In this episode of Dev Propulsion Labs, Anuraag Gutgutia, co-founder of TrueFoundry, breaks down how to sell enterprise AI infrastructure when nobody trusts you yet. He also covers how TrueFoundry evolved from an ML deployment platform inspired by Meta’s FB Learner into a full enterprise AI gateway, why every company now can’t avoid AI, and why voice agents are the next big niche to bet on.
Watch the full video on YouTube.
Transcript:
[00:00:00] Victoria Melnikova: Hi everyone. Welcome to Dev Propulsion Labs, our podcast about the business of developer tools. My name is Victoria Melnikova. I’m the head of new business at Evil Martians, and today I’m excited to introduce Anuraag Gutgutia, co-founder and CEO of TrueFoundry. Hi Anuraag.
[00:00:24] Anuraag Gutgutia: Hi Victoria. Great to be here. Thanks for having me.
[00:00:27] Victoria Melnikova: How are you today?
[00:00:28] Anuraag Gutgutia: I’m good. I’m very excited to talk about the dev GTM and you know how it kind of impacts the way AI can propel in the world, and very excited with the last episodes that I’ve seen, especially of some of my co-fellow co-founders in the similar space. Yes. So looking forward to it.
[00:00:45] Victoria Melnikova: Yes, yes.
So as I mentioned, Dev Propulsion Labs is the podcast about developer tools and the industry of developer tools is changing a lot these days. So let’s start with an introduction. What is TrueFoundry? Who are you? What brought you to founding TrueFoundry?
[00:01:05] Anuraag Gutgutia: So TrueFoundry is an enterprise AI platform. Our goal is to enable enterprises to build, scale, and deploy gen AI and agentic applications in their enterprise and do it without any constraint of what compute they’re running it on and within their VPC.
In order to power that, we offer two products primarily. One is an AI gateway product that allows you to bring all the components that go into agentic AI, whether it’s the models, the MCP servers, the agents itself, and gives you full observability and governance around that. And the second is our deployment platform, which allows you to deploy your agents, deploy your large language models, and scale it without worrying about how you manage the underlying infrastructure.
[00:01:49] Victoria Melnikova: Can you give us your kind of background? Because you managed a hedge fund, right? Yeah. That’s your background. What brought you to found an ML tool basically in 2021? Yeah. 2022.
[00:02:04] Anuraag Gutgutia: So I used to work for a hedge fund called WorldQuant, and we used to be an algorithmic trading fund. So we used to use like signals across the globe to kind of forecast how stocks will move.
And I used to be a portfolio manager, trading equities as well as currencies, got a chance to manage around 600 million in assets for them. So that was a really good experience. And at that time, like, you know, my fellow co-founders, Nikunj Bajaj — we went to school together at IIT Kharagpur in 2009 to 13. So I’ve known them for like 16 years and they were at Meta.
Meta has a really solid internal platform for machine learning, which was called FB Learner. And you can imagine that companies like Meta are like at least five to seven years ahead of the curve. So even though Meta released Llama in 2022, they have probably been working on this for years. Right. And when we used to discuss the way Abhishek and I used to kind of [00:03:00] take models into production at Meta versus the difference that existed in the public domain, we found there is a huge gap.
We thought if we can build and bring that platform to other companies around the globe, it can accelerate AI in general for other companies. And that’s how we started TrueFoundry. Our goal was to bring FB Learner, which was Meta’s internal platform, to every company around the globe and do it at scale.
[00:03:23] Victoria Melnikova: Interesting. So this wasn’t your first time founding a company, right?
[00:03:28] Anuraag Gutgutia: Yeah. This was not, yeah.
[00:03:29] Victoria Melnikova: What does it take to take that leap? Like was it your friends that reached out and said, Hey, let’s start this new business? Or how did it start for you?
[00:03:38] Anuraag Gutgutia: It was kind of a mix of several things. So in my college days, I wanted to build something and we had started something in the education segment.
But obviously as a college kid, you probably don’t know much. We raised some funds from the government, but finally we exited it and we realized that, okay, there is a lot more that needs to go into building a startup. And after a while, once I was in the job, [00:04:00] spent six years there, got a sense of the industry.
You know, reached out to my friends and they were on the same boat. Uh, you know, they have been at Meta for several years and wanted to create an impact for the world. And we felt that there could be a huge impact that we can create, especially as AI is going through an inflection point. And it is kind of very, you know, surreal for us that when we started, AI was still going through small trajectories of growth.
But suddenly in 2022, after ChatGPT launched, it actually went through an inflection point. So a lot of things that we wanted to happen suddenly propelled with speed. So I think the main reason is, you know, being able to create impact and then working with co-founders you have known for a long time just makes the life easier.
