Nobody Knows What's Running

Jake Aaron Villarreal: I'm Jake Aaron Villarreal, born and raised in Silicon Valley, here to take you behind the scenes to share what it's like to be a startup founder. The journey they're on, the problems they face, the products they build in an effort to make our lives better. I'm excited to have with us today Vincent Dermont, principal engineer and co-founder of Guild AI. Vincent, welcome to the show.

Vincent Dermont: Thank you for having me.

Jake Aaron Villarreal: Well, I'm excited to have you. I know we had a call a few weeks back and it was really insightful about where you're at, what you're [music] building, and where things are headed when it comes to AI. Vincent, before we jump in here, where are you calling us from?

Vincent Dermont: So, right now I'm in San Francisco. I'm usually based in Los Angeles, but uh, we are in our office in the Financial District in San Francisco.

Jake Aaron Villarreal: Really cool. I love that area. Well, I spent some time in there as well when I was working at Oracle, and I was commuting back and forth down the peninsula. So, hopefully, it's not too much of a, what is it called? June gloom.

Vincent Dermont: It's okay. We're okay today. Good, good day.

Jake Aaron Villarreal: Okay. Very cool. All right. Well, um, a little bit more about Vincent. He was the head of engineering at Lightspark and a member of the leadership team there. Most recently, he held engineering and management positions at Meta, where he worked on a wide range of projects, including Libra and Novi. Throughout his career, he has focused on distributed systems and developer experience. Now, he is helping build up Guild AI, which, as you can imagine, is an AI startup. So, give us a little background, Vincent, yourself. We know you've been in technology for quite some time. What were your early experiences that kind of led you down the path of getting into technology?

Vincent Dermont: I think, I mean, it's really, you know, it goes back to when I was a kid. I loved building stuff, and uh, I grew up in the '90s and, you know, it was the explosion of computer science and the internet and all of that. And I kind of, you know, went really deep into it. Uh, I started pretty early on with gaming, and instead of playing the games, I would try to build them. And then I got into engineering school in France, studied computer science, uh, got really into it, worked in a couple of startups, created one before I finally moved to Silicon Valley. I spent a lot of time in San Francisco working in startups, working at Meta, and now I'm back to co-founding a startup and trying to build something new.

Jake Aaron Villarreal: I like it. Well, it starts with having curiosity for technology, and then learning it, and getting involved, and then creating systems that hopefully make our world better. You know, you mentioned you were at startups, and I know that you had spent, I think it was up to five different startups you were part of, and at Facebook or Meta, you were there helping build out Libra. And you know, you watched a company you were part of evolve into what became TikTok. Looking back, walk us through that a little bit, because everyone knows about TikTok, but kind of where did it start?

Vincent Dermont: It was a company called Mindie, uh, that I joined uh, when I moved to the US. So, it was a US company created by French people, and they had this really cool idea of uh, creating looping music videos. And so, you would basically take your phone out, shoot something, and pick music uh, from iTunes, and just like send that looping video and share it with your friends. It was really fun. It didn't work out for a variety of reasons. Maybe we were a little bit too young and inexperienced to actually go through. But ultimately, what we had done had given birth to Musical.ly, and that, you know, ultimately became TikTok. But I wasn't part of that adventure. I was just at the very beginning.

Jake Aaron Villarreal: Yeah. Got it. Well, you've been part of lots of companies that have done some incredible stuff. Uh, talk about Guild. You know, everyone's building AI systems and agents, and you know, there's an evolution of how you build it, what it solves, why we're doing it, how do you manage it all. So, give us the inspiration behind Guild.

Vincent Dermont: So, I think, really, Guild is an AI control plane. So, what we've seen when working at Facebook, and then when working in other startups, is that you start using more and more agents, but you kind of need to know what they're going to do. And you know, we hear stories all the time about dropping databases in production, or someone last week told me that uh, they had an AI agent just completely publish their entire sales pipeline in like a public Notion document. So, that type of stuff happens, and it's okay when you're prototyping, when you're playing, but when you start to have autonomous agents that run in your production systems, you really want to have some sort of control and governance around them to know what's happening.

