Ask Any C-Suite This One Question About Their AI Agents | Hector Hernandez, COO @ Speakeasy

We actually had someone on our sales team set up an assistant, give it all these different access things. Did you send these emails? Oh yeah, yeah, sorry about that. I did send that email. Well, I specifically told you not to send any emails. You're totally right. You told me not to send those emails and I did it anyways. You know, it was kind of one of these things where scolding a puppy. What would happen if tomorrow you hired 500 people. You would never just hire 500 people, ask them to start tomorrow, and just give them like root access or passwords and credentials to everything and say, hey, good luck, go do some good things for the company. You heard a friend of yours doing that, you would think that is insane. Like, what are you doing with your company? That is what we're doing oftentimes with agents. And I was just kind of clicking like, yeah, yeah, accept. And next thing you know, I hadn't realized that I asked it to actually Connect to Slack and like publish the thing. You know, I hadn't— I didn't mean to do that last accept. It'll come through and it'll say, hey, Angelina just deleted all of your customer records. That's probably not something that you intended to do. Maybe not even something that you thought you could do. I'm an optimist. I'm a founder, Angelina. You have to be an optimist. There are problems that can be really scary. I think the best days for all of us are in the future. We can't let everything be unchecked. Humans know humans better than anything else.

Today I have Hector Hernandez. He's the co-founder and COO of Speakeasy. So today everyone is racing to roll out AI agents, and Hector's team is building the guardrail layer that decides what those agents are actually allowed to touch. And his customers include Google, Mistral, and Glean. Today he's gonna teach us what really breaks when you go from a handful of agents to hundreds of agents. Welcome, Hector.

Thank you so much, Angelina. Really excited to be here as someone that's listened to your podcast a few times. I'm excited for the conversation.

Thank you so much. When you were building Speakeasy, have you ever thought about what is this single question you're trying to solve?

Yeah, and we wrestle this with customers all the time as well. For example, we can come up with like a hypothesis or an idea around something that we're really excited about or that we really want to like tackle, right? Something that we're like, okay, like this seems like the right direction to go and something that we want to invest a lot of our lives in. But it doesn't really become true until you start talking to prospects or the customers that share that pain.

Yeah.

You know, they'll, You can say, hey, how are you thinking about XYZ? And they're like, oh yeah, that's nice, but it's not a priority. Like, we're not going to spend money to solve that problem. What I really need to solve is, you know, and then they give you all of this rich information and then you start pattern matching. If you have 2 or 3 or 10 conversations that are down the same vein, then you're like, okay, I'm onto something. And then you can go, you can build, and nowadays you can build so quickly and you go back to those folks and say, hey, you know that problem that you had? You know, we're kind of building in this space and then you can share notes with them. And that is really, I think, when the magic happens. So you definitely have to be passionate about a space. You need to have something that you want to wake up in the morning and you want to talk about, like that is definitely important. But then what that passion is going to lead you down, you need to find problems that people are Uh, not, not only problems that people have, but problems that people are willing to pay money to solve.

What is that problem?