So, you know, that was a really good segue to start something of my own.
[00:04:47] Victoria Melnikova: That’s interesting. Did you always envision TrueFoundry as a top-down company? Do you always target enterprise in terms of how the GTM works?
[00:04:57] Anuraag Gutgutia: Not really. Like when we started, our primary target audience was a lot of enterprises.
Because what we realized is if you look at, let’s say, the Fortune 1000 companies and you kind of go into how each company is doing and impacted by AI, you’ll be surprised that a lot of them aren’t really using AI in production. And that kind of was not aligning for us because we felt that the tech companies, when you think of Meta or Google, they’re using AI to scale.
So when we tried to decipher what is different, we realized that it’s actually the core platform that can make it really good. And our goal was, can we actually enable all these enterprises to really scale AI in the right way? And that’s how we started. And because of that, initially our approach was, Hey, you know, we’ll reach out to the people who can have the say in these organizations, which will be like the chief AI officers or heads of platform engineering and so on.
But as AI has grown, I think we do work with a lot of [00:06:00] smaller and mid-market companies and in order to get the developer love, we have worked on a lot of bottom-up motion over the last year as well.
[00:06:09] Victoria Melnikova: Kind of like, go through the last year. And what did it mean for TrueFoundry? Because you started it in 2022, right?
[00:06:16] Anuraag Gutgutia: Correct.
[00:06:18] Victoria Melnikova: And the original pitch was ML workflows. Like what was the original pitch?
[00:06:24] Anuraag Gutgutia: Yeah, maybe I can tell you what the original start was and how it transitioned.
[00:06:29] Victoria Melnikova: Yes, yes.
[00:06:29] Anuraag Gutgutia: So when we originally started, our pitch was we want to bring an enterprise AI platform that enables data scientists and machine learning engineers to deploy models in the best way with the best SRE practices in a cost-optimized manner, and with the right governance within their enterprise, within their VPC and a multi-cloud or hybrid environment, and do it at Meta scale.
That was our pitch when we started in 2022 and one of our first enterprises we were working with was [00:07:00] a pharma company, one of the big pharma companies in the US, and while they were working with us, ChatGPT got launched and three months after that they wanted to deploy their own large language models. And they kind of told us, Hey, can you enable this deployment platform to support the deployment of, you know, Llama, which was launched by Meta and some of our other fine-tuned models that we can run within our own compute.
And that’s where we extended the platform to support deployment of large language models. And slowly as we went into more depth, like, you know, it was not just about models, it was a lot of other components. Like today, you want to deploy MCP servers, you want to deploy agents. One of the things that stood out for us that helped us was from day zero, we built it as a software engineering platform.
So it was never like an MLOps platform. It was always a platform that took any code you wrote, Dockerized it, and deployed it. So even if you had to deploy agents, it’ll just work. Even if you had to deploy MCPs, [00:08:00] it’ll just work. So we just extended the positioning to kind of cover that. So that was one major change that happened.
So from deployment of models to deployment of any services that go into AI — models, MCPs, agents, et cetera. The second change, which is an even bigger change, I would say, is as AI progressed further, like maybe end of 2023, when gen AI was one year into its cycle, a lot of companies that we were working with realized that it’ll not be one model.
It’ll probably be OpenAI along with Bedrock, along with Anthropic, and they wanted to have a way to be able to control how these models are used within their company. Who has access to which models? How do I ensure there are the right limits so that someone does not end up creating a huge cost bottleneck?
How do I ensure if one model goes down, the other model is able to pick up, and I’m able to put the right guardrails? So that is [00:09:00] where again, one of the enterprises we were working with wanted to have this layer, which at that time was called an LLM proxy. It has evolved to an LLM gateway and now an AI gateway, and that’s where we kind of built our AI gateway.
And from day zero, we built it in a way that it could cover all the components that go into agentic AI. So these are the two product lines we offer, and any company that is building agentic AI in production is starting to want an AI gateway internally where they can manage their models and MCP servers with observability, governance, and reliability.
That is something that has changed. So earlier it was just a deployment platform, but today we have a deployment platform but also an AI gateway.
[00:09:40] Victoria Melnikova: So it’s like a natural evolution of something that you technologically already had. And now you’re just serving the needs of today, basically.
[00:09:49] Anuraag Gutgutia: Exactly, exactly.
[00:09:50] Victoria Melnikova: Has your customer profile changed in these years? Because right now everybody probably wants to have AI in their enterprise.