If something wrong happens, you need to be able to roll it back and maybe put some guardrails around what the agents are allowed to do. And honestly, there is a lot of, you know, similarities with how you would handle a large employee base, right? Like, if you have a large company, you don't give full access to every single employee to all your systems. You usually have a lot of permission systems, you know, whether it's data access or networking, you usually have those systems that kind of put boundaries around what people need to do their work. And an AI control plane is basically that: like, how do we take all the AI agents that you have created and that you want to use and make sure that they run in a safe way?

Um, we saw something happen at Facebook internally. So, our CEO James was leading the developer, um, developer experience team at Facebook, and so he was building the systems there. They built a system that, you know, basically would bring agents to engineers and builders across the company, and they would share those agents in some sort of marketplace, and people went crazy for it. They started forking the agents, copying the code, adjusting it to work for their particular IDE or their specific system, and so that's something that we are going for. So, we want to build that control plane that will run your agents securely. But we also want to have the agent hub that is something that can be internal to your company or even completely public where you can build and share agents that are not your secret sauce, but that really make you faster.

Jake Aaron Villarreal: Wow, that's amazing. You know, any type of technology that you get in front of customers takes some time to try it and experience it and see how it works. How long has Guild been around now, and what's the traction look like so far?

Vincent Dermont: Yeah, so we started the company in, I think, October last year, uh, maybe end of September, last fall. We spent, at the time, there were not a lot of, um, agent SDKs and harnesses. Like, you would have your Claude codes, you had LangChain, you had a few of those. But we started by building our own agent harness, um, that could basically wrap LLMs and run agents efficiently in production and in the cloud. We were really focusing, uh, on automated agents. So think something where you get a trigger from somewhere, maybe, you know, a PR is published on GitHub, and then an agent gets started and starts working and calls other agents to continue the work, and then you get your results.

So, we started with that. We built an orchestration layer, and we built our control plane into that, where we can run basically agents and have full logging permission controls whenever the agent is trying to access external services, and you can define policies much like you would do on AWS or Google Cloud or Azure. You would be able to define policies that control your agents. So, we started doing that. Now we've opened the door to more agents, right? Like, if you're using Goose, for example, you can bring your Goose agent on the platform, and it gets integrated automatically on the control plane that we have. So yeah, that's been the journey.

Right now we're in public beta. So, we have the website that is open, but most of our focus right now is to work with our design partners, which are our first set of customers, and we're trying to make them extremely successful. Like, the goal is really to prove the value with a small set of customers and then expand. Uh, so we're working on the agent hub that I talked about where we're going to share agents with the world. We have a variety of customers. It's actually very interesting because, you know, our product really helps different types of companies.

If you're a very large company, governance is something that is part of your DNA most of the time. Our product resonates really well with very, very large companies that you've probably heard of. Um, and we're working with a few of them on how to bring AI to their world. Um, if you are a smaller company, maybe you're more interested in cost management, and you're more interested in, you know, with the latest trend of token maxing, controlling the AI cost, and all of that becomes actually like really part of the game. And so, we have like a wide array of types of customers that we are making successful right now.

I heard a quote from one of our sales calls last week and it was pretty interesting. Like, we were talking with the CTO of—they had, I think, 150 engineers on their team. So, not a small company, but not a very large one either. And he was telling us that he felt terrible because people were using AI everywhere, and he loved that, and they were moving faster, but he had no idea what was running, where it was running, what was being accessed. And he was kind of panicked 'cause he didn't want to slow anyone down—like, that's really not the goal, like, you want to move as fast as possible, especially with what AI gives us—but at the same time, he had a responsibility to his customers and to his stakeholders and builders to keep things secure. And so, like, that's the kind of thing we can really help with.

Jake Aaron Villarreal: So when you say control plane, is it really like a pilot? I mean, I got this image in my mind like it's a pilot kind of driving a, or flying a plane, and controlling all the parts of the plane that need to work to make sure you're going in the right direction. So in this metaphor, is it understanding the agents, what they're doing within the systems, understanding access to certain, I don't know, parts of the company and data, and then making sure that who has access gets the right access? You can turn them on, turn them off. Like, give a use case, maybe not talking about the company name specifically, but, you know, who is it that's most interested about your platform? Are you selling this to like the CTO? Are you selling to the CEO? Are you selling to the developers within an organization? Who's most interested about what you're creating? And is there anything out there in the market today like this?