For us, it's quite frankly that everyone, everyone we talked to today is looking to adopt AI. That's like the most obvious statement I can make, right? Everyone is looking to use AI for, at different levels, board and C-level. Everyone wants to be more efficient. You know, everyone wants to, you get this supercharge of efficiency that AI is going to bring to us. At a director level or at a manager level, these folks are being squeezed and they're being asked to do more with oftentimes fewer resources. And they're trying to figure out, how can I supercharge myself? How can I be both— think strategically about my own team, but also be able to do a lot of IC work, which I'm being asked to? And at the individual level, I think most people know that, you know, they're, they're not going to be replaced by AI, but they are going to be replaced by somebody that knows how to use AI better than them. And so they want to, you know, dip their toes and, and go, go into this as fast as they can. The real problem that has been presented as we talk to different people is, number one, how do we go from having a couple people using AI or solving some problems to actually scaling this out at a true enterprise? Like that seems like a really big jump to make. And number 2, once we have that going, how do we have some guardrails, some controls in place so that, you know, you don't have, you don't end up in the headlines or you don't have secrets being, you know, kind of passed around your entire company. That is one thing that people are concerned about. So first of all, they're like, how do I get adoption? How do I get started? But then very quickly they're like, okay, how do I do this in a secure way? How do I do this in a way that I am confident in the strategy that we're, that we are going down? You and I in the, you know, in the past I've talked a little bit about what would happen if tomorrow you hired 500 people. You would probably go through a pretty intensive Onboarding and you make sure that everyone has like the right solutions, they have access to them, that they have the right roles, they have the right permissions within those solutions, and you would get them going, right? You would never just hire 500 people, ask them to start tomorrow, and just give them like root access or passwords and credentials to everything and say, hey, good luck, go do some, go do some good things for the company. That would just never work. If you heard A friend of yours doing that, you would think that is, that is insane. Like, what are you doing with your company? That is what we're doing oftentimes with agents. We are just saying, here you go, here's a credential. Now go, you know, go, go do all these amazing things for me without thinking of, oh wait, there are no guardrails to these. What is the access? What are they touching? And that is a really fundamental problem people are trying to solve today.

That's what happened. That would really remind people this could happen when the initial OpenCLAW incident was exploded earlier this year. It was just so new and people are still talking about it.

Yeah, it's funny that you bring up OpenCLAW because we were really excited about OpenCLAW and we still are. In fact, part of our solution allows you to create some assistance, but give them a harness. Give them guardrails for what they can use and be appropriate and have some observability over what they're doing. But we started using it ourselves and we actually had someone on our sales team set up an assistant and give it all these different access things. And it very specifically said, hey, only do this. Don't do anything else. And then she woke up one morning and she got a bunch of emails from a customer of ours. And she's like, oh, these are weird emails from a customer. Like, I, what, what are they coming from? And then she's like, she saw that there was a thread actually that had, like, she had sent out an email starting this conversation with an existing customer. It just happened once. And she quickly realized that, wow, like this claw that I had created, even though I very much said you are never allowed to send an email without me knowing. You're never allowed to do any of this. It still went ahead and did it, right? And then it was, when you asked it like, hey, did you send these emails? Oh yeah, yeah. Sorry about that. I did send that email. Well, I specifically told you not to send any emails. You're totally right. You told me not to send those emails and I did it anyways. You know, it was kind of one of these things where scolding a puppy. They're like, yeah, I know I'm not supposed to do that, but I did it anyways. There was no harm, nothing was done, but it was a very quick warning shot for us that we need to innovate, but there is also a big layer of control that needs to be instilled as you're doing so.

It's so powerful. It's alarming, Right. I really like, I actually read your OpenCLAW blog post. It's really well written. You had like the 6 layers architecture and you, the complete architecture of an agentic system, like each layer, what can go wrong and how you invert the default so your version can make sure if it fails, it fails close. I took your diagram and I turned it on its side. So from your vertical layers into a horizontal flow, I want to look at from the user side instead of architecture side. Imagine like, this is how humans can feel when we put in a request in the chat window, and then something happens through this whole pipeline with AI, and then at the end, some response comes back. Or more complex situation, like you're a salesperson who have email exchanges, the more actions are taken, where does it sit? Each one of these steps.

I can take a very like high level first overview at it. One of the really important things that you're kind of calling out here on all of your risks is that there is a lack of understanding as to who is actually taking on some of these actions. There's no real like knowledge of what the identity is and there's no real understanding as to like What the default is that they have access to, whether it's a human or an agent. I think that that alone, like, that underpins a lot of these risks.

What is that identity? Is it identity of the person who initiated the call, or is it the identity of each one of the agents actually? are doing the work, you know?