[00:09:57] Anuraag Gutgutia: I think it has changed quite a lot, [00:10:00] Victoria. So when we started, we used to sell to like the heads of data science, heads of machine learning. And generally the interface with the engineering was not as high because not all platform engineering teams were involved in AI — they were mostly involved in their software development toolkit and so on.
But with gen AI, what has changed is suddenly AI is no longer a data science thing. It is like anyone is using it, right? Like within the organization, all the engineers are using these models as APIs and building agents that they can scale. So today the customer profile when we look at the ICPs has changed from, you know, the heads of data science to also heads of platform engineering.
You know, heads of IT infrastructure who manage these endpoints within the company. Sometimes because security is so important, we end up talking to the CISOs in organizations who want to care about, oh, what are the guardrails that are being imposed? How is [00:11:00] MCP authentication secured? So there are a bunch of personas that have now become important and relevant to us, and that has been one change.
The second change has been what you pointed out — pretty much like earlier, out of 10 companies we used to talk to, eight used to say, okay, we are doing some AI, but there is less business impact. But today, every single company we talk to, whether it’s a mid-market company, whether it’s an early enterprise, or whether it’s a Fortune 100 company, all of them cannot avoid AI.
It’s now a boardroom discussion. It’s like a boardroom topic for them and they want to invest into it. And suddenly, you know, there is impact on the bottom line in terms of how can I scale my customer success processes, or automation in customer success or customer support, or how can I improve my internal processes. And now it has gone into launching AI products to actual customers to deliver ROI, because people have realized that if they don’t do it, they’ll start becoming irrelevant. And [00:12:00] now when we talk to companies, 10 out of 10 is pretty much the number.
[00:12:05] Victoria Melnikova: As your ICP has transformed into something new, how does that affect your developer experience? Do you have to think about your product differently and how it’s served?
[00:12:16] Anuraag Gutgutia: Yes. There have been some changes, Victoria, but one of the things has been from day zero, we built our product to be very software engineering focused, and the good thing for us was suddenly software engineers who actually weren’t the direct users a lot of times have now started becoming users.
So the product already took care of that. We always had GitHub integrations built into the system. We always had YAML that software engineers like to use. We always had APIs that people like to consume. It’s just that now there are, you know, software engineers who are using it at huge scale.
[00:12:53] Victoria Melnikova: Yes.
[00:12:53] Anuraag Gutgutia: So the experience layer needs to become even better.
And there are a lot of infrastructure-level [00:13:00] flexibility requirements that you want to build into your product. For example, because AI has become so important, because data is so important, suddenly people want things like, oh, you know, out of all this data that is passing through OpenAI, I want certain pieces of data to be stored somewhere with certain access controls and certain pieces of data to be stored somewhere else.
So those architectural patterns and design changes are what the product encounters. And the second change has been that now even smaller organizations with developers are signing up automatically and they want to kind of connect their models, because now everyone has an OpenAI key. They want to connect it. They want to see the metrics, so you want to make the developer experience in terms of signing up really good. Those were two major changes, I would say.
[00:13:43] Victoria Melnikova: So as a founder in this space, you probably see or hear a lot of conversations about agent experience, right? And what it means today. I’m not sure if it impacts you directly because you’re providing an infrastructure layer for enterprises. But if we [00:14:00] think about what are the main pillars of agent experience in 2026, what would you say is the most important?
[00:14:07] Anuraag Gutgutia: So maybe I’ll explain this from a fundamental basis — like what exactly is an agent? So an agent is in some ways a non-deterministic workflow that has intelligence built into it, and that comes from a set of components within it.
So when you look at a real-life agent that any company is launching in production, it has primarily five components. One is models, which could be your models as APIs or your own hosted models like OpenAI, Llama, whatever. The second is MCP servers, which give the models the ability to access real-time information or specific information that could be specific to companies, et cetera.
Third is prompts, and prompts have evolved a lot because now models have much better [00:15:00] capabilities compared to three years ago. So how well your prompts are, and how you could version them, how you could pass the right prompts depending on the right situation, is the third pillar.
Fourth is guardrails, because if I have to launch it in production, I want to ensure that it’s saying the right things, it’s not revealing my PII data, it’s not saying things that are hate-speech based, and so on. And the fifth is an agent itself — so an agent could call another agent that could help it do a subtask.
So as long as you have these five components, you can build any agent in the world, right? By combining them into workflows and all. So to your question, these are five components that you want any company to have the ability to use in a very simple way, and that means they should be able to access this, they should be able to quickly look and debug things across these five components, and then build an agent workflow, test it out, and launch it. So that’s what [00:16:00] it takes to build a real agent workflow. And that means now that even people who are in data science or ML are starting to think about APIs, which they didn’t have to interface with a lot earlier.
