# Why Enterprise AI Fails Without a Context Engine - with Eran Yahav of Tabnine

- **Show:** The AI in Business Podcast
- **Published:** 25 Mar 2026
- **Source:** https://podcast.emerj.com/why-enterprise-ai-fails-without-a-context-engine-with-eran-yahav-of-tabnine
- **Transcript:** https://podcasts.opennash.com/shows/ai-in-business/why-enterprise-ai-fails-without-a-context-engine-with-eran-yahav-of-tabnine-ec9467/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Eran Yahav of Tabnine argues that AI agents fail on complex enterprise work, around 80% of the time, because they don't know the organization. They behave like new hires who never got onboarded. His fix is a context engine: a layer that maps a company's code, documents and systems so agents can navigate them instead of guessing.

The main tension is power versus direction. Agents are fast and can generate code faster than humans can review it, yet in old, undocumented systems they wander, burn tokens and produce wrong answers. Yahav says a pre-built map improves accuracy and cost, but it must be detailed enough to help and light enough to maintain, and it must stay inside the company's security perimeter.

## Key insights

1. **Agents fail because they are un-onboarded newcomers** Yahav says human developers at banks take months, sometimes six to nine, to become productive in legacy systems. AI agents get no such onboarding, which he links to roughly 80% failure on complex tasks.
2. **A pre-built map beats agents exploring alone** Context agents continually crawl the organization and record dependencies, so a task agent can skip irrelevant areas. In a vulnerability example, the agent goes straight to the three relevant areas instead of searching everything.
3. **Starting with code, Jira and Confluence pays off** Yahav recommends giving the engine access only to code and closely adjacent sources such as task tracking and design documents. He says this can double success rates and cut token use by up to 80%.
4. **Legacy systems hide many wrong answers** Yahav's example is a company with fourteen ways to fetch employee details, of which only one is correct. An agent that finds the first one may use an outdated or deprecated route, wasting cost and time.
5. **The map must be coarse enough to maintain** A street-level map would take too much effort to keep current. Yahav says the goal is a map precise enough to help agents succeed but light on incremental maintenance.

## Questions this episode answers

### What is an enterprise context engine?

Per Eran Yahav of Tabnine, it is a layer that connects to a company's systems, builds a map of how they relate, and shares that map with AI agents. It works as persistent memory alongside the language model and the agent.

### Why do AI agents fail so often on big company projects?

Yahav says they fail around 80% of the time on complex tasks because they don't understand the organization. Like newly hired engineers, they lack knowledge of legacy code, business logic and internal practices.

### How can a context engine save money?

Agents with a map don't wander through the organization or retry wrong paths, which uses fewer tokens, the units AI models charge for. Yahav cites up to 80% improvement in efficiency and about double the success rate in some cases.

### Is it safe to give an AI my company's internal information?

Yahav says many customers run the context engine behind their own firewall or fully air-gapped, meaning disconnected from outside networks. He also argues that agents given proper organizational knowledge are easier to trust.

### Where should a company start with a context engine?

Yahav suggests starting with the code base and closely related tools like Jira task tracking and Confluence design documents. That is the lowest barrier to entry and already improves agent results.