Vincent Dermont: Yeah. So, it's like, the control plane is really, you have your system, your existing production systems, you have your data, and then it, uh, sorry, your agents on top of it. And then it's a plane that goes like really in between. And so, your agents are going to access your systems through the control plane, which gives you a centralized place where you can see what's happening and you can control what's happening. So, it's really, that's really how I see it. It's like a thin layer that goes like in between the two and that gives you complete control. It's something that is like frequently used in networking, where you're going to have like, all your network is going to go through like a single plane and you can manage your controls there.

I think right now we have like the two sides of the solution, because usually when we talk to C-level type people, so CTOs, CISOs, CEOs, they're very excited about the value that we are bringing 'cause they have a company to manage. They have a team to manage, and they want to move as fast as possible and as safely as possible. But usually what happens is that those C-level people, you know, ask their engineers to go and check it out and see if that's something that would work for them and that they could use. And so we have this like two-pronged approach where we really have the governance aspect that really speaks to the C-level people, but then the ability to build, deploy, and share agents is actually really speaking and resonating with builders in general. So we start mostly with engineers, but we're kind of opening it to builders because now, you know, designers, PMs, marketing, they're also building things. Uh, and so we really have that two-pronged approach, the Guild sandwich, if you will, and uh, it resonates with both sides.

Jake Aaron Villarreal: Yeah, that's good. Thanks for clearing that up. There's always tension between security and speed. How do you help organizations embrace AI without slowing builders down?

Vincent Dermont: Yeah, that's a really good question. So, I think we are not opinionated on what you should do. We give you the tools to control what's happening. And then we have a team of forward-deployed engineers that actually help companies go through the transition when needed. So, some companies are already very AI-forward, and they want to go from a place where everything is open to starting to restrict things a little bit more. And when I say restrict, it's not about removing the ability to do things. It's about making sure that you don't do things you're not supposed to. So, it's not about slowing down. It's really about making sure that we, you know, for example, an agent that is doing a marketing task should probably not have access to deploying or deleting your production database. And it's really about pulling those boundaries in.

But we also see some companies that just have no idea how to do AI. You know, like CTOs that get told by their board and by their investors, like, "Hey, you really need to move faster. You really need to go and do AI." And they, some of them have no idea how to get started. Like, as soon as you get out of the Silicon Valley AI bubble that we live in. Um, there are lots of companies that actually need guidance and help on, "Okay, like maybe they have a few engineers using Claude or Codex on their laptop and that's about it. Like, how do we level them up? How do we create those autonomous agents that are going to automate some of their processes and make them better?" That's also something we've seen and we help with.

Jake Aaron Villarreal: Yeah. What do you think is the biggest misconception executives have about deploying AI agents inside their organization?

Vincent Dermont: Yeah. I, so I see AI, at least at the moment, I see it as bionics for builders rather than robotics that would completely replace builders. Um, it makes engineers much more efficient, but it doesn't remove the need to actually design systems properly, understand what's happening. And so you can very easily get to a prototype like that, you know? You can one-shot a prototype with Claude or Codex very easily, and that's amazing. That allows you to iterate very quickly, to test the market, to put things in front of your customers.

But when you have a complex distributed system in production that has many moving parts, you still need engineers and builders to understand what's going on and how to make it reliable. So I think, you know, we're still hiring a bunch of junior people. I've seen in the industry there has been a lot of layoffs and there is a lot of fear from junior engineers that they might not find a job. We do hire a lot of junior people because I'm, I'm getting old. I've been in the industry for a while, been building, you know, software in a certain way for a very long time. So, you know, most of my career I've been mentoring engineers and teaching them how to do things. And I think what's very interesting is that now we're going both ways. When I hire more junior people, they're AI native. They've been using AI their entire career, and so they bring a lot to the table, and they teach me a lot of stuff on how to use it more efficiently, and I teach them a lot of things around how to design systems properly. And I think that that's been very surprising to me and very cool.