[00:16:14] Victoria Melnikova: So we talked about what an agent is, right? But if we think about another SaaS that an agent will be using — we’re talking about agent experience, right? So how do we make it easy for the agent to actually do the job?
[00:16:31] Anuraag Gutgutia: Yeah. So once you have an agent, now imagine a company — let’s say we take an example of Automation Anywhere, right? It has document extraction and document processing. Earlier it was a rule-based workflow. Today it’ll probably be an agentic experience for a customer where they can pull off models. So the customer who is offering SaaS to other customers is also thinking, how can my product evolve and offer an experience layer for my customers which is agentic.
So that is one level [00:17:00] of change that has happened because now people are not thinking about just internal workflows. It’s being launched through their customers. Which means that the way you design your product, it has to be built in a way that allows you to ship this agent component — which means these five components in a box — that you can ship to your customer either in SaaS or on-prem in the case of an air-gap customer.
That is something you need to build a capability for, and that’s not easy because it means you’re pretty much shipping now complex models and APIs and other things in a box to a customer, and that needs you to be really good at the infrastructure level.
[00:17:35] Victoria Melnikova: So now there is this whole debate about how to work with the models, right? And how to prompt your AI workflows and things like that. There was this study — I was just talking with TJ from Chroma about it — the study that showed that using skills and agents.md, whatever, it actually burns more tokens and produces worse results. That’s what they say, allegedly. Do you have an opinion on [00:18:00] that? Like, do you guys have to study it at all, at a certain scale, to understand how to serve enterprise customers better?
[00:18:08] Anuraag Gutgutia: Victoria, like any new technology when launched has certain flaws, right? So skills as a concept has come very recently. So it’ll start getting better. I think we probably need to just look at what OpenAI was when it came versus what the experience is today — like it used to only access historical data, and today it can access real time. So this is just going to change. I don’t think that debate holds, that if you use skills and other things it’ll burn more tokens, because tokens will start becoming less and less costly over a period of time, and people will start consuming more tokens.
It’s like in the industrial age, you are consuming water or coal to produce electricity. Here it’s almost like you’re consuming electricity to produce tokens, and people are consuming tokens, and the tokens are going to produce further things that create ROI. So that’s the world today.
In order to be ready for enterprises, you need to [00:19:00] just ensure that as enterprises or even other companies are consuming these tokens, there is a way for them to control it. You know, for example, let’s say you are in an enterprise and there are 10,000 other people. If I open it up and there are people who do not know what they’re doing, they might end up burning millions and billions of tokens, and that can cost a huge amount to the enterprise because these are not cheap.
So the only thing is ensuring that there are all these so-called policy engines that your product enables, so that as people consume these tokens, or use models or prompts, there is some way to put limits on that — which could be in terms of rate limits, budget limits, team-level limits, customer-level limits. There are a huge number of different ways in which you can define these limits, and that’s what the product needs to enable.
[00:19:58] Victoria Melnikova: Okay. Let’s go back in [00:20:00] time. Can you explain how you approached your first sales? So you and your co-founders created, let’s say, an MVP v1, right? And now you’re ready to get your design partners on board.
You obviously have an interesting background — you’ve been on both sides, on the hedge fund side but also on the founder side. How do you approach sales? Do you just map out your ICP and start cold outreach? Do you travel to San Francisco and go in person and wait by the door for the CEO to come out? What are some actual workable tactics that you’ve used to get your first customers?
[00:20:45] Anuraag Gutgutia: So I would say initially getting the first customer — let me actually break it down. The most important thing is the person, no matter what position that person is, will buy from you only if they [00:21:00] trust you.
So the only exchange currency literally is trust. Now as a founder, you’re trying to solve this problem: how can I establish trust with my potential buyer? There are different ways to do that. One is credibility. Like as a founder, if you have a great background and you have built tech products in the past, that builds credibility. That takes the currency of trust maybe from zero to 20.
The next thing is, can you actually get on a call with them and give them a view of the vision of the future? Because enterprise buyers are not buying just for the state of the product today, but they are buying into you — that this person has credibility and this person also knows how the future will look. So I can trust him or her to bring that product to me, which I know will evolve because I trust this person. So that takes you to level two.
And similarly there are other ways, right? So when we started, our first goal was, okay, founders, we have the credibility, [00:22:00] that is good in terms of the backgrounds. We mapped out our ICPs, which were early enterprises and mid-market companies, and we figured out who in that was relevant, whether it’s the VP of AI engineering, the VP of data science, or the chief AI officer, depending on the type of enterprise or company. And then we started reaching out largely from our LinkedIn.