Yeah, that's a great question. And I actually think it's both. I think it's really, uh, I think it's really important to note both the identity of the person that has kicked off, you know, the additional— the initial prompt. What we're seeing a lot of folks now though is that they have this idea of like they have swarms of agents that are running, right? And they have like Swarm 1, Swarm 2, Swarm 3. So it's important to understand first who kicks it off, who the owner is, but then it's really important to understand the identity Uh, of all of the individual agents that are doing something as part of that entire flow or process. We have seen various degrees of adoption and maturity for how folks have visibility into agents. Some folks oftentimes, especially in the early days, don't really have any visibility or there's no identity. If they are having some logs, it'll look like the, the person not only started the prompt, But did all those actions. It'll look like, you know, in this instance, it'd be like, hey, Angelina started this prompt. And then, you know, Angelina took all of these different actions and accessed this database and then, you know, ran this command and then compiled this. And it would just look like you the whole time, right?

If I work for a company and I'm using this powerful agentic systems and some decisions made without my permission, I would be really hesitant. Like, what if there's some consequence? And it's not something I can control. Do I take the accountability on the decisions?

Yeah, that's a great question. And some of the times, what you're triggering, the various tool calls that you're triggering, you may have inadvertently elevated, escalated access to some of the systems, right? So, like, if you were logging into your CRM, you would have a certain control, like, oh, I can read all these records, but I can't do anything destructive. And inadvertently, if you make this call through, you know, your preferred AI platform, you can have an agent take on destructive, destructive, you know, perform destructive things. You didn't mean for it to happen, but it'll come through and it'll say, hey, Angelina is, you know, Angelina just deleted all of your customer records in Salesforce. You know, like that's probably not something that you— Intended to do, maybe not even something that you thought you could do. You were just simply trying to, I don't know, clean up your forecast before the next call with your director or with your VP. What Speakeasy does, what it helps with, is it provides a consistent identity across all of your AI platforms. It's very possible, Angelina, that you today use More than one yourself. At the average company, we see a large enterprise has 4 different AI platforms that they use, that they officially use. There's a lot more that are being unofficially used at a lot of places, but there's an average of like 4 different platforms that they officially use. One of the really critical things is that we are able to provide consistent identity across all of those. So that if you're doing something in Cursor and all of a sudden you move over to Claude, that it's not like 2 different Angelinas, right? Claude Angelina can view everything, but Cursor Angelina can destroy everything. You provide one consistent identity across all of the AI platforms, and then you provide a consistent role, a consistent set of permissions, all tied to your identity provider. Whatever that may be. So now when you go off and you trigger an agent to, you know, just to follow on this example, hey, clean up my forecast. I have a call with my VP on Friday and, you know, take a look at all my calls in Granola and update next steps or clean things out that are closed, lost, whatever. When you go in there now, you can have some assurances that this agent is going to take on actions that are permitted, that are within the same level of access and permission that I have. And your administrators can also have some level of assurance that that is happening. Your administrators can also see like, okay, this was triggered by Angelina, but it was an agent that went off and did this. The agent has a specific identity. This is also important because sometimes you may want to create an agent that actually doesn't have the same level of permissions that you do.

Mm-hmm.

Maybe you want an agent that is just going through there and reviewing some of the PRDs or going through Linear and just, hey, here are all the things that got done today. And you want to just have it have— and you want to have that agent have lower levels of permission than what you have because there's no need. This agent is just here to let me know what's going on across my organization. There's no need for it to have any write access, any destructive access whatsoever. That is something that a lot of organizations don't have or don't do today. The best they do is that they tie the permissions of all the agents you create to you. The next level is let's have different agents that are being used or being created by folks within the company, and let's give them all different levels of permission based on the tasks that they're meant to have.

If I imagine like for a human employee, first of all, we have, you know, after the onboarding and the background check, we have single sign-on, right? We also have the role-based access control for as a human. I wouldn't be surprised that agents should have some level of that. Is that what you're talking about?