## Facts

- **Eran Yahav** [0:03:19]: "Uh, some of the more complicated businesses will tell you six to nine months before the engineer really becomes, uh, productive."
- **Eran Yahav** [0:03:48]: "This is, uh, you are entering into this legacy system, potentially consisting of like millions, tens or hundreds of millions of lines of code"
- **Eran Yahav** [0:06:43]: "There's some component to that, but the third layer that is really important in enterprise is what you called the context layer, or what we call the context engine."
- **Eran Yahav** [0:06:43]: "They don't know anything about my particular business, my particular bank, my particular enterprise, right?"
- **Eran Yahav** [0:09:11]: "it does pre-computation and builds a map of the organization to understand the dependencies and relationships between components."
- **Eran Yahav** [0:09:11]: "These context agents, their job is to do the homework, their job is kind of like constantly crawl the organization"
- **Eran Yahav** [0:10:26]: "it pre-builds, call it the highway map of the organization for the agent."
- **Eran Yahav** [0:09:56]: "typically, if a vulnerability lands and you give it to a task agent with no context, it will have to start digging through the org"
- **Eran Yahav** [0:11:38]: "they would like the context engine to run inside their perimeter."
- **Eran Yahav** [0:15:59]: "giving context engine access to code and to, um, um, I would say something like a Jira task management or Confluence design documents"
- **Eran Yahav** [0:16:33]: "In some cases, we see two times on the success rate"
- **Eran Yahav** [0:16:33]: "In some cases, you you can see 80% improvement, right?"
- **Eran Yahav** [0:18:28]: "But in reality, there are fourteen different ways, uh, to get the employee details, and agent just bumped into one of them"
- **Eran Yahav** [0:21:24]: "So we're already kind of generating code faster than any human can review it"
- **Eran Yahav** [0:24:00]: "It's not unusual to see people spending thousands of dollars getting nowhere"
- **Eran Yahav** [0:23:09]: "you need to give agents kind of swim lanes in which they can operate without kind of interfering with one another"
- **Eran Yahav** [0:22:41]: "the agent itself is like this very powerful car. It's like a Ferrari, it can go really, really fast, but if it doesn't have a map"
- **Eran Yahav** [0:17:06]: "Context engine also provides you, maybe for the first time, full visibility of what is going on across different projects"
- **Eran Yahav** [0:14:41]: "we are focused on software engineering agents, so we are only looking at software engineering tasks at this point."

## Transcript

**[0:00:12] Host:** Welcome everyone to the Emerj AI in Business podcast. Today's guest is Eran Yahav, CTO and co-founder at Tabnine. Tabnine is an enterprise context engine that provides the infrastructure layer for AI systems to access organizational knowledge and reasoning. Eran joins us to discuss moving beyond AI pilots by using a context engine to reduce the failure rate of agents in complex tasks. Our conversation also covers boosting ROI by doubling agent success rates and cutting token costs. Mapping organizational knowledge to stop agents from driving in circles and deploying AI within secure perimeters to ensure enterprise grade trust.

**[0:00:48] Host:** Today's episode is sponsored by Tabnine. Before we begin, a quick note for our executive listeners. Emerj invites enterprise leaders who are driving meaningful AI initiatives to share what they're learning with a peer audience. If you're moving real projects forward and want to be part of that conversation, you can learn more at go dot Emerj dot com. Slash expert, that's G o dot e m e r j dot com. Slash e x p e r t.

**[0:01:11] Host:** Now the conversation with Eran. Eran, welcome to the AI in business podcast. How are you doing today?

**[0:01:18] Eran Yahav:** Doing great, thank you for having me.

**[0:01:20] Host:** Thank you for being here. So we're gonna head into it. Eran, everyone is always talking about AI in business, but it seems as though the honeymoons honeymoon phase is coming to an end. Uh, you've seen this from all angles. In your opinion, as enterprises try to move beyond AI pilots, what core business challenges and operational pain points are they currently running into?

**[0:01:43] Eran Yahav:** Yeah, first of all, I, I have to say the honeymoon period is not completely over. There's still a lot of kind of low-hanging fruit that you can pick, uh, by deploying A I in in the business. I think people are successfully doing that. Problems start when you are, uh, starting to go into the more complex tasks, the more, uh, high-value tasks, call them. And then what you're seeing is A I agents failing in around 80% of those cases.

**[0:02:11] Eran Yahav:** And you know, uh, the the the reason, the underlying reason, or the main reason, is that they just do not understand the organization well enough, right? They act as if they are, uh, say, newcomers to the organization, rather than an onboarded developer that has been with the organization, understand how things are done around here.

**[0:02:31] Host:** That is a it's. It's definitely something that we see, especially after pilot or demo versions of AI, is where the the AI is not evolving with the company, or it's not evolving in the same culture or in the same way that our companies need to evolve. What is the reason for that?