So that was one thing that helped us get into those first initial calls. And we used to do the discovery and we used to share the vision that we feel that ML is like software engineering, and ideally there should be no difference between your software engineering platform and your ML platform. And that’s where some of the folks who understood this probably bought into that trust in us. That is how some of our initial customers converted.
This was obviously propelled by a lot of other things. For example, we were doing a podcast called the TrueML podcast where we used to invite AI leaders to talk about their AI journey, talk about the deployment infrastructure they [00:23:00] had and how they had built it, which also helped us establish credibility as someone who’s bringing thought leadership into the ecosystem.
But ultimately it boils down to what do we need to do to build trust. I think that is pretty much it.
[00:23:14] Victoria Melnikova: Okay. And how soon did you start seeing results? Like did it take — you know how they say that enterprise sales take about one year to close one deal — did it come sooner to you, or what was the process like?
[00:23:26] Anuraag Gutgutia: I think the first enterprise customer did take, from the day we incorporated, a little more than a year. I’d say it is a process that takes time. But there’s one of our investor friends who calls this a good reminder — like one big enterprise is equivalent to like 50 seed companies. So the time is, I feel, worth it, because if you crack a good enterprise, that can actually be the right design partner that can guide you depending on what the tool should look [00:24:00] like in a large-scale setup where there are multitudes of teams, developers, and business units. That actually makes your tool evolve quite a lot to enable you to sell to other enterprises. We have gotten insights from enterprises that have really made our product today one of the first-class enterprise AI gateways. If we didn’t work with enterprises, a lot of those insights are not easy to get.
[00:24:25] Victoria Melnikova: Since the process is so long —
[00:24:27] Anuraag Gutgutia: Yeah.
[00:24:28] Victoria Melnikova: And especially at early stages, it’s a very nerve-wracking position because you have to raise money, you have to prove something to investors, there’s a lot of legwork that needs to be done, you need to hire the team. There’s a lot that’s happening. Did it ever feel soul-crushing, or was it more like, okay, we’re doing the right things, we’ll see the results — kind of slow and steady?
[00:24:53] Anuraag Gutgutia: Along the pathway of landing a big enterprise, we did land a few mid-market customers. So [00:25:00] I think we mixed it up. It was not just Fortune 1000 enterprises. We had companies that were anywhere from a hundred million to 200 million in revenues that we were reaching out to, and we landed a bunch of them along the way.
So our first customer we got in the first six months, and that was a decent-sized mid-market customer and that helped quite a lot. And obviously, once you have some customers, it helps quite a lot. It’s also dependent on the type of the product. Like at that time our product was a deployment product, which is going to be the core centerpiece of a company running all their AI, which means it needs time before they can trust you.
Today there are products that can be used at different life cycles of agent development, and some of them don’t necessarily have to be used in production. So that gives you a faster turnaround, an easier way to start. So it depends also on the type of product and what problem it’s solving.
Like when we launched our AI gateway product around two years back, we were able to sign customers on this much faster because [00:26:00] initially the AI gateway was more an experimental product — people wanted to use it to just switch between models and all. It was not impacting things directly in production when it was launched. Now today, it is the centerpiece of production. But it depends on the type of the product, is what I would say.
[00:26:14] Victoria Melnikova: So let’s talk about the more recent part of the journey. You recently raised a Series A from Intel, right? Walk us through your plans for this round. Right now the market is kind of being cut up between different players, and you need to establish your space. And obviously AI for enterprise is a very desirable spot to be in. Do you have certain milestones for this fundraise? What are some pressures that you’re experiencing? Kind of take us through this recent phase.
[00:26:51] Anuraag Gutgutia: So I would say first of all that our investors have been really great. We have had very nice relationships [00:27:00] with each of our investors from the seed to the Series A and even now. A lot of it is on us as to what plans we set. We are always obviously inspired and motivated by them to push what we can do, but I would say that pressure is not in any way bad — whatever there is, is more like a joint effort along with our board and the rest of the investors.
But having said that — our plans for this: when we kind of did the Series A, we were around close to maybe 20 customers, a little more than that. These are paying customers, right? There were obviously others who were using the free version of our platform, and at that time it was largely mid-market and some enterprises — like seven to eight bigger enterprises.
Over the last year we have grown to now have more than 50 paying customers, and we have doubled down on enterprises because the governance, observability, and control of AI is becoming an even more central piece when you look [00:28:00] at these enterprises. They have also gone in their journey from gen AI in POCs to now gen AI in production where this thing starts mattering.
So they also realize the importance of a central control plane through which you can manage all your AI traffic. So earlier our motion was, as I was telling, largely outbound. One of the parts of this fundraise is also to build our thought leadership and build the inbound motion, which we actually worked a lot on in the last eight months.