That's correct. Oftentimes we've seen, depending on the different solutions or different vendors that you use, There can be some disconnect between the access that you're supposed to have and the access that you have when you access a solution via the API. When you log in to a, you know, a product of your choice, you know, sometimes the UI itself handles a lot of the roles. Like, oh, hey, you can see these screens, but you can't see the other, you know, these other screens are for, you know, your manager or someone or an administrator. When you're accessing it through the API, we have seen that oftentimes that breaks down and all of a sudden you are given access that you didn't necessarily intend to have. A lot of this is just going through and going through this experience with large enterprises, understanding what different MCPs do, understanding how different tools work, understanding the identity providers that different people are using so that you can then properly map and then actually set out on rolling out AI with a level of permission that you're already comfortable with. You don't need to start from scratch. You don't need to say, hey, we know that leadership wants us to use this, and so we're just going to do it regardless of what the permission or security impact is. You can actually roll this out knowing this is leveraging all of the things that we have spent You know, years or, you know, decades in some instances setting up, like you said, RBAC.

For the identity providers you mentioned, are you one of them? I'm not aware of vendors who will give identity to all the agents generated within a company. How does that actually work?

That's cool. There are, I mean, there are a lot of big identity providers out there. Okta, there's Ping Identity, there's OneLogin, there's Entra from Microsoft. What we do is we partner, we work with all of them, and we do proper mapping for the users. When an agent is created, we can actually reverse skim. So when a new agent comes up, we can actually then go in and tell your identity provider, hey, you have this new agent, and make sure that they're able to properly catalog and track and be aware of this new persona. And it is kind of a persona. You know, I don't know about you, but some of these agents have taken on a, on a human-like level of visibility within our organization. There's a couple that we have that we joke around and we talk about them as if they are, you know, a person that we're, that's sitting next to a desk. No. Hey, you know, hey, did you see that thing that Nigel did? Like, oh my goodness, that is so cool. You know? And it's like, oh, it's, yeah. It's not a real person, but you forget that. They become part of the fabric of your company. They become part of your team. They become part of the thing that makes the culture, the thing that makes your company unique.

This is so intriguing because there's so many different sorts of agents. Who's responsible for these agents? The identity services you were talking about, Right. Gives different access controls and categorize each one of the agents that's being created within the company. But do they all trace back to the creator, the human? Like this human is in charge of managing all these agents. Where does the accountability lines end?

Yeah, that's a great question. And I still think that this is a hotly debated topic today because some of this is so new. There is a lot of responsibility within the creator, whoever kicked off these agents and whoever created it, but there's also a real responsibility within the company to make sure that you have the right framework so that not every single individual within the company needs to think about, you know, all of these elements the same way that not every single individual in your company thinks about, you know, am I using this SaaS tool in the right way that it's meant to? They are given a framework and allowed to be successful. They are set up for success. for success from the beginning. Who sets up this framework? That's also a big conversation. We are seeing oftentimes there are CISOs and CIOs involved in this. Sometimes it is a CTO, especially in a, maybe a mid-sized company that doesn't have a fully dedicated CISO, or when the CTO is not just responsible for future-facing roadmap and innovation, where they're also— in charge of internal IT type of things. We see those 3 personas come up a lot, but we're also seeing brand new roles be created. We're seeing, you know, folks like head of internal AI, you know, folks that are responsible for how do we run AI at our organization. We're seeing some folks from the identity space, whoever was previously managing your identity provider is now being asked to like now extend this And help us understand how we should think about agents, how we should think about AI. So I think this is still a fast moving space. Most folks last year were trying to figure out, you know, what is the value? What is the magic use case? Where are we going to use AI? This year everyone is starting, not everyone, but most folks are starting to really move into scaling this out. And so these roles and the responsibility and all the thoughts about this are starting to get figured out.

Tell me about scaling, scaling to hundreds, thousands, even millions. Everyone becomes a manager of AI agents. So how do we handle that?