**[0:02:47] Eran Yahav:** Yeah, uh, again, I I think it's not just the, as I say, e evolution. I think the A I agents are really facing this critical challenge of not having the understanding that human engineers do. So when you go to a bank, for example, and you ask, uh, you know, how long does it take a human developer to be productive, to become productive, someone that was just hired into the bank? Um, typically, if you ask that question, they will tell you a few months.

**[0:03:19] Eran Yahav:** Uh, some of the more complicated businesses will tell you six to nine months before the engineer really becomes, uh, productive. And the reason is not that these people are not good, these people are great, right? The the issue is that, uh, there is an onboarding process. They need to understand, uh, the entire context in which they operate, they need to understand the organization, they need to understand the existing systems, they need to understand how existing systems are being maintained and manipulated.

**[0:03:48] Eran Yahav:** So this is a very different use case, uh, than what we call greenfield development, in which you are creating a system from scratch, right? This is completely brownfield. This is, uh, you are entering into this legacy system, potentially consisting of like millions, tens or hundreds of millions of lines of code that encode very complicated business logic that is not typically not documented anywhere beyond the the code itself, right? And so you really need to kind of ma make your way around that, navigate around that.

**[0:04:22] Eran Yahav:** That becomes the main challenge, and this is what human engineers are getting into when being onboarded, and we need the same onboarding process for the A I agents in order for them to become productive. That's kind of the the hypothesis of everything that we're doing.

**[0:04:37] Host:** So what we're saying is is that the the problem or the delay comes in where organizational knowledge is still kind of somewhere. It's buried into the company policies and documents, and they they don't. They're not quite there yet to understand exactly what the company is is about and what direction to go in. Am I getting that?

**[0:04:56] Eran Yahav:** Yeah, I think maybe again we we focus on software engineers or software engineering agents, and for software engineering agents, a lot of the decisions that they have to make, a lot of kind of the the micro decisions that they make when developing the software are based on very rich contextual information. Like, you know, what is really the business case, what is the feature that we're trying to build here, what is it really trying to serve in the business?

**[0:05:22] Eran Yahav:** How does it relate to the design of the entire system? Uh, what libraries can or cannot we use? Um, you know, what was the recent incidents that were around this code base where people were trying to maintain the same thing? All that kind of organizational knowledge, all that thing that are around the code, that are sitting in the head of the senior engineers. Um, this is what you need to somehow convey into the mind of the agent, right?

**[0:05:53] Eran Yahav:** Does that make sense?

**[0:05:54] Host:** It does so. I've, I've heard you refer before, um, I, I've heard you refer to context layering in A I. Is that what we're getting at? Is that what, what you mean when you say the A I has to start to think like that?

**[0:06:07] Eran Yahav:** Yeah, I I think abstractly, you can think about uh, agent universe, or the the agent kind of stack as having three components. One component is the L M, the large language model. Uh, these models are becoming more and more powerful, right, more and more capable, uh, but they are not aware of the context in which they're operating. Uh, in order for them to understand context, context has to be provided to them, right? They know what they have been trained on, and typically they have been trained on open source code, the entire internet, et cetera, et cetera.

**[0:06:43] Eran Yahav:** They don't know anything about my particular business, my particular bank, my particular enterprise, right? Uh, and so that's one component of the mix, the L M. The second component is the agent itself that definitely has a lot of kind of a U I U X uh as aspects to it. How does it interact with the human? How exactly does it, uh, invoke the L M, et cetera? So there's some component to that, but the third layer that is really important in enterprise is what you called the context layer, or what we call the context engine.

**[0:07:18] Eran Yahav:** And that is the component that is, um, connecting to all the enterprise system, digging deep into those systems, uh, aggregating and correlating insights across systems in order to, together with the agent and the L M, give you real, uh, organizational intelligence. So just to to recap, really, three components, the the language model, uh, the agent, and the context engine, all have to participate in order to get real organizational intelligence.