We wrote a lot of content in terms of thought leadership and engineering deep-dives on the need and the importance for an LLM gateway or an MCP gateway in an enterprise. And that has led to a huge amount of traffic for us in terms of architects or VPs or directors coming to talk about where they are at in terms of the AI journey and asking what we would recommend. And that helps quite a lot. [00:29:00] So that is one muscle that we wanted to build and expand on, and our goal is to reach more than a hundred enterprises by the time we get to Series B.
[00:29:11] Victoria Melnikova: So when it comes to thought leadership — we at Evil Martians write a lot of technical content. We have about half a million readers every year, which is pretty big for a small team. What are some things that you find are working in building that trust with your enterprise buyers?
[00:29:27] Anuraag Gutgutia: I think I’ll again double down on the word trust. So I’ll give you an example, right? We have an enterprise, one of the biggest private companies across the globe, and they are on a journey where they want to democratize the use of MCP servers. They have around 80 use cases in gen AI that they have kind of listed down, but they do not know today what is going to be the net use case or net scale. So what they want to do is enable people to use these MCP servers [00:30:00] and models and see what will happen.
Now if you think about that, they also do not know fully how these MCP servers will be used. So we have done at least more than 10 sessions with them just educating them on what MCP servers are, what they enable for you, the different ways in which people can use them, and how you could even expose your internal things as MCP servers, because that’ll give you an edge.
So a lot of that is not sales — it’s just enabling them. So I think the biggest part of trust in this era would be being able to go on a call and educate a lot of your customers, because a lot of customers are deeply immersed in the work that they are already doing. And you know, the AI world is moving so fast that sometimes it helps when an AI-first company that is actually building in that space can share insights they’ve learned from working with other customers.
And that is [00:31:00] something that has been appreciated. And I think that is the biggest part. In fact, we have launched something called TrueFoundry Academy, which is available to all our customers and even prospects, where we educate them on different aspects of how to build an agentic AI app, what are the components that go into it, or how do you ensure that you have the governance in the right way.
And later on for our customers to actually onboard into our platform and learn about that. And in this, we not only talk about TrueFoundry, but we talk about all the adjacent pieces of tools — it could be any tool that is on the agent side, it could be any tool that is on the evaluation side or on the security side. We try to create that awareness that this is the architecture that you need to have to be able to scale agentic AI, and that I think is the most important thing.
[00:31:48] Victoria Melnikova: So we are in San Francisco and we’re obviously in a bubble. We talk about this with founders like yourself a lot. When it comes to enterprise sales, it’s very clear to me why Fortune 1000 companies are interested in purchasing AI infrastructure tools and learning how to do it well.
When we speak about the broader world, do you feel like it’s still true? Like are they aware of what’s going on, or is it still gonna take a few years for this AI wave to kick in?
[00:32:19] Anuraag Gutgutia: I think if you had asked me this question maybe one year back, the answer would’ve been that a lot of people are still getting used to it. But it has really accelerated over the last nine months, I would say.
So not only enterprises, but even among enterprises, the companies that you would think could be legacy companies which will be behind in tech — they are sometimes the foremost adopters of trying to improve their internal workflows and bring huge efficiency. And even among the other set of companies, which are mid-market, early-stage companies, we are seeing that people are now trying to think of [00:33:00] things as AI-first or agent-first — like how do I have an experience in my app which earlier would’ve been a set of rules, but now I want to power it as an agent, power it as a chatbot. And that I think is starting to become more and more vibrant.
It’s still not a hundred percent, but it’s growing rapidly.
[00:33:25] Victoria Melnikova: Where does your focus lie? Do you focus on American customers, or are you thinking more global?
[00:33:30] Anuraag Gutgutia: Our focus geographically has been the US market in terms of how we kind of reach out and how we ensure we build credibility. And the US has a lot of say on other geographies — like when we have a US reference customer, we have inbound from an Australian customer who says, okay, you have this customer, let’s also talk. So I think the centerpiece remains the US, but we have customers across Latin America, Australia, Japan, [00:34:00] India, the UK, and the spread and density of AI customers is increasing rapidly across geographies.
But the US is still definitely the leader, I would say.
[00:34:12] Victoria Melnikova: When we talk about your current needs as far as the team goes, I can imagine that obviously you need thought leadership, you need sales, you need engineering. What are some people that are really hard to come by? What is the hardest person to hire for you personally and why?
