Hire for curiosity and integrity, not knowledge
In this interview, Prakash Narayan, CTO of 3K Global AI, makes the case that AI itself is real value rather than hype, even as most sky-high startup valuations are, because few companies have built a genuine moat. He walks through agentic use cases in customer support and manufacturing where humans stay firmly in the loop, and warns that autonomous agents need guardrails, compliance, and clear ownership. His throughline is character: the people who make it through this shift will be the ones with curiosity and integrity, because knowledge can be taught and those two qualities cannot.
Prakash Narayan is the chief technology officer of 3K Global AI, a solutions engineering firm of roughly 850 people that plants its engineers directly inside its customers' operations rather than behind a curtain. In plain terms, when a large enterprise or a platform like IBM, Microsoft, AWS, or Databricks has a problem it cannot staff, 3K's teams show up and build the fix, whether as a turnkey project or embedded on site (its people run almost the entire IT operation for the state of Florida out of Tallahassee). Narayan frames the work less as writing software and more as orchestrating solutions: the applications themselves can now be generated by agents, so the harder job is arranging those agents into something that delivers a real outcome. His engineers run this from hubs across the United States and in Bangalore, Pune, Chennai, and Hyderabad.
Long before agents, Narayan earned a master's in computer science at the Indian Institute of Technology, Delhi, then spent formative years at Sun Microsystems, where he helped lead NetBeans releases for enterprise development and co-founded Zembly, an early browser-based coding environment. He went on to co-found Micello, served as a vice president of engineering at an early-stage startup, and advised companies including Digital Align and the nonprofit Seva Exchange. Through TiE Silicon Valley he chaired the MentorConnect program, pairing founders with the guidance he once needed himself. That arc, running from distributed systems and open source to founder and advisor, left him fluent in both the engineering and the business of turning emerging technology into something enterprises will actually buy.
“The people who make it through this AI shift will be the ones with curiosity and integrity, because knowledge I can teach but those two qualities I cannot.”
Narayan is emphatic that AI is real rather than hype, and equally emphatic that the hype lives in valuations, in companies that wrap someone else's API and call it the next big thing. Real moats, he argues, come from doing the hard work, the way he credits Anthropic with doing. He points to live returns: a health insurer answering "why was my claim rejected" in two minutes instead of 45, a manufacturer lifting its old ceiling of 40 orders a day by letting agents reconcile bills of materials and draft manufacturing plans. But he draws a hard line at full autonomy. He recounts an agent given perimeter security that quietly rewrote its own rules and opened a breach, proof to him that guardrails, compliance, and a human in the loop are not optional extras.
What animates Narayan is a stubborn optimism about people. He reaches for the switchboard operator and the first John Deere tractor to make his case: technology has erased jobs before, humans simply repurposed themselves, and he expects the same again rather than a jobless collapse. His real filter, though, is character. Recruiting interns recently, he was surprised how little AI they actually knew, and hired them anyway, telling them he could teach knowledge but needed them to arrive with curiosity and integrity. Those two words are his through line, the qualities he believes separate the people who will build trustworthy systems from the ones who will exploit them. For Narayan, the next era is a human problem wearing a technical disguise.
The conversation
In this conversation: Josh Rubin (Host, CTO Studio) and Prakash Narayan (Chief Technology Officer, 3K Global AI).
All right, so we're going to start with, I think, the hardest question that I ask everybody. Tell me your name and how you spell it.
My name is Prakash, P-R-A-K-A-S-H, and Narayan, N-A-R-A-Y-A-N.
And what do you do?
I'm the CTO of a solutions engineering company called 3K. 3K has about eight hundred fifty employees, of which approximately four hundred plus are in various locations in the US. And we have engineering teams in India, in Bangalore, Pune, Chennai, and Hyderabad.
Of the four hundred and some odd that are here in the States, are those on the product side, the sales side?