**[0:07:47] Host:** So our context layer will essentially become the brain of the AI operations. Is that correct?

**[0:07:53] Eran Yahav:** I I think it will become part of of of the of the brain. I think that the brain is the kind of the computation engine, which is the L M in a sense, but also call it the the memory or the persistent memory, or kind of the insights that you built over time, uh, from the organizational systems. So so, the way, maybe throw a few words on kind of how context actually works. It's not just connecting to to the systems, actually extracting valuable information from them, and you can think about it more as building a map for the organization of the organization for agents to navigate through.

**[0:08:36] Eran Yahav:** So what the context engine kind of does is define the universe in which agents operate.

**[0:08:45] Host:** From my understanding, the context layer will then also have to be able to reason, uh, think more than just basic things. So, with that reasoning taken into consideration, can you explain how an A I with context understanding will understand what the effect of any damage could be? If we think about blast radius, um, or anything that will change, how does a context layer do that in a better way than just a generic A I?

**[0:09:11] Eran Yahav:** Yeah, that, that's a, that's a, that's a great question. I think the the gist of the context engine is that it does pre-computation and builds a map of the organization to understand the dependencies and relationships between components. Such that what we call task agents do not have to reconstruct them. So think of the context engine as having its own internal, what we call context agents. These context agents, their job is to do the homework, their job is kind of like constantly crawl the organization, find the interesting, relevant information, build the dependencies, build a representation, build this map of the org, which is exhaustively correct.

**[0:09:56] Eran Yahav:** So they kind of see everything all the time, right? And then they expose this map to the agent, so to to a task agent. So, for example, if we go with what you mentioned, blast radius or some vulnerability lands, typically, if a vulnerability lands and you give it to a task agent with no context, it will have to start digging through the org to find what is the blast radius like. Hey, let me just scroll everything and try to find what this affects.

**[0:10:26] Eran Yahav:** But in a large enough organization, that agent stands no chance. It does not have not the visibility across all systems and doesn't have, like, the long horizon reasoning or kind of the the capacity to, uh, visit all of the of the organization code bases or other things to find what this vulnerability affects. What the context engine does, it pre-builds, call it the highway map of the organization for the agent. So the agent, the task agent, does not have to rediscover the map every time, it just looks at the map and say, look in the highway map.

**[0:11:02] Eran Yahav:** I already know that these five general areas are completely irrelevant, and now let me focus on the three areas of the system that are relevant for this vulnerability. And let me dig deeper into that. So that makes things more efficient, more accurate, and and also increases the success rate dramatically.

**[0:11:22] Host:** It makes sense, and it feels as if, though there will be a lot of trust given to the the context layers. So, with that being said, trust and things like data leakage and all the other challenges that we have, what does it take for A I to deliver meaningful, reliable value inside a complex enterprise?

**[0:11:38] Eran Yahav:** Yeah, I think there are two aspects to your question. So, first of all, if you look at the context engine, as I said, it really builds a deep map of the organization, uh, understanding relationships and correlating things across system. Which means that it has access and gotta have access to many of the most precious sources of information inside the organization, right? And typically, uh, for our customers, uh, that would mean that they would like the context engine to run inside their perimeter.

**[0:12:13] Eran Yahav:** So it means behind their firewall or even completely air gapped, because this is now becoming, uh, really, a map of the organization or containing very precious information across many of the S of the most again sensitive systems in the org. So many of our customers want to have that inside the perimeter, so that's one aspect of security or or trust. Uh, I think that the other aspect of security and trust is as we delegate more tasks to agents in order for us to trust what these agents are doing.

**[0:12:49] Eran Yahav:** Uh, we need to know that they know enough about what they're doing, right? So when we hire an engineer, uh, we hire them because they are great professionals and because they have common sense, and once we onboard them, we can trust their decisions on how to, you know, manipulate the system, uh, in order to get the same level of trust for agents. We need to give agents the same kind of onboarding, the same kind of organizational awareness in order for them to be successful and solve problems in ways that we can trust.

**[0:13:25] Eran Yahav:** With without information, it will be very hard, uh, for us to trust what agents are doing.

**[0:13:29] Host:** And it's a it's a big problem. Well, we have to trust them because we have so many regulated industries that they need to to make use of this. They need to make sure that there's no data leakage to them. It could lead to much bigger disasters than just leaking sensitive information. It can also be other people's information. Uh, you've said that the enterprise context engine makes this information available to any agent or tool. Uh, is the goal to make the company's internal knowledge basically the the entire head of operations?