[00:34:31] Anuraag Gutgutia: One is a product marketer who understands tech really well and is able to connect with the developer audience. Because product marketing in the traditional age was very different, right? You needed to understand the SaaS product and you could create thought leadership around that. Today because there’s so much noise — [00:35:00] when you look at some of the headlines of some companies, you’ll be just confused as to what they’re doing, and it’s very hard to differentiate and understand.
So for someone who is a developer, or an architect in a company, or a head of security engineering or head of infrastructure, they want to really know that piece that gives them the real architecture, the real workflow. It’s very important that product marketers can understand developers, understand the tech, and know how to talk in an architectural and infrastructure-friendly way. And also those who can keep upskilling themselves with the new tech that is coming. So I think that is one role that matters so much to the growth but is always harder to get and harder to scale.
And the other role I think I would call out would be pre-sales engineering.
[00:35:54] Victoria Melnikova: For pre-sales, or something else?
[00:35:56] Anuraag Gutgutia: Pre-sales — basically when you go into, let’s say, [00:36:00] a call with a customer. In the traditional world, if it’s a SaaS product, they could do the demos themselves.
[00:36:08] Victoria Melnikova: Yeah.
[00:36:09] Anuraag Gutgutia: And later it evolved to solutions engineers who could do these demos, but the demos were more static. Today the product changes every few days, so you need these pre-sales engineers to be really tech-focused. You can no longer just be a sales engineer — you really have to be a platform engineer who is ready to talk about the product to these customers and tell them why they should use it, what is the infrastructural complexity, what will this product grow into, what will the roadmap be.
[00:36:36] Victoria Melnikova: Yeah.
[00:36:36] Anuraag Gutgutia: So you almost need like a —
[00:36:38] Victoria Melnikova: — technical founder —
[00:36:39] Anuraag Gutgutia: — technical founder to be there on those calls. And that is a really, really hard space to fill. So these are the two roles I would say have been the most difficult ones.
[00:36:48] Victoria Melnikova: That’s very interesting. But also the most needed ones, because they’re both interdisciplinary — they both lie between technical and something else. Technical and marketing. Technical and business.
[00:37:03] Anuraag Gutgutia: Yes.
[00:37:04] Victoria Melnikova: When you hire people, what is one quality that you feel like they all have? Like an ideal hire for TrueFoundry.
[00:37:14] Anuraag Gutgutia: When we hire people, I think at least one quality that we want everyone to have is extreme ownership. We talk about this internally all the time. So there’s this book by the Navy SEALs called Extreme Ownership, and the way they define ownership is very nice.
I will probably just tell two lines of it: when you go to a war or a battle and let’s say you are a hundred people, everyone has some responsibility, right? Like the person next to you is probably going to fire a missile or man a tank, and the person next to you is supposed to save the other person next to you. But things go wrong in a battle. Things go wrong in a war, and maybe the person next to you is [00:38:00] no longer able to do that. At that point, you cannot say, hey, you were supposed to do this — I will only do my part and wait for you or find someone else. You do everything to win.
I think that is what extreme ownership is. You cannot have people who are just, okay, this is my job and A, B, C is your job. You need people who own the end objective, and no matter what, they’ll achieve the end objective. They feel responsible. If something goes wrong, it hurts them and they actually go to correct it.
I think that is probably the biggest quality.
[00:38:28] Victoria Melnikova: Do you think it’s possible to retain a lot of people like that as you grow, like as you become a big company?
[00:38:34] Anuraag Gutgutia: I think so. I mean, it depends. Fingers crossed — we have had some great teammates who have joined us since the beginning and who have stayed with us, and we are really fortunate to have them.
I would say we would not be where we are without the teammates in our team. And we hope that we are able to build a team that has extreme high ownership, extreme high agency, and who can take up the founder’s mindset and actually [00:39:00] pretty much make us redundant in a lot of places.
[00:39:04] Victoria Melnikova: How do you motivate people to come and to stay with you?
[00:39:09] Anuraag Gutgutia: The way to think of people staying with you is also in some way the trust. Any person who is joining you is joining you because they trust that if I work with this set of founders and this team, I’ll probably be building something that is relevant in the next generation context, or I will be getting great teammates to work with, or I’ll create a huge impact.
So I think you just need to ensure that at every single moment they are able to feel that impact, either in terms of a customer really doing something wow, or in terms of their teammates teaching them something they would not have easily learned elsewhere, or in terms of the future vision.
So I think these are three pillars, and as long as you do that right — and we are learning, we are also pretty [00:40:00] new founders and we are learning this — but hopefully with that, if there are good people, they’ll stick around. Because they’re not just thinking about what am I getting here, but they are also motivated by the bigger thing: I’m creating this impact and that can change the life of a customer.