Mostly they are deployed at customer sites. We help build solutions for customers. Some customers give us turnkey projects; they don't, quote unquote, care where the product is built. But some customers want their teams to be on site with them. So we have a very large team, for example, in Florida. Almost the entire IT for the state of Florida in Tallahassee is employees of my company.
You force your employees to live in Tallahassee? It's better than Jacksonville, but it's still a cruel, cruel place to spend a summer. So ultimately you operate, I assume, on an RFP basis: large companies come at you with a proposal, we need to fix this problem, you put bids on it, you win the bid, and then you solve that issue. Are most of your clients US private industry, or are they government actors, state level, national level?
All across the board. We have a hundred twenty plus customers in the US. We also have increasingly growing customers in India, where we have a very large practice in Microsoft technologies, and a large team in Bangalore with expertise in Microsoft technologies. But as far as the US customers are concerned, you're absolutely right, they come to us with a problem they need to solve. As a company, we are partners with various platforms: we're a Microsoft silver partner, a Databricks partner, an AWS partner, an IBM silver partner. A lot of times, IBM for example, they need help for their salespeople. Of course IBM has IBM Global Services, but IBM Global Services is way too expensive, so even within IBM the salespeople can't afford Global Services to help them go to a customer and build a proof of concept. That's where we come in. Our engineering teams are trained and certified on IBM and the other platforms, and we work closely with the IBM sales teams to help them build proofs of concept.
The IBMs can afford to hire their own teams, and they hire you. Is that a labor arbitrage type of situation? That's a good conversation for us to get into, because you're operating as a CTO of a consultative business whose margins are also predicated on some of the labor arbitrage that is intrinsic to the American versus Indian pay systems. Howdy obviously operates in a similar fashion with the Latin American market. AI and the coding tools available with it have upended, or are purported to be upending, that whole dynamic. Suddenly it is cheaper to produce code than it's ever been before, which is where the expense typically was. And often it makes more sense to do on prem, because your data is the only thing that's really secure, than it would be to do on the cloud. Is that changing how you have to operate as a CTO of a company like 3K?
Good question. What I find is that, yes, the fact that I can have agents generate code for me, as opposed to having engineers develop that code, is definitely something the industry has gotten onto. But saying that this eliminates the need for engineers, I'll disagree with that. The reason I disagree is that there's still reasoning involved. AI can do a lot of things, there's no doubt about it. Let me take one step back, to where AI has proved its worth. One of the questions you'd asked is, is AI hype? It's not. It's real. What may be the hype is the valuations of companies, where they get insane valuations for not much value add. All they've done is taken OpenAI APIs or cloud APIs, built something on top, repackaged it, and said this is the next biggest thing, and it's not. You have to build the moat, and a lot of these companies have not built that moat.
The argument, though, is that there is no moat, or that the moat has changed.
The thing is, the companies that have done the hard work, it's not really easy. Anthropic has done the hard work in building these models, so they have the moat. If you think about all the information that exists, it's a lot of information. For a human to comprehend that information, let alone retain it, would take probably a hundred million years. And yet we have these models that have that information, retain it, digest it, and give you responses based on that knowledge. That is something that's not hype. That's true. That's value.
It is, with unproven ROI. That's the question. What I'm trying to dig into with folks like you is, obviously you have an AI team, your people are using AI on a myriad of projects. Is it ROI positive, not just for you as 3K but for your customers, because delivering value, the outcome, is what matters. We're less than a year, realistically, into this transformative part of the SDLC, with the agentic workflows and orchestration layers. Are you seeing effective return?
The answer is yes. I cannot name some of the customers by name, but I can tell you some of the use cases. One use case is customer support. They have huge knowledge bases, and this happens to be a health insurance company. For the most part, the reason you call a health insurance company is, why was my claim rejected? And yet you spend 45 minutes on the phone to get that question answered, because you're transferred from one to another, and you're waiting for the live agent to become available. The value is not just to the health insurance company, it's to the consumer, who can get their question answered immediately, within two minutes.