**[0:13:58] Host:** What exactly do we see the future of every tool being used, and how do we keep that information safe?

**[0:14:05] Eran Yahav:** The the vision of the context engine is to be, maybe you said it earlier, the the kind of the the brain of the organization that captures information, um, that is similar to what a senior engineer would have in in their head, right? And so the idea is to provide a common data source or data plane and control plane for all the agents that are running around the organization. And make sure that they're all following the same map, the same guidelines, and to give the organization some control.

**[0:14:41] Eran Yahav:** O over what these agents are are doing. Uh, we are focused on software engineering agents, so we are only looking at software engineering tasks at this point. But the software engineering tasks are informed by systems that are not just code, right? By design document, by incident reports, uh, by uh, production information, telemetry from things that are existing in production. So you are getting visibility across the entire software development life cycle, from planning requirements, um, implementation, documentation, test generation, et cetera, et cetera, up to deployment, visible to the context engine.

**[0:15:25] Eran Yahav:** So future tasks, um, have this rich information to them as they're being implemented. Does that make sense?

**[0:15:34] Host:** It does, it does, but what I'm thinking about also is we have a lot of decision makers in enterprises, uh, C E O s that might not be into the technical part of it. And they might also still be a bit uncomfortable by giving some A I model their entire playbook or their entire internal handbook of what the organization is about. Where should they start to get comfortable with the idea of integrating that context engine?

**[0:15:59] Eran Yahav:** Yeah, I think the the kind of the lowest barrier to to entry, I think, is like, uh, giving context engine access to code and to, um, um, I would say something like a Jira task management or Confluence design documents, something like that. And closing the loop, uh, around things that are really adjacent, closely adjacent to code. And you will see that if you give agents that level of contextual information, uh, you will see success rate going significantly higher.

**[0:16:33] Eran Yahav:** In some cases, we see two times on the success rate, and you will see also the the efficiency, uh, on in token consumption, speed and otherwise being significantly improved. In some cases, you you can see 80% improvement, right? So by giving these agents more information from sources that are first of all the code base, but then the things that are adjacent to code base, you're making them more successful because they are better informed on what is it that they're trying to do and how they're trying to do it.

**[0:17:06] Eran Yahav:** And how these things, was maybe potentially have been done in in the past, right? So that's what we are trying to to capture there, maybe to to your, somewhat to to your point, not C E O level, but maybe architect level, um, for architects or team lead, or V P of engineering. Context engine also provides you, maybe for the first time, full visibility of what is going on across different projects, across different silos in the organization.

**[0:17:35] Eran Yahav:** Because again, context engine kind of integrates and correlates information across the different sources. And now, as an architect or as a V P of Eng, I can sit back and kind of start asking cross-cutting question across my project, across my organization. So the context engine is primarily, uh, designed to feed agents or to give map for the agents, uh, but the same kind of information is definitely, um, usable or useful to humans that are trying to navigate organization as well, right?

**[0:18:08] Eran Yahav:** So the context engine is valuable to agents, but also valuable to humans.

**[0:18:13] Host:** And that sounds like a possible cost saving strategy as well, because you'll everything will go faster. You mentioned speed, cause I assume that there will be less cleaning up to do for our engineers. They won't have to go and recheck everything sixty times.

**[0:18:28] Eran Yahav:** Yeah, and not only, uh, that the engineers would not have to recheck things, uh, sixty times. The agent itself will not get lost in the organization, right? One of the problems that I can send agent say and tell agent, Hey, um, how do I get in? How do I get employee details in my organization? And agent will come back and say, here's like one microservice that gives you employee details. But in reality, there are fourteen different ways, uh, to get the employee details, and agent just bumped into one of them, uh, which may be just outdated, the wrong one, whatever, you know, think legacy.

