theCUBE + NYSE Wired: Mixture of Experts Series | Prashant Bhuyan, Accrete

Prashant Bhuyan, Accrete

In this theCUBE + NYSE Wired: Mixture of Experts segment from the New York Stock Exchange, theCUBE’s John Furrier sits down with Raj Verma, CEO of SingleStore, to unpack how the intersection of technology and finance is shaping enterprise strategy. Verma shares why SingleStore is “on course” for the public markets, reflects on brand-building through the company’s partnership with golf Hall of Famer Padraig Harrington and connects that ethos to how SingleStore helps organizations fix struggling data “swings.” The discussion zeroes in on what’s next as Wall Street watches the AI infrastructure buildout: after chips and systems, the software and data layers set the pace for value creation. Verma outlines why enterprises must modernize “brown” data estates into “green” ones to safely bring corporate context, governance and compliance into LLM workflows via RAG – and why commoditized data-at-rest puts the advantage at the query layer that unifies data in motion with data at rest. He predicts agentic AI will gain reasoning capabilities in roughly 18 months, cites industry indicators like Google reporting ~25% of its software now built by AI and argues that high switching costs will give way to disruption as buyers reassess legacy vendors. The conversation closes with concrete momentum: ~33% YoY growth, ARR in the ~$135M range, gross dollar retention ~98%, cloud NDR ~130, ~50% of business now in the cloud, landing ~3 new customers per day, a path to cash-flow breakeven in the next two quarters and a teaser for AI-related announcements in the next two months. Listeners will find notable stats, real-world use cases and forward-looking views on how databases power reliable AI at enterprise scale.

Share this session

Clips
More from theCUBE + NYSE Wired: Mixture of Experts Series

Prashant Bhuyan

Founder & CEO

Accrete

play_circle_outline Accrete AI's Cognitive Infrastructure: A Digital Brain Unifying Tacit Knowledge to Avert the Agent Tower of Babel

play_circle_outline RAG/search is insufficient; agents need global reasoning from a single system of record

play_circle_outline Continuous learning/classification: agents observe interactions to capture tacit knowledge without explicit annotation

play_circle_outline Labor impacts: knowledge loss risks disruption; top experts become more valuable, others at risk

play_circle_outline Technical stack: multimodal ingestion, proprietary knowledge functions, multi-agent framework integrating LLMs

Info
Transcript

Prashant Bhuyan, Accrete

Prashant Bhuyan

Founder & CEO Accrete

In this interview from theCUBE + NYSE Wired: Mixture of Experts, Prashant Bhuyan, chief executive officer and co-founder of Accrete AI, joins theCUBE + NYSE Wired's Gemma Allen to discuss why cognitive infrastructure — not retrieval-augmented generation — is the foundation autonomous AI agents need to make complex decisions. Bhuyan explains how scaling agents across an organization creates a "Tower of Babel" problem, with each agent operating from a fragmented version of reality. Accrete AI's knowledge engine solves this by building an autonomous graph that e... Read more

explore Keep Exploring

What does the company do? add

What problem is your company trying to solve about context, and how do you see AI/intelligence evolving organizational decision-making? add

Does "classification" here mean continuous learning — and how do knowledge engines and agents continuously build and refine contextual understanding, capture tacit knowledge, and serve as a dynamic organizational memory (for example, replacing manual CRM entry through conversational interaction)? add

What are the risks to organizations from tacit-knowledge loss and fragmented information, and how could AI-driven cognitive substrates or autonomous knowledge-graphs help capture judgment, improve decision-making, and reshape the labor force and competitive landscape? add

How does the described knowledge-engine platform solve the problem of context? add

bolt Powered by CUBE AI

Prashant Bhuyan, Accrete

search

Gemma Allen

>> I'm Gemma Allen coming to you from theCUBE Studio here at the New York Stock Exchange. This is Mixture of Experts, one of our programs with NYSC Wired. And joining me now is Prashant Bhuyan, CEO and Co-Founder of Accrete AI. Welcome, Prashant.

Prashant Bhuyan

>> Thank you.

Gemma Allen

>> So Prashant, I was getting ready for work this morning. I asked Perplexity to describe what your company is in 30 seconds. My eight-year old was listening in the kitchen. She said, "Oh mom, it's like a mega computer brain." How accurate is my eight-year-old? Start there.