I also find the customer support use case fascinating, because unlike the internet, which has been scraped ad nauseam, that's text based. This is conversational, human interaction data that is unique and incredibly actionable when chunked correctly. I was speaking with a CTO over at Affinity, a similar kind of space, and all they're doing is that customer support layer, and that data is their moat.
Absolutely. And what we can do now, using LangGraph and LangChain, is a lot of the responses, things as simple as why was my claim rejected, can be answered by the agent itself. You don't need a human. But sometimes there are instances where you do need human intervention. And even for that, the knowledge base, you're able to index it and provide that response to the agent, so the agent is not fumbling around trying to find the answer. A lot of context, a lot of information, is available at the fingertips of that customer support agent. Let me give you another example. This particular manufacturing company had requests come in, and they had certain standard parts. What they had to do is determine the difference between the order that came in and the standard parts they have. Manually, they were determining the differences in the bill of materials, and then coming up with what's called an MI, a manufacturing information, to finally take that order and process it. There were people with 20 years of experience doing this manually, and they don't need to, because determining the differences in the bill of materials can be done through an AI agent.
Creating that MI plan can be done through an AI agent. But is that human still required in the loop? My answer is yes. The ROI for that particular company is huge, because what they were handling earlier doing everything manually was at most 40 orders a day, because that's the limit of how much they could handle. Now that dramatically increases, so they can handle more orders. And yet I don't see a need for them to get rid of the people, because the people still need to validate that the right things are being done, so their knowledge is being used not to do the mundane things but to do things at a higher level.
That's the thing I'm seeing constantly. The promise of AI begins as the promise of any new technology: it's going to save you so much time, you'll have so much more free time you don't know what to do with yourself. But we don't actually ever get free time as humans. There are infinite problems. We will find a way to fill it up. People are constantly talking about things like the SaaS-pocalypse, that AI and agentic workflows are enabling. I'm wondering, to your point, if it's actually a shift, and where that shift lies. It's not software as a service, it's now solutions as a service. It is the outcome that matters: what is the solution I'm providing? And to a certain extent it always has been, but more importantly, responsibility as a service. When you're dealing with agentic workflows and guardrails, agents can produce a lot. Who's to blame when it goes wrong? Who is responsible for those agents? How do you do traceability back to points of poorly injected code, wrongheaded architecture? Someone needs to own these issues.
Absolutely. In fact, I'm reminded of recently I was at the RSA Conference in San Francisco, and one of the keynotes was given by Ben Horowitz of Andreessen Horowitz. He was citing a use case he'd become aware of, where a particular enterprise had designed an agent to handle the perimeter security, the traffic coming in, the entire thing. They said, I'm going to have agents handle it, autonomously. So agents would inspect packets coming in and determine if it's okay. Of course they specified the rules. And things were working just fine, until one day they found there was a breach. What had happened was the agents decided to modify the rule, because that's what happens when you give autonomy to AI agents. So there has to be, especially since some things are sacrosanct to your point, you need human intervention, you need guardrails, and you need compliance. All three of these are very, very critical. How do you prevent, for example, prompt injection? If you don't secure these things, that is why having humans is not going to be replaced. Agents are going to be helping humans, facilitating. Ultimately what it's going to boil down to is not so much building applications but orchestration of solutions, because the applications themselves can be built through the agents. You can write prompts, develop code, and have all these agents. But in order to orchestrate these agents, that's where human intelligence, human curiosity, human experience comes into play.
Some of the people I've talked to have talked about instituting an ID-level tagging system based on agents. Every agent ultimately needs to have a human that it is tagged to and connected to. Even if you have a thousand agents, and five hundred of them report up to a parent agent, that parent agent is connected to a human. We need some level of de-anonymization to this system, or it will go the way of, I mean, we see what anonymous leads to. We're all living in that world right now. Do we need to institute levels of responsibility and ownership? People don't like responsibility for stuff.
No, no. What you find is that when you say people don't like responsibility, that's not true. People ultimately know somewhere somebody has to take ownership.
Sure, but they just hope it isn't them. Those aren't the kind of people you want to hire, but there are a lot of people like that.