Yeah, everyone is an architect now because we all have an army at our disposal. And again, going back to it for a second, that's why the framework is so important, right? You don't want every single individual trying to figure out how am I supposed to do this correctly? So when, you know, we are fortunate enough to work with a lot of companies that are starting to scale aggressively their use of AI. And what we've noticed is that the first thing that breaks when people scale their use of AI is often not cost, even though that's in a lot of headlines. It's, it's also not always security, even though that's something that we definitely help with and we support. Most of the time it's complete loss of visibility of what's going on. If you take a company with like 1,000 employees, just a, you know, maybe large mid-market type of company. And you say, hey, we're rolling out AI. Here is your shiny enterprise cursor and Claude license. You know, let's, let's go become productive. One person, maybe day one, one person will connect to their Salesforce instance and another person will build little agent to provide updates to the team. And another person will connect to an external-facing MCP, and another person will create an MCP for an internal database. You know, multiply that times every other employee, times another 30 days, and very soon you just lose track of who is connected to what and what are, you know, what's the permission they're supposed to have and where is data flowing and what data's going in there. Oh, what's going into these prompts? Like you, you can see how this can quickly really kind of scale out. We had one customer that said, hey, our sales team has set up, you know, hundreds of skills in the first couple of days that we give them a license where we don't know what these skills are doing. Like we actually don't even, we're not really sure that these are, these are skills that we want them sharing with other members. And so the first thing is really getting full observability over what is going on in my organization? What is being created? Who's creating it? What platforms are we using? We rolled out Cursor and Claude, but I see a lot of Codex going on in here. So the first thing is visibility. And I understand that observability is a very reactionary thing like that. Like that happens after something has happened. And a lot of folks in the space want something that's a little bit more proactive, which is definitely where we step in and help. But Just having that visibility in the first place is table stakes, you know, really, really having that understanding. So that is kind of where we come in and we really help folks understand like, hey, here's everything that's going on, on your local machines, on your enterprise accounts, on personal accounts, on local machines. We help you get a really good understanding of everything that's going on. And then you can take actions on that. Then you can stop Or block different things. But the first thing is like, you know, next time, Angelina, you're at a party with a fellow CISO friend of yours, just ask them, hey, do you know how many agents you have in your organization?

Speaking of this invisibility of what agents and AI are doing behind the scenes, even as a single user myself using Claude Code and all the other tools, I have no idea either. I have some visibility and created a little tool for myself. but it's not enough, right? I'm not going to know like everything. But the next question is, once I know what's going on, wouldn't I, you know, or a company like yours or your clients would naturally want to know, so what can I improve? Once you know what your employees are doing and then you want to know like, how do we improve it?

A couple of things. I would— I want to go back to, you know, your statement about like, as a single, you know, as a single user that you can see some value, because I think there's something that I've started playing around with as well, where I've noticed that models are, and I don't think this is that controversial. I've heard a lot of people talk about this. Models are becoming much more of a commodity. Every week or every day, there's a new model. There's a new platform that is more and more powerful, more and more impressive. It doesn't make them inexpensive commodities. Oil is a commodity that can be really expensive, right? It just makes it so that You know, the difference between one or the other can really change. And so I had, you know, my daily driver for a long time was Claude, and I had it, I had that environment set up with everything that I wanted. I had all the connectors, you know, set up in the right way, and I had my history and all this stuff. And then a couple weeks ago, someone said, hey, have you checked out the latest version of Cursor? Like, it's awesome. Like, I love all, you know, and they were really like amping it up. And in my mind, I was like, oh man, like, I just, I, I don't know if I can, I don't know if I can take the admin time to go and set everything else up in there. But one of the beauties of having a control plane, even for just, you know, a solo player or for myself, is that I was really able to just log into Cursor. And because of the control plane nature of what we provide, I'm fortunate enough that obviously we use our own product. Everything was already set up for me in Cursor. Like all the permissions, all the connectors. And that was like, okay, cool. Like I've really taken the overhead away of me being able to go around and test different solutions. Because otherwise that would have happened on, you know, if you're, if you're anything like me, that would have happened on a Saturday or a Friday night, you know, or something like that's, that's when that would have really happened. And I was able to do it just by logging in and saying, oh, wow, this is, this is great.

Wait, you're, you're saying that as an individual, operator, anybody can actually use Speakeasy to have visibility into what they're doing?

Today, we're mostly focused on large enterprises. That's who we work with. We will soon have plans for, you know, folks to be able to do this on their own. Me, again, because we use Speakeasy here, we