**[0:19:03] Eran Yahav:** So in in legacy systems, there rarely just one way to do thing, and rarely the first thing that you bump into is the right way to do something. So you need to put some controls around it and need, like the agents, to have, uh, exhaustively correct view of the organization. So the agents say, like, hang on, I already kind of looked at that problem of like, how many ways are there to get employee detail? Like, I did my homework, there are actually fourteen different ways, and the only correct one is number seven.

**[0:19:33] Eran Yahav:** The other are either outdated or being deprecated, or et cetera, et cetera, or just more expensive and efficient, right? So by having this pre-computed view of the organization for certain things, not for everything, uh, you can save a lot of confusion from the agent, which translates to also, uh, cost, time and success rate.

**[0:19:57] Host:** And and again, cost saving sounds like cost saving sounds like more efficient output. That sounds like just something that, uh, is no longer something that you can afford to ignore. Now, shifting into this idea of context layering and everything with it, uh, in an enterprise, the different layers or different levels of management will also have to get used to this and see how they can integrate it into their part of the business. What are the most important things each role can consider as their next step?

**[0:20:27] Host:** So what should a C E O, what should a C F O, what should the architect be thinking about right now?

**[0:20:32] Eran Yahav:** Oh, that's a hard one. Uh, I I think, uh, C E O should really think of how do we build the infrastructure that can make the company move a hundred x faster, uh, with the, uh, deployment of agents. And the key understanding here is that unless you define the world the agents operating very carefully, uh, you are going to, uh, get into havoc. Because you are putting very powerful machines into a universe, uh, that is not clearly defined, and letting them manipulate things beyond what humans can review.

**[0:21:24] Eran Yahav:** So we're already kind of generating code faster than any human can review it, and our only hope is to define the universe such that the agents are doing the right thing almost by construction. And you know, the the context engine also informs the code review process, so you will get better code reviews as well from the context engine. But just thinking about the C E O level, uh, you don't really want, like things being generating incorrectly, going to review, getting pushed back to fixing, getting pushed back to review again, and this ping-pong between generation and review.

**[0:22:07] Eran Yahav:** That is kind of just conceptually wrong and inefficient. You want the agents to be doing the right thing from the get-go, and in order for that to happen, you need the context engine. You need to provide them sufficient context to to do the right thing up front. So that's kind of the C E O level. It's really laying down the infrastructure that will make agents successful. And this is true for software engineering, but it's also true more widely in the organization for the other systems in the org, which are all also going to be, um, uh, revolutionized by agents, right?

**[0:22:41] Eran Yahav:** And a a, and so again, the agent itself is like this very powerful car. It's like a Ferrari, it can go really, really fast, but if it doesn't have a map of where is it trying to go, which is to just drive in circles very, very quickly and and basically burn a lot of fuel and get nowhere, right? So it's all about kind of like defining the map in which these cars or these agents can go really, really quickly and do the right thing.

**[0:23:09] Eran Yahav:** So that's C E O level, uh, architects really need to think of how to structure the project, the software projects themselves, in a way that decouples a lot of the dependencies. Because, again, this is no, not really different than what we do with human engineers, but you need to give agents kind of swim lanes in which they can operate without kind of interfering with one another, with one another very deeply. So the modularization, the breaking of systems into components, and that part become even more important.

**[0:23:42] Eran Yahav:** It was always important, by the way, but becomes even more important, uh, these days. And the last persona, I don't remember which one you mentioned, so sorry, was that the C F O? Or is that?

**[0:23:52] Host:** Our C F O? Yes, the C F O. Let's think about, uh, R O I, and the quantifying reworks and things like that.

**[0:24:00] Eran Yahav:** yeah, so definitely thinking about the R O I of agents, these things can become really expensive really quickly. Uh, the basically L M tokens are, uh, not that cheap, and while they will become cheaper, I think the the agentic workflows are consuming many tokens. It's not unusual to see people spending thousands of dollars getting nowhere, uh, and this is even, uh, more common in again, what we call the brownfield projects. Think not project that I'm writing from scratch, something where I actually have to go into an existing system and change some small things in many places.