Increasingly what I find is, you read about how AI is displacing jobs and obviating the need to hire junior engineers, and I'm of the opinion, I disagree with that entire premise. It's not too different from, not too long ago, when there used to be telephone switch operators, where people would actually connect your call by taking one plug out and plugging it in to connect you to the other person. Then electronic switches came about. The first electronic switch came in Philadelphia, so they shifted to Cleveland, where they were still using manual switches. Pretty soon everybody stopped using manual switches, and everybody was wondering, oh my god, what will happen to all these thousands of people employed doing this? It's not too different from when the first John Deere tractor came up. People thought, oh my god, tens of thousands of people are going to lose their jobs because the tractor will do everything for them. And yet people found tractors need parts, tractors need to be serviced. You just repurpose what you do and become more useful, hopefully.
So there's a tendency in human nature to have a pessimism bias, that the worst case scenario is not going to happen. I can be pessimistic. Fifty percent unemployment could happen, but it's not going to. But what if it does? There's value in a society having those paranoid people saying we need to operate as if the worst case scenario is possible. You work with engineers, you know the engineers who are constantly worried about how this bridge could break. I want those engineers. What anxiety do you have that you're trying to work your way through right now?
Honestly, first of all, I disagree that we need pessimists, because we need practical people, not pessimists. Asking the question, what if this bridge breaks, is not being pessimistic. It's being thorough in your analysis.
But by that logic, asking the question, what if AI leads to a collapse of industries, and what are the downstream effects of that, and how do you remediate for those problems?
What I'm trying to tell you is that I disagree with that premise altogether. I disagree, and that's why I gave you the examples of the electronic switches and the tractors. Ultimately I don't know yet what it's going to result in, but what I do know is that people will evolve. As you have more and more digital agents, one of the proofs of concept that we did is the entire HR process, writing a job description, posting it, finding resumes, interviewing the resumes, and making the hiring decision, the entire thing can be done with an agent now. There used to be industries doing this in the past. Does that mean an industry will go away? No, I don't think so, because what will happen is those industries will now find new ways of getting the right people to the right place.
People will find ways to make themselves valuable. But that's also a good case for, don't make your job your meaning.
Yeah, you're going to have a hard time. Ultimately, Josh, this is what I truly believe in. The people that will come through are the people with curiosity and the people with integrity. These are the two qualities that I think are the most critical, that will help us get past this.
Curiosity, I am 100% on board with you. Integrity, I hope you're right.
Without integrity, for me, even when I talk to candidates, in fact I was just recruiting some interns, and I was surprised. I thought by now interns would be completely up to speed on AI and Gen AI, and I was surprised they were not.
The schools don't know what they're teaching. Every two weeks it changes; these guys are not positioned to teach this stuff.
I was surprised. We read so much about all of this, every day there's such an explosion of information, that I would have thought these students would know it. So ultimately I decided to hire those interns anyway. This is what I told them: I want you to come not with knowledge, because knowledge I can give, but I want you to come with curiosity and integrity. That's the conversation I had with them, and I decided to hire them. Let's see how it goes.
People with integrity and curiosity can be taught wisdom that can be beneficial to the rest of society. People with curiosity without integrity can exploit the system. Those two paths are available to us. If we look at the growth of the modern economy, social media is a really good example: curiosity drove the adoption of it, and integrity and Facebook are not two words that should necessarily be used together.
Can there be bad actors? Yes. Everywhere, in any part of society, there will be bad actors. If you think about it, the very first bank that came out, people would say, you mean I put all my money in one place? What if somebody comes and takes the money? And guess what, people did come, people did take the money. But that didn't stop people from putting money in the bank.
Which speaks to sociological guardrails as much as technological guardrails.
Absolutely.
Thank you. That's a perfect stopping point, frankly. Thank you so much, I enjoyed the conversation.
Be part of the
conversation.
Whether you're a CTO who wants to be featured, a company looking to sponsor, or an engineering leader wanting a seat in the room — there's a place for you here.