[00:40:18] Victoria Melnikova: That’s similar to what you said about customers — it’s all about the vision and trust.
[00:40:22] Anuraag Gutgutia: I think a lot of it generally boils down to how do you solve for trust — whether it’s for your customers, your partners, your employees, your own co-founders, your investors. Things go wrong all the time, but as long as that trust is built, people will want to work together with you.
[00:40:39] Victoria Melnikova: How is your companionship with your co-founders? Do you complement each other as far as the skills go? Are you friends beyond work?
[00:40:48] Anuraag Gutgutia: We have been friends for 16 years. We were in the same dorm room in college and in our college we used to have inter-hostel competitions, [00:41:00] and we were captains for a set of competitions for our hostel or dorm room. So I used to lead the product and case study, Nikunj was the captain of open software, and Abhishek was the captain of hardware. So you know, since then we have been close friends and even as co-founders we complement each other.
We obviously have discussions around hard things, but it’s always because we trust each other. We know that if we are saying it, it’s because we feel that might be the right thing, and we ultimately always align.
[00:41:31] Victoria Melnikova: Do you have to be here in SF from time to time to make sure the business is growing?
[00:41:37] Anuraag Gutgutia: Yeah, so one of my co-founders is based here in SF and we have an office in San Mateo, so we have a team there and Abhishek and I keep traveling every few months.
[00:41:47] Victoria Melnikova: If you were to give advice to somebody who is thinking about starting a product or starting a company — AI-native, agentic, whatever it is these days — as somebody who actually speaks with enterprise [00:42:00] and mid-market clients all the time, what would you say is a good niche to look into? As far as the product goes?
[00:42:09] Anuraag Gutgutia: We are seeing that there has been a huge inflection in the world of voice agents. And fundamentally, if you think of the globe as your customers — a lot of people, the way they communicate is primarily via voice. They don’t use writing or skills. Probably about one third of the globe is able to write and code and so on, right? The rest of the globe primarily interfaces via voice, and a lot of those interfaces and interactions can be AI-powered. So I think voice is something, and if there are niches in terms of workflows that involve voice and speech, you could build unique agents — that’s a great area to look into.
We are also thinking about [00:43:00] how we can plug in to be not only the infrastructure for LLMs, but also for voice agents, because they will become a huge portion of the net set of agents that are created in the world.
[00:43:12] Victoria Melnikova: Yeah. I feel like as AI becomes a commodity for just regular people —
[00:43:16] Anuraag Gutgutia: Yeah.
[00:43:16] Victoria Melnikova: — they’re probably not gonna be using a keyboard or something else to use AI. That’s interesting. Okay. So we actually arrived at my final question, which is called the Warm FUZZ, and it goes like this: what makes you feel great about what you’re doing today?
[00:43:33] Anuraag Gutgutia: So the best thing is, you know, we are powering the AI traffic for some of the companies that we used to look up to. And it’s just amazing to see that a Fortune 10 company — one where we buy pharma drugs from — is actually powering its IVR via our system. And some of these moments where customers succeed and they’re able to launch high-scale applications powered [00:44:00] by our system — it feels great that we have been able to build something impactful.
[00:44:04] Victoria Melnikova: Finally, I would like to provide you with space to promote TrueFoundry. I don’t know if an average user can use it somehow, but maybe they can learn something from your academy or about how to build agentic workflows.
[00:44:17] Anuraag Gutgutia: Yeah. For everyone who is watching here, I would say that the world is changing really fast and being on top of this agentic world requires that you, in whatever role in your company, are able to look into an infrastructure that can power all the components of agentic AI. And that needs control, that needs governance, that needs observability, and that needs the right reliability, because you cannot launch a production application at a hundred million users scale without the reliability built in.
So at that point, we at TrueFoundry [00:45:00] come in, providing you that central control plane, which we call our AI gateway. And we believe that can help solve a lot of firefighting and chaos that you will otherwise face in your company. And if you feel that could be interesting, we are happy to walk you through what it means and even give you free access to TrueFoundry Academy where you can learn about what it means to actually scale agentic AI, and a free trial to our product where you can test it for yourself. So really appreciate you, Victoria, for having me here — it was a great conversation.
[00:45:35] Victoria Melnikova: Thank you, Anuraag. It was a pleasure. Always great to learn about new tools, but also to understand how you practically approach enterprise sales and things of that nature.
Thank you so much. Thanks for having me. Thank you for catching yet another episode of Dev Propulsion Labs. We at Evil Martians transform growth-stage startups into unicorns, build developer tools, and create open source products. If you are a developer tool that needs help with product design, development, or SRE, visit evilmartians.com/devtools.
See you in the next.