Jake Aaron Villarreal: Yeah. Well, you're talking our language, because we're in the business of helping companies find engineers and product leaders. And you know, it's, there's a shift of, you know, as you're building AI into a company, whether it's, you know, systems or agents or managing, you know, whatever you need to manage when it comes to AI, you know, the talent that has to do that has to change, too. So, you know, I've heard this from many founders as well: we can't find the right people, they don't have the core skill sets yet that we need. You have a lot of engineers that are well established in their career, but they're still also ramping up. So, the combination of having that senior engineer with sort of that junior engineer that only knows AI and development, or at least is very in the trenches of what, you know, today's world looks like, that happy marriage seems to be working out well for, sounds like for your organization and other ones we're talking to. Which is great to know.

Vincent Dermont: Yeah, which is great to know. I read about that, uh, also in, like, the CTO of Shopify made a post about it, and she was giving really good advice, I think, which is that our job is changing, right? Like, before it was about writing the code in the right way. Now, writing the code is virtually free. What matters is understanding, you know, one, two, three layers of abstraction below [clears throat] so that you can judge if the AI agent you're using has been doing the right thing or not. And so you kind of like shift your focus on different parts of the problem and a lot less on the lines of code themselves.

That being said, I think there is still, I see a lot of similarities between what makes an agent understand your codebase easily and what makes a human understand the codebase easily. Like, there is a lot of overlap, and so it's still, you know, as an engineer by trade, like, I still care a lot about making sure that you can't make mistakes. In the same way that 5, 10 years ago we were building frameworks that would put, you know, guardrails around what an engineer could do in a codebase, I think doing the same with AI agents is extremely efficient.

Jake Aaron Villarreal: Yeah, well, let's talk a little about the younger engineers. We get calls all the time from, you know, people in college trying to figure out their path going forward. They're studying computer science, they're going to come into the market. What's your advice to them? What should they be investing their time in learning, and what would make them more hireable as they get out of college?

Vincent Dermont: Yeah, that's a really good question. I think there is one thing that doesn't get replaced yet, at least. It's taste, right? Like having the right taste for a product that is easy to use, a product that solves like an actual problem, a product that really will speak to customers. So it's less about how do we make it work and more about what should we do? And I've seen, um, I've seen engineers, you know, go crazy and build a lot of things, and then you end up with like a Frankenstein monster of a product that is unreliable, because sometimes it's not because you can build that you should. Sometimes choosing what you're going to focus on is actually like more valuable.

So I think that I would advise younger engineers to like try to build a lot of things 'cause now it's free, but develop some sort of taste for what you should build. And I would ask them to be very curious. Don't just paste something in an LLM and get the response and send it out. Try to understand, is that the right thing? Why did it do it that way? What should we do? Because I feel like, um, if the value you bring to a company is typing a prompt, it's not extremely valuable. If the value that you bring to a company is typing the right prompt and making sure that you understand the output and that when you're going to send your code for review, it's been thought through and it goes into what the, you know, the company wants, that is very valuable. So, like, stay curious, learn things. This is one of the biggest technology changes I've seen in my lifetime. But there have been other changes that have been pretty impactful, and you know, you just have to be curious and learn, and learn, and learn more.

Jake Aaron Villarreal: Yeah, I like that. Well, that's good advice, and that's certainly something that we'll continue to share. How about for the engineers that have 5 to 10, 15 years of experience, and they're in a company where they've done very well, but, you know, there's a changing of the seas, and they might, their company might downsize? I mean, we're hearing a lot of fear from engineers that have done a really good job. What advice would you have for them? What should they be thinking about? What should they be working on? What should their future look like?

Vincent Dermont: I mean, honestly, it's similar advice. Like, I think if you're not learning, then you're duplicating yourself. So sometimes it means getting out of your comfort zone, right? Like, I know for me, like at the very beginning of AI, I was like, "Well, okay, this is cool, this is exciting, but I like the craft. I like to write a beautiful piece of code." And I quickly realized that like, no, this is not where we're going. Where we're going is actually those tools are giving us superpowers. And so instead of being afraid of them and not learning them, you should embrace them. You should just like experiment and try.

