right approach I think your view is that the ontology should be built bottom up and discovered over time as a consequence of the agents walking all these different systems. >> Correct. Correct. >> And both of those things can be described in a property graph. Right. like you can use the top down ontology and you can use the bottom up inferred on ontology. my personal point of view and here's probably where we where we disagree is that I actually think being able to do both is the right approach and at least for the foreseeable future being able to guide the system a little bit like yes if you have to go all the way out and describe your entire company and how everything works in a top- down way with yourontologists that are that never works right the world moves too fast and they're always behind and you can never get a consistent so that never works Right. But I actually think the other extreme of complete bottom up too has drawbacks as well. And being able to say some of the things top down and have them being met and informed bottom up, I think is probably the right approach, you know, for for a while. But ultimately, we're we're agnostic to whichever of these approaches.
43:24 >> Yeah. Yeah. It makes total sense from from an enterprise perspective. I think I think it actually solves a lot of problems. Like I think one of the big problems ends up being like how do you secure them? Um because by definition these these context graphs end up you know spanning different relationships across different you know systems of record and different people and all those different things and you're trying to centralize this context which in theory sounds awesome. Um but the context you have access to is different than the context I have access to. Um but this agent somehow will will have access to all of it right and that that needs to be secured and and actually being able to enforce certain security kind of principles there I think is is important. There is like a counterpoint here though which is you can think of like the analogy of of um how we used to think about what AGI would look like 10 years ago versus what it like how it's actually you know progressing today. And I think like one of the biggest things for me was realizing that this next token prediction task um is actually the thing that could kind of scale into AGI over time. Um it's extremely fluid. It's extremely, you know, nondeterministic.
44:33 Um but it seems to encode the right kind of task uh to be able to you know model very very complex relationships. Um, and the reason the reason this analogy is like interesting to me is because like while like that that wasn't intuitive to me um 10 years ago, right? Like that next token prediction task being the right way to scale neural networks and what later became LLMs um to be able to handle very expressive and like reasoning intensive tasks. Um even actually after JBT came out uh like it wasn't obvious to me uh that just actually spending more tokens would necessarily give you exponentially better answers. Um and so the AI native way of thinking about this generally implies that the model can learn any pattern, right?
45:37 like there that that that these relationships can actually be implicitly learned um in in like transformer, you know, intermediate representation space in the LLM or in some sort of like external model as opposed to it actually being described. And the moment we as humans try to get our hands into it and try to define any sort of ontology, right? or maybe ontology is probably the wrong word, but like we try to define any sort of schema or record type or anything like that. Um that we end up constraining its expressiveness, right?
46:16 The same way that if you tried to fit if you tried to prescribe what each dimension of embedding actually was meant to represent, right? If you like went and tried to do that, embeddings wouldn't be embeddings. >> Yeah. Yeah. Um and so like that that that's kind of the counterpoint that I often struggle with where it's like you know if you if you throw a graph like a graph transformer on top of these like observed trajectories or um you you continually train an LLM and fine-tune it right for a particular domain these things tend to learn very interesting relationships that I could never even imagine modeling.
46:49 >> Yeah. >> Um in a graph database in a relational like any sort of a database like just does not make sense here. It's actually just it's an externalized form of continual learning. >> Yeah. >> Um and if that's the case, then does everything just get eaten by the model. >> Yeah. >> Is are we is is that approach are we like on some wrong side of the bitter lesson type thing going on here? >> Yeah. Like I I don't know. I don't know.
47:14 But that's that's kind of the counterpoint like because I I think I understand it from an enterprise side and this is something that we see every day and we've had to take you know opinionated positions on on how do you secure these things as you deploy agents into production when you have thousands of engineers or or things like that and I think you've seen versions of this play out across your customer base too. Um but at the pace that these things are moving.
47:34 >> Yeah. >> Right. The platform shift is happening and you know when we first started the company it was before chatbt people would laugh at us when we said you're going to deploy AI into production >> and now all of a sudden in production right like it's like no longer it's no longer even a conversation. Um and so like the way people are thinking about these things has changed so much and I'm wondering if this is one of those things that's going to change too. Yeah, it's I think it's a really really interesting observation and certainly like trying to predict speaking of next token prediction like the next year or two like I can't even predict the next weeks what's what anthropic is going to launch you know speak coming back to where we started with the stock market and you know that that kind of stuff um the neo forj like my point of view from like running the company and building the product that we do is to stay orthogonal to that like ultimately yeah if it's if it has an ontology or not we should be the best substrate to express kind of graph shaped data >> and I do believe that over the next time unit >> hopefully not just quarters more likely years at least having some kind of an enterprise world model which on some level is what we're all building up here. And you know, we've spent 101 15 years doing digital twin implementation, you know, for our for our customers, right? And then you add kind of aic memory and the decision traces >> to that and you start getting a pretty robust kind of world model. So maybe not quite the the physics-based one or the Minecraft playing one of the fail and the only coons of the world but like the equivalent I think filling an equivalent function in the enterprise right I do feel like that is going to be really valuable to pair up with these models but I think you're exactly right like over time >> who knows like where where this where this goes and may maybe maybe our entire layer gets eaten by whatever claw pod 67 or something like that, right?