Prashant Bhuyan

>> It's pretty accurate in a sense. The way I like to describe the company is that we're building cognitive infrastructure for autonomous decision systems. So what does that mean? Inside of a complex organization, you have all of this context that's scattered across the organization. Fragmented systems, fragmented data, tacit knowledge that resides in the heads of your best experts. And when you think about an agent today, the way it's commonly defined, you're putting a large language model on your data and you're going to let it go take action. But the truth is, if you scaled that up in an organization, you had a thousand agents taking action, the question is you run into this Tower of Babel problem where they can't really work together because they're operating off of different realities. What we do is, we build this cognitive infrastructure. It's an autonomous graph that sits on top of all of these fragmented systems, encodes your tacit knowledge, your decisions, and creates one ground truth, one system of record that enables a multitude of agents to operate and have global reasoning, so you can offset complex decision making to the agents. So in a sense, it's a digital brain for teams of agents so they can collaborate, make complex decisions that really you can't, because the world is getting so complex and you can't reason that fast.

Gemma Allen

>> So I guess in the world that most users are familiar with where you prompt an LLM, it's the kind of RAG plus tooling model, it's telling you a question you asked, in your world, you're helping users understand questions they should be asking. It's getting ahead of a curve from the perspective of context.

Prashant Bhuyan

>> Context is really the key problem that we've been working on ever since starting the company in 2017, and for many years before that. And essentially when you think about the way we work today, we hire people inside of an organization that have some sort of experience and education in a particular domain. They search through a lot of information, they analyze it, and then they get together to try to form a decision. Our thesis is that complexity is rising at such a pace, it's outpacing biological reasoning capacity. So that reactive search process is going to lead to bad decisions and bad outcomes. The problem is that with a search system, you have to always know what to search for, and that's not going to be the case as the world gets more and more complex. So we think that intelligence is about building systems smart enough to tell you about the problems that you don't even know you have. Smart enough to proactively go out and solve them and make the decisions based on your tacit knowledge, your judgment, your values, your standards, contextual to your organization, such that those systems are operating to drive your objectives forward. And so we think that there's going to be this massive evolution from companies that are built around reactive search and analysis systems, to proactive, predictive, autonomous decision systems. And we think that RAG and search is really the wrong data model for agents.

Gemma Allen

>> Wow. Okay. So you have some big, big workloads with the US government. You also have some very big contracts with commercial players. Talk me through the typical buyer persona that you're seeing right now.

Prashant Bhuyan

>> Well, we started the company and we have this platform, this knowledge engine platform that its first application really was in defense and intelligence, where our customers are dealing with high stakes decisions in complex information environments. An example would be customers that are looking to thwart foreign influence in complex vendor supply chains. Typically, this takes teams of very smart and experienced people months to years to map together complex supply chains, and to understand the degree to which these different entities are under foreign influence. When a bad actor infiltrates the supply chain, it can have consequences measured in the billions of dollars to the US economy, and then even further consequences to national security. So these are the types of problems that customers want to become more proactive in and not reactive. So our knowledge engine essentially is able to understand the context, model influence, map together entities and relationships, figure out where the vulnerabilities are, write intelligence reports in the matter of seconds, compressing that timeline down to minutes rather than months and years, and actually drive significant outcomes that are making a real impact on national security. That's an example on the defense and intelligence side.

Gemma Allen

>> So we're hearing a lot right now about this idea of autonomous systems and the risk that they potentially bring, especially to folks who are hearing this, I guess anecdotally, who's, maybe world might not necessarily collide with the tech space on a day-to-day basis. It certainly seems as though there's a lot of fear at the idea of agentic autonomy. There's also, I guess, a very interesting competitive race happening globally in this space broadly. What are your thoughts on the human-in-the-loop element to this?

Prashant Bhuyan

>> My take on the human-in-the-loop, it's a little bit different the way I think about it conceptually. Essentially, the way I think about human-in-the-loop is that, are these systems grounded in the tacit knowledge, the experience, the judgment of humans? And if so, can they reason from that ground truth, explain their reasoning, and take action that's aligned with the objectives of those humans? What's very scary is the idea of autonomous systems that are taking action based on ungrounded LLMs, hooking OpenClaw up to ChatGPT or Anthropic to go and perform tasks in a way that it's not grounded or aligned. That is very scary. So if you think about the large language models as an interface layer, but the brain that underlies that LLM is actually rooted in your tacit knowledge and your judgment and your values, it can lead to tremendous breakthroughs in every field of endeavor, and massive productivity gains and competitive advantages at both the sovereign level and the enterprise level. But if you have a lot of agents running around taking action on hallucinations, for example, it could become a very scary world where these agents are operating off of fragmented realities with objectives that aren't fully defined.

Gemma Allen

>> So I want to ask you about classification, because it's a space where there seems to be a lot of money going into this startup landscape and companies that are doing some interesting things in this space. When I looked at your company this morning, I thought, where does the role of classification and companies like Toptal and these folks who are hiring domain experts to help them really build out the brain play in this world, in the world of Accrete? It's iterative, right? But it's iterative from the perspective of an AI contextualizing at scale. Do you see a world where there is continued, I guess human touchpoints, human interface as these brains build?

Prashant Bhuyan

>> So with classification, I'm taking that to mean continuous learning and how these knowledge engines continuously build on themselves and refine their contextual-

Gemma Allen

>> Exactly....