**[0:24:42] Eran Yahav:** Uh, this is often quite hard for the agents, and and so I think putting on the R O I hat, uh, you will need to measure how much are you spending on agents and have some visibility into the velocity of the team. What is that I'm getting as output? I have to say, these are like subtle, uh, questions of how exactly do you measure the the velocity or the quality of what you're getting out? But we are working on that, and I think there's like, there, there's gonna be, there has to be, uh, more meat around this in in the next, uh, I would say, uh, let's call it a couple of years, even, right?

**[0:25:29] Eran Yahav:** Like, more holistically, and not just online, but the entire industry. Because the the current ways in which we have to measure this are not very, not not sufficiently sophisticated.

**[0:25:39] Host:** But it's something that definitely should be measured. So it's, uh, uh, the importance of context, uh, layering is important to every senior person within an enterprise. It's not just for one part, one department, or or one person's use, it's it's something that can change the entire enterprise.

**[0:25:56] Eran Yahav:** Yeah, hun 100%, 100%. And I I think again, if you just think about it abstractly and you say, hey, uh, context engine is really doing the homework, building a map that other agents can use to be more successful. That's pretty easy to to understand. Uh, I think there's a nuance there, which is like, what is the granularity of this map? Uh, if you build it at a too fine-grained level, uh, you spend a lot of time and effort maintaining it and updating it.

**[0:26:27] Eran Yahav:** So definitely, the intent is not to build like a super detailed street-level map of the organization, call it, uh, but rather, I think we found very interesting, uh, design points in which, uh, the map is precise enough to make agents successful and fine-grained enough to make agents successful. Without requiring heavy lifting on the maintenance, the incremental maintenance of this map, which is kind of the the engineering balance that you have to strike, uh, for this to really work out.

**[0:27:00] Host:** Absolutely so, Eran, from our conversation today, I think my biggest takeaway is that context engines and the idea of context A I is not a buzzword, it's not just a buzzword, it's it's the competitive edge that every enterprise needs at this point. Um, and that it's possible with trust. We can have context A I and give this A I all of our information, the ins and outs of our organization, and still be able to trust that there won't be data leakages or things like that.

**[0:27:28] Host:** And then it is industry wide, and it is C-suite wide, whether you are a C E O, a C F O, a C T O. Uh, it is something that needs to be integrated in all parts of an enterprise for it to be successful.

**[0:27:39] Eran Yahav:** Yeah, 100%. This is a question of, you know, the the maturity of A I in enterprise. I think as as this has matured, uh, I think the problem of context has become right. Our expectations from agents have become, um, I would say, higher. And as these expectations become higher, we find out that in order to meet these expectations, we need to equip the agents with better understanding of the universe they operate in. And this is what the context engine does.

**[0:28:11] Host:** That's perfect, it sounds essential. Thank you so much for your time today and for giving us such insightful information around context. A I

**[0:28:17] Eran Yahav:** Thank you very much, thanks for having me.

**[0:28:22] Host:** Wrapping up today's episode, there are three key takeaways for C E O s, V P s of engineering, and architects from our conversation with Eran. First, moving beyond A I pilots requires shifting from generic models to a context engine that acts as an organizational brain, providing agents with the specific onboarding and legacy system knowledge they need to be productive. Second, implementing a dedicated context layer can significantly optimize R O I by doubling agent success rates and reducing token consumption costs.

**[0:28:49] Host:** And finally, to maintain enterprise grade trust and prevent data leakage, this infrastructure should ideally operate within the company's own secure perimeter or air-gapped environment. Position your brand alongside the Fortune 500 leaders defining the enterprise A I roadmap for the opportunity to showcase your solution to the executives currently funding and scaling global initiatives. Partner with Emerj to reach the decision makers holding the strategic mandate. Secure your partnership at go dot Emerj dot com. Slash partner, that's G o dot e m e r j dot com.

**[0:29:19] Host:** Slash p a r t n e r. For further executive level analysis and to join our network of leaders delivering workflow impact with A I, visit Emerj dot com. On behalf of the team at Emerj, we'll see you on the next episode.