I have a few friends that I've been talking to, and they work in very large companies, and they don't like AI 'cause they see it as something that's going to come and steal their job and make them useless. And I keep having this debate with them where I'm like, "Yes, if you react like that, that's what's going to happen. If you don't..." Because you can't just like learn AI tomorrow. You need to use it for months. You need to understand the tools. You need to stay on top of what's happening, what's expected. And then it means that you become more productive, you become better, and you're actually more employable. If you push back, and if you try to keep things as they are today and prevent change, then when the change is going to happen anyway, you're going to be obsolete, and that's a bad move. So I would really try to rekindle your curiosity and take it as an opportunity of learning something new that's going to change the industry.

Jake Aaron Villarreal: Yeah. Okay. I want to get a little more tactical here on this. Now I'm going to talk about an employee that did really well as an engineer, maybe had been part of startups, had been part of a larger organization, and has been downsized, and now is on the market trying to get their AI experience up and work on things that would be valuable to get, you know, hired again. What specific tools should they be learning? What specific projects should they be building on their own? What would be valuable to know that they can do if you wanted to hire someone who's on the market, who's got a lot of experience and maybe spent the last three, six, maybe nine months learning AI? Where should their focus be?

Vincent Dermont: So, I think that what I would look for when I interview people, I want to understand a few things. First, do they have, like, you know, solid computer science knowledge? And it sounds like they do because they've been doing that for a while. The second thing is, have they graduated to using AI? We don't do, you know, LeetCode whiteboard interviews anymore. Like, we don't do that. What I do when I do coding interviews is I just, you know, give them a problem that is like really large, and I look at them using the AI tools they want. So, you know, things like Claude, Codex, and others, they can use whatever they want, but they need to really master the tools. So they need to have built a bunch of things with them to understand exactly what the limits of the tools are, what the right prompt is going to be, what it can and cannot do. And I want to watch them fly through the problem. I want to watch them, you know, use AI to understand what happens, um, be critical about the output, like understanding it deeply, and move fast. So that's for like the technical side.

But then there is the, I would say the like, more behavioral side. I want to see them automate things. Like, if I was to hire someone today, and I do, I want to see how they've been automating, uh, things that in the past would take them a long time and now are kind of like, you know, they can actually do it with agents. So for a few examples of things that you can do on the platform is I hooked all my GitHub accounts to Guild, and I have a bunch of agents that search for different things in PRs to like flag to me when there is something that probably needs my attention. I have a bunch of code reviewer agents that do like a pre-reading of any PR that I have to review to give me some pointers on what might or might not need my attention. At Guild, we have built a bunch of, uh, agents that basically look at the logs in production and automatically, you know, try to find bugs and suggest fixes for them. So like, any part of your workflow, you should probably try to automate in some way with AI, or at least have the curiosity to see what happens. And I'm not talking about removing yourself entirely from the equation, but I'm talking about making your daily processes faster and easier for yourself.

Jake Aaron Villarreal: Yeah, I love that. So, if you were looking at someone who maybe is still working, and your internal team reached out, and they were interested to talk to Guild, and you were talking to them about their day-to-day work, and they said, "You know, 90% of my code is actually done by Claude, and you know, I'm an engineer, but you know, most of the work that I'm building is through, you know, tools like that." Is that a positive or a negative?

Vincent Dermont: To me, it's a positive. I, you know, I want Claude to write most of the code, but Claude writing the code doesn't mean Claude thought about it, designed it. Mostly, like, you know, you're on the hook when it goes wrong, right? Like, I think building something in a, you know, in a greenfield space where you start a company, you start a new project, you have nothing, it's great, there are no bugs. But whether you use code or not, there will be issues in production, so your job as an engineer is to catch those, fix them, make sure that they don't hit the customers too hard. But also as an engineer, like, engineering I think is a team sport, right? Like, you usually work as a team, and so how do you make sure that what you write with Claude is actually helping the rest of the team?

So there is a lot around, you know, how do you update your specs to make sure that other teammates and other agents can understand them easily? If you have code review, like, and most companies do, because most companies have SOC 2, um, SOC 2 assessments and they need to have some sort of code review. Uh, you know, how do you think about your reviewer? How do you make it easy for them to review? Do you have an agent that helps them focus on the right portions? Do you write, you know, gigantic PRs that you haven't reviewed, or are you more tactical and thinking about your reviewer? There is a lot of things that you can improve in your process.