Prashant Bhuyan

>> understanding of the world. So you can think of the knowledge engine as a kind of a dynamic context, memory, and perception layer, and the agents are an interface that integrate into this universal system of record where they can now have global reasoning across a complex organization. The agents really have two purposes. One is to go out and use this knowledge and apply the tacit knowledge of the knowledge engine to accomplish tasks and make decisions. But the second is really to further understand the people and collect more tacit knowledge. So the agents learn through observation and through natural interaction. It is not a question of explicitly giving your context to the machine. That's an impossible task because all of the value in knowledge, most of it, is tacit, which means it's very difficult, if not impossible, to express tacit knowledge formulaically into a system of rules. I'll give you an example. Imagine you're a salesperson. The most painful thing, it seems, is to take a complex deal and enter it into a couple of cells in a CRM system. That is an abstraction of, really, the ground truth and the complexity of the deal they're trying to close. Rather, it's much easier for that salesperson to interact with an agent, have a conversation either through text or voice, and the agent basically takes that information, what's relevant, puts it back into the brain of the organization, and now you have a much more nuanced understanding of what's happening with that deal without any sort of explicit human-in-the-loop annotation. Let's put it that way.

Gemma Allen

>> I've been listening to Jensen a bit lately. It seems everybody is, right? And he talks a lot about task versus mission. He gets a lot of questions put at him around the AI bubble, but also the world of AI and the labor impacts, etc, which it's hard to argue that they are not becoming significant and there's a lot of fear out there. He talks about a world where we're mission-led, where we are essentially automating some of the most onerous tasks of the day and looking at it from the perspective of mission-led activity. And tacit knowledge is hugely instrumental in ensuring that that happens in the most effective and also fluid way. What are your thoughts on the impacts of human labor in this entire agentic era that we're entering into?

Prashant Bhuyan

>> Right now, I think that the explosion in complexity and knowledge loss that you see in organizations, you have an expert that got moved to a different division or left the company, all that knowledge walks out the door, these are existential risks for organizations. Organizations that are built around this reactive search and analysis system that don't encode their tacit knowledge, that don't understand why decisions are made ... Most decisions are made at the end of the day based on intuition. If you don't capture all of that, I think the organization is at existential risk, and that will lead to disruption in the labor force. And I think unfortunately, that will probably happen at an accelerated rate. However, those that can evolve to building some kind of cognitive substrate for agents, a shared reality for these agents to operate, to encode this tacit knowledge, to understand why decisions are made so they can model better decisions going forward with real world outcomes, they'll be able to harness the complexity to make breakthroughs in all the different fields that will lead to incredible productivity and competitive advantages. Core to that are going to be the people that you trust the most inside of your organization, the judgment, the expertise. Maybe you don't need a lot of them, but the most valuable people will become incredibly valuable when you think about it. And I think that ultimately most software that sits on top of these fragmented databases will eventually kind of give way to one universal system of record, which I think will take the form of an autonomous graph of some sort, a knowledge engine, which is dynamic, etc. But ultimately, the people who are fundamentally tied to searching those systems of record to try to make sense of it, etc, but they're not actually part of the ground truth, they would be at risk of disruption, in my opinion.

Gemma Allen

>> Those systems of record and that cognitive substrate is also an interesting space for SaaS tools, right? Because if you think about in the agentic world, the value in having an agent act on your behalf and engage with Workday and engage with Ramp and engage with all of these different tools that are so prevalent across enterprise, is only realized when these tools talk to each other. And there's a risk in these tools talking to each other from the perspective of defensibility in a SaaS model that's becoming exceptionally squeezed. What are you seeing there? What are your thoughts?

Prashant Bhuyan

>> Well, it's interesting because we've also deployed our knowledge engines at our own company. We have hundreds of employees. They use different software. And it's interesting because when you talk to the agent that basically sits on top of a myopic database that has the software layer on top of it, one of these service providers, you realize it's kind of an abstraction to that ground truth. And ultimately, as you build enough organizational context from the knowledge engine, you can learn the workflows and then you don't necessarily need the software. So it's less about orchestrating the agents that sit on top of these software systems. It's more about getting to the underlying ground truth that you can essentially reason from and build agents on top of. So what we've been doing internally is we've been canceling certain software subscriptions. Not all of them and not all at the same pace. Some will take longer. But our goal is to create a universal system of record that really is ours, and there's no reason ... We have an amazing chief information security officer, former NSA guy. He uses multiple tools in his stack. You can think of Wiz and Datadog and KnowBe4 and all these different software. He's like, "The vulnerabilities sit in between the software." And now that he's got global reasoning with the knowledge engine, he's built agents. He's built one agent. He's built agents for autonomous pen testing and things like that, which are fairly simple use cases. But he's built another agent that will constantly attack our code, identify vulnerabilities in code, remediate the code, submit merge requests, write reports, align stakeholders. He sent me an email the other day saying ... Sent my agent an email the other day with his agent, and basically it said, "Hey, I just saved $400,000 by canceling these software subscriptions, and look at these vulnerabilities. I got this agent to do this every single day. We don't have to ..."