49:50 >> Fair enough. Fair enough. Well, what's uh what's next for you guys at Neo Forj? >> Yeah. So, plenty of things, right? Like um one of the areas that I think is underexplored right now is the interplay between call it your classic kind of rag corpus and your domain data and memory, >> right? And um I feel like these systems are growing up in in parallel right now. And and I think here we are on a call.
50:22 Let's say that you and I are co-workers and we kicked off a a project and we have a code name for that called Project Trinity, right? Or something like this, right? Then we also have Google Docs called Project Trinity. And then maybe there's some kind of an agentic logger kind of a thing, a conversation log of this. We talk about project trinity. When you do your agentic retrieval, there should be a project trinity concept and it should be able to link to the convers like the agentic memory piece and whatever is coming out of your rag corpus that should be interlin it feels like right and I haven't seen anyone doing that in a good way and you know we've talked about it in different ways one way is like short-term memory versus long-term memory and you know that kind of thing right >> so that's certainly an area that we're um really interested in and exploring a lot.
51:20 >> Um >> yeah, >> another piece is and you know we we talked about this before, right, which is your first you know as a 12-year-old project around kind of named entity recognition and entity resolution and like on some level out of a bunch of unstructured mostly text, right? How do you actually extract real entities? So we have this conversation and we talk about William Gates, right? Who works at Microsoft, right? Okay. Well, we need to be able to extract that there's a William Gate. That's a person and Microsoft is an organization and connect the two >> and then we also call him Bill Gates and then we need to be be able to do the entity resolution of those are actually the same individual that we're referring to, >> right? And as I think in these four quadrants right that on all the different data sources more and more of this is going to be unstructured data of course right and then having that as a first class primitive like both named name named entity recognition and entity resolution ner and er I think is crucial >> the outcome here is ultimately knowledge based construction is that right like based on understand >> yes a consistent one across the the four quadrants Right. So that we know that we're talking about the same same concept, the same entity, the same individual, the same company, >> you know, that kind of thing.
52:44 >> Very cool. Very cool. And and I I I genuinely resonate with the the term you used world model earlier. >> Yeah. >> In fact, actually like the the what we what we call the kind of core piece of technology that we built is an engineering world model. Um and that's exactly right. Right. I think a a context graph that is well tuned allows you to model reality in a very expressive way and you know the the threshold for greatness there uh for us looks like can you ask what if right can you predict uh in a simulated way um what would happen under certain conditions um and that's that's >> before you experience the what if of that terrible engineer called Emil who shut down the system in production or whatever the example was from before like before you actually experience it in production you want to be able to simulate it in your world model. Yes, that makes a ton of sense to me.
53:40 >> Exactly. >> Um all right so so as as as we kind of wrap up here um I wanted to kind of like zoom out a little bit. I mean you've you've been a founder for how many years now? 20 something. >> Yeah, 20 years something like that. >> Not quite but 20 years. Yeah. >> Um no I mean it's it it takes incredible amount of kind of grit and passion. um to to stay at to like really stay at something for that long and you know as as a really early in life user of Neo forj uh first of all thank you right for everything you've built um but I mean as as you kind of think about the platform shift that we're that we're in right now you think about you know other founders building in the space um you know what advice do you have uh for for people building today?
54:28 Yeah, I think first of all like a platform shift is a massive opportunity, massive opportunity for disruptors, right? Like if you look at the last one when we shift in like the the shift to to to cloud and mobile that created 40 to 50 companies in the enterprise space, there's like consumer stuff too, but I tend to think of kind of infrastructure B2B enterprisey type stuff. There's 40 to 50 companies that were created that reached a billion of ARR or more right in the last platform shift.
54:58 >> I think this platform shift coming back kind of full circle is even bigger because it's not just a technology shift but it's literally an industrial revolution. So I think there is hundreds of companies that were founded in the last few years or haven't yet been founded that will grow up to be a billion dollars of revenue not unicorn billion dollar valuation billion dollar revenue type companies happening right now >> and whenever I give like do whatever early stage startup type talks right I always say that my first slide is always ignore most advice including probably all of the stuff you see in my presentation because that's I've I've I've always sought out advice. I love hearing advice just to inform myself but then I ignore mo most of it right and everything is so situational right and my early stage was 15 20 years ago >> like how much of that applies and it's an end of one it's only my experience right >> having said that I think the one thing >> that you cannot overindex on the one thing is customer empathy like having a deep deep deep understanding of the crucial personas for you like what do they where do they get their information? What do they do first thing in the morning when they wake up, right?
56:18 What what do they think of as their friends, their frenemies, their enemies? What are their key challenges? Like all of that have like a super nuanced understanding, all the callers of that that persona, th those real life individuals, right? That is the I think the only thing that you can't spend too much time on as a as a >> Yeah. Love that. Love that. You know, I think uh I often think about like the early days of Player Zero and and I think in the grand scheme of things, we're we're still quite young. Um yeah, our customers basically told us what to build.
56:52 >> Yeah. >> Um and it was just Yeah, the the fact that they would call me at, you know, 2 a.m. in the morning that, hey, I have this idea and you know, I was using your product and like those are the things that actually move the company forward. Um I I completely agree. >> Um and yeah, I hope that never ends. And and then I think like the other piece and you kind of queued it up that way too, but I guess it's just the persistence of it all and the grit of it all like because there's so many ups and there's so many downs and you think you're the like the the absolute top of the world. Like you're the hero and you're the zero seven eight times per day, right? Like so just having the like the the staying power to be kind of with it, right? I think would be like the the other piece, right? Now, how much of that is can you listen to someone saying that versus an innate trait? Like I don't I don't know. But like knowing what what drives you like Jeff Bezos has this like the dichotomy of customer obsessed versus competitively obsessed companies, right? Like Oracle is a classic example of the the latter, right? Like you know he Larry Ellison wakes up every day like thinking about which company to kill like which c like competitor to kill kind of thing, right?
58:02 >> Yeah. I ultimately, much like I think sounds like you do, I I'm obsessed with our customers. What gives me energy is there's 25 projects right now trying to find the cure for cancer using Neo Forj, right? And 99% of all flight tickets in the world are calculated with Neo Forj. NASA says that thanks to Neo Forj, humanity is going to get to Mars two years earlier, right? Like that's the stuff that really kind of inspires me and gives me gives me energy and like kind of that that staying power but just like knowing that like what is it that truly kind of drives you? Yeah. I think would be like the other piece probably >> and then and then uh like kind of other macro question here is um I mean we'll do another we'll do another version of this in a year uh together but in a year do you have any kind of hot takes or predictions u for what you think is going to play out in the next 12 months?
58:56 >> Well, I'm not going to try to predict the stock market because like I'm not qualified in >> going well yeah >> in any way. But I do think things will be a little bit more settled down. I think the volatility is just exceptional right now, right? And and I think a lot of this is just a massive uncertainty. It does feel like we're at the absolute elbow of the curve right now of everything changing at at the same time.
59:22 Right. And so I think may maybe some of that right in terms of the the technology it is the two areas that that I mentioned that that were and unsurprising that that we are really keen and are exploring right now which is some kind of connectivity between agentic memory and rag right like on a on a deeper level than just there are some documents there's some text over here but you know right some like how tied together and that's of course intimately connected to the ER and NE topics that we that we uh discussed before.
60:00 >> I also would not be surprised if we I think well let me not say it in the negative. I think we're gonna see a lot of people go up the stack like data platforms that add applications which may just be agents running right you know but something solving direct business problems because like the data platforms tend to solve technology problems for people who build things for end users and I think some of the data platforms are going to go up the stack and start solving problems directly for uh for for end users right and I think that's a year from now. I think that's going to happen. And I'm sure I'm going to be super humiliated when a year from now we're do this again and you play this recording and none of it ended up happening. But but those would be my >> like on the spot guesses a year from now.
60:50 >> Yeah, totally. I mean look like I think I think given the the customers you get to work with, the customers we get to work with, I think we get very different set of context, right? Um and and so that's why these conversations are interesting. >> Yeah. But uh yeah man without further ado I I I really appreciate you making the time for this. This is like a >> fun conversation. We we should definitely do it again sometime. Um and we are going to do a version of this again uh in in a few weeks. Um talk that. Yeah.
61:17 So what we're doing in a in a few weeks is that we are doing a panel on context graphs for nodes AI and nodes has been our online conference for many many years actually before the pandemic even and it it's Neoj's online developer expo and summit or something contrived like that just to be nodes because of course we love nodes right and now we're doing an completely AI centric one >> um and So we you know hopefully we have a chance to to to link to it and the keynote for that is going to be a panel discussion around context graphs including both of us us truly other people as well. Yeah be a lot of fun.
Summary
- Emil's journey led to the creation of Neo4j due to frustrations with relational databases in handling complex data.
- The concept of context graphs is introduced as a way to capture decision traces and institutional memory within organizations.
- Emil highlights the importance of understanding customer needs and the nuances of their workflows for product development.
- The conversation touches on the interplay between operational databases, data warehouses, agentic memory, and context graphs.
- There is a discussion about the potential for AI to change the shape of work and the importance of securing access to context across different systems.
- Emil predicts that data platforms will evolve to solve more direct business problems, integrating AI and agent capabilities.
- The importance of both top-down and bottom-up approaches in data modeling is emphasized, particularly in the context of evolving organizational needs.
- The upcoming Nodes AI conference will feature discussions on context graphs and their applications in AI.