But right now we're mostly focusing, you know, when we go and we talk to companies, we're mostly focusing on automating processes that don't really need a human in the loop. Like, we have a lot of use cases for example around ticket processing where we've seen companies that have, um, you know, a customer creates a ticket, and then you have an agent that reads the ticket, that tries to triage it, and then it's sent to engineering. Engineering tries to triage it again, and then it's sent to the actual real engineer that needs to work on it. There are a lot of parts in that process that are, you know, very underoptimized and can be really helped with AI. I'm also very excited about the self-healing factories that we are building. That concept of your agent is going to live in your infrastructure, and it's going to see what issues happen, and it's going to fix them before you even have to wake up. That type of stuff is really interesting, and we have a large variety of use cases that we support on the platform that we're always excited to show to our customers.

Jake Aaron Villarreal: Yeah, I love it. Well, you're building what I think is the future of how agents will be managed and operated and deployed and built, just orchestrating it and controlling what's being used. We have some companies we're talking to now that do other parts of that, which is listening to what the agents are saying and what they're doing and are they doing the right things that they should be doing. And so there's some interesting concepts that are happening all around that space. As you look at Guild going forward, I know you raised capital, you've got a team, you're building a team, continuing to expand that team. What's the next 12 to 18 months look like for you? What are you excited about? What's on the roadmap?

Vincent Dermont: Yeah, I mean there are many things I'm excited about. We just released our insights dashboard. So now when you go to Guild, we have our first version of this observability piece where you can go and see, you know, what your agents are doing. You can see token spend. You can see which agents are, you know, consuming a lot and which agents are not. You can slice and dice in many dimensions, which is really interesting. I think the next steps for us are going like fully, completing our launch basically. We have a public beta today. We're doing a lot of, you know, white-glove approach with our customers to make sure that they are successful. The goal for us is going to be to scale that. Uh, so making sure that we manage to build a self-service product that actually brings a lot of value.

We want to launch the agent hub, which is, you know, that place, that marketplace where you can actually publish, share, find new agents, fork them, and adjust them for exactly what you need. We think that's going to be extremely important because a lot of people don't really know where to start. It's hard to write an agent. Engineers usually can figure it out, but builders in general, it's really hard. There is a lot of care and a lot of effort that goes into building an agent that is actually good and not just AI slop. And so like, being able to find those agents on a public marketplace is actually very interesting. And then there are multiple parts, right? Like when we talk about observability and governance, I think we're only at the very beginning. Right now you can define permissions that are very, um [clears throat] static I would say, just like you would do with, you know, regular SaaS services. But how can we have more automation there? Like how can we detect bad patterns, how can we trace the data through the process between agents, how can we actually help maximize your token usage so that every token delivers more value, and how can we do that at scale on your entire fleet of agents?

Jake Aaron Villarreal: That seems to be a hot topic that everyone's trying to figure out, is the token maximization, how much you're spending, you know, who's controlling what the spend is. And uh, I think Uber, if I'm not mistaken, went through like most of their spend in the first quarter and it [laughter] was like allocation for the year. So, how does your platform help manage that? Because I know that's going to be something that's top of mind going forward for, you know, months and maybe years to come.

Vincent Dermont: Yeah, those tokens are expensive, and they're still, I mean, you know, when you look at the major AI providers, they're still heavily subsidized by investors, but at some point we're going to have to pay for the real value of the tokens. You know, granted, like, it's likely that the cost of a token goes down, but usage is going to go up. So, yeah, cost is a real problem. Our platform helps you in many ways. Like, that's really one of the problems we try to solve is, so since you are running all your agents in the control plane, you can see, you have the data on what agents are doing, and so you can slice and dice that data in our insights dashboard to understand like, where am I, like, where is my money going? Are there like some agents that are consuming a lot?

And then you can ask yourself, if that agent is consuming a lot of tokens, is it actually valuable? Like, how do we get into measuring the return on investment on that? Like, was this run that just ate, you know, 30 million tokens, was that a successful run? Did you actually get the value you were hoping out of that, or should you revisit the agents? Maybe run it less frequently, maybe build it in a different way or ask your prompt in a different way, so it doesn't have to do as much work and you can get more value out of it. So, we really like, give you a lot of visibility in all of that today, and we will in the future build on top of it to actually proactively advise you on how you can optimize your usage.

Jake Aaron Villarreal: That's great. Well, you'll probably get a lot of followers and people interested in learning about how you're doing that. So, I like to leave a little space at the end of our show of as you grow, the type of roles you're looking to hire for, and it's kind of just a free way to talk about, you know, what's top of mind for you in terms of your growth as a leader on the engineering side, and what type of roles are you looking to fill today, and what should people be thinking about if they want to, you know, apply directly to you or to the company?

Vincent Dermont: Yeah, so right now we have a few roles open. You can check out guild.ai/careers, I believe is the link, uh, where you can go see our open roles. I know on top of my mind that right now we're hiring, uh, production engineers or a DevOps person basically, uh, because running all those agents basically means, you know, having a large fleet of servers, um, that we need to manage, and um, you know, we containerize everything, and so there is a lot of work to be done there. Uh, so we're really looking for someone who's going to be mostly in the, you know, San Francisco is preferred, um, at least like, you know, West Coast time zone for now. We're still a small team, so we're optimizing for being close together. Um, in the future we will likely expand more. So that is a role that we're really looking for.

Another type of people is, um, you know, I call them the LLM whisperers. People who really want to build agents, and that means, you know, taking a problem and writing a prompt, writing evals, helping us build our eval platform to make sure that we improve our agents all the time, and really going and being passionate about solving our customers' problems using AI agents. And this is interesting because it's very different, it's a very different type of job than traditional engineering, I would say. In traditional engineering, usually, you know, you have a specification of what you want. You write the code, you write unit tests and integration tests, and things are fully deterministic, and when it works, it works, and if there is a bug, you can add a test and it never happens again.

Building an agent is non-deterministic by nature. So if you ask the agent the same question twice, you're going to get a different answer. Hopefully not too different, but sometimes very different. And so writing evals, which are like the new type of test for LLM agents where you're basically trying to statistically verify that your agent kind of does the right thing most cases and doesn't fail too often, is a very different thought process. And it's often hard for traditional engineers to like, move to that type of mindset. Um, but we're really looking for people who have that type of mindset and are interested in going through the grind of like making that agent amazing.

Jake Aaron Villarreal: So if you're someone out there listening to this podcast, where, Vincent, do you think they would be? Would that be an engineer somewhere building something and hearing what you're asking them to do, and maybe they're already doing that in their current company, or is it someone like an AI lab that's, you know, building something fundamental, or where do you think that skill set lives today?

Vincent Dermont: I think, I mean, I've seen a lot of people who were machine learning engineers before who have that type of mindset. So, like people who have, you know, been used to working with machine learning and data in the past, but they can come from anywhere. I mean, AI foundation labs of course, but it can be engineers who are just very product-oriented and really want to solve problems and are ready to like, test and iterate many, many, many, many times. I think it's more of a mindset than like actual skills. Like the skills can be taught, like how to write an eval, like we can teach that, but it's more of a, you know, there is a problem and I really want to like go deep in that problem and make it work for our customers. So you need empathy. You need empathy for the user because there is a lot of, a lot of that job is about looking at what customers have typed and why it failed and understanding why, and iterating and trying something different and trying something new and improving your set. So um, it's a very rewarding activity, honestly. Like when you get that agent to work, it makes everyone better. It solves large classes of problems at once, but you need to have that mentality of going deep into the problem.

Jake Aaron Villarreal: Yeah. Got it. Well, exciting. Well, I'm really happy to have you on the podcast and what the future holds. Be great to see how things go over the next 6, 12 months. Maybe have you come back and give us an update of how Guild's doing. So again, Vincent, thanks so much for joining. I'm your host.

Vincent Dermont: Yeah, thank you for having me. This was great. And yeah, I would love to come back in, you know, 6 to 12 months and talk about what happened in between and hopefully how we grew and solved a lot of problems.

Jake Aaron Villarreal: I love it. Well, thanks for coming and thanks for listening to our audience. It means a lot to me. You spent your time with us here today. I'm your host Jake Aaron Villarreal signing off for now, but can't wait to catch up with you all on the next episode. Until then, Vincent, the world, take care. If you like what we're doing, don't forget to subscribe, leave a review on Apple Podcasts or wherever you listen, and follow us on YouTube where we go behind the scenes to learn what it takes to be a startup founder.

‍