Gemma Allen

>> Give that agent a bonus, right?

Prashant Bhuyan

>> Well, that's what he said. And then he went to HR and he said he didn't need to hire X number of people. And so that is really the dynamic that we're experiencing.

Gemma Allen

>> So interesting. Talk to me about the tech. We're always interested in understanding the technical build-out here on theCUBE. Are you building everything proprietary? Are you using multi-models? Talk me through it a little bit.

Prashant Bhuyan

>> Yeah, we've spent the better part of a decade working on this underlying problem of context. And so what we've done is we built this platform, knowledge engine platform, and it's a dynamic context, memory, and perception layer. So there's really three buckets to it. The first is it ingests data across multiple modalities, audio, images, video, text, etc. Sits on top of existing tools. It sits on custom tools directly on data. But the idea is that it's constantly streaming in information. The second layer is what we call knowledge functions, and these are proprietary models that essentially perform higher level reasoning on that information. Namely, they're able to, based on context, infer relationships in the data. And that really is the key thing that drives our memory, because when you integrate it with the third layer, which are the multi-agent framework, that's where we integrate with LLMs, we don't wrap around LLMs, we integrate the LLMs into the knowledge engine. Because then that agentic multi-agent collaboration framework can reason across the relationships discovered by the models. And that's what gives our agents the global reasoning across shared context, and that is the key to creating a shared reality for the agents to collaborate. And that really is how we solve the problem that we call Tower of Babel.

Gemma Allen

>> Wow. So 2017, 2026, that's a nine-year span, but by God, has a lot happened in nine years. From the perspective of ongoing releases, features, staying abreast of what's happening in the frontier space and building that into your own product stack, talk me through how you handle that. It must be an onerous journey day to day.

Prashant Bhuyan

>> Well, we've had the same hypothesis for the last 15 years, even before launching the company, which is ... I used to build algorithmic trading systems, and they would follow rules to make decisions to buy and sell stocks. But I figured, how do we get the machine to understand why it's buying and selling stocks? Because if you could figure that out, you could actually beat someone else looking at the same information and capture an inefficiency or something to generate alpha. And so that scope has expanded, but the hypothesis is still the same, and that is, how do you solve this problem of context at scale and get it into the machine? So we've always been focused on the same problem, which is building autonomous graphs to capture how people use intuition to make decisions, so the machines can model how to make better decisions going forward in a way that you can trust. And as frontier models started to proliferate, we realized there was a way now for humans to interact with that context through the agents, and there was a way to collect more tacit knowledge and build up our context. So these models, from our perspective are commoditizing, but what doesn't get commoditized is that tacit knowledge. It's not just about organizing data inside of an organization, but it's also about organizing the tacit knowledge and the experience, the intuition, the judgment. And so we think that that really is the bridge between Generative AI and, dare I say, super intelligence.

Gemma Allen

>> Wow.

Prashant Bhuyan

>> Yeah.

Gemma Allen

>> Okay. So Prashant, close us out. Talk to us a little bit. Where are you guys at from the perspective of the investment cycle and what's on the roadmap for the next six to 12 months?

Prashant Bhuyan

>> Yeah. We largely bootstrap the company. To date, we've been growing kind of organically.

Gemma Allen

>> Wow.

Prashant Bhuyan

>> We have aggressive plans moving forward. And so we're excited about our growth plans. We are looking to basically expand in defense and intelligence, and then on the enterprise side simultaneously. The fact is that the demand for what we're doing is actually greater than we expected, and so I think that we're very excited about the future. I mean, ultimately on the roadmap, we think that our knowledge engines today are powering agents. We think tomorrow that cognitive substrate could be instrumental in organizing robots and having them manipulate three-dimensional space in the industrial world. And so there's a lot that we're thinking about going forward.

Gemma Allen

>> Wow. Compounding superhumans essentially, right? The super brain. Incredible.

Prashant Bhuyan

>> Exactly.

Gemma Allen

>> Well, listen, Prashant, wonderful to have you on theCUBE. Thanks so much for joining us.

Prashant Bhuyan

>> Thank you.

Gemma Allen

>> Fascinating conversation. I wish you guys all the best.

Prashant Bhuyan

>> Thank you. Appreciate it. Thank you.

Gemma Allen

>> I'm Gemma Allen coming to you from theCUBE Studio here at the New York Stock Exchange. This is Mixture of Experts, one of our programs with NYSC Wired. Thanks so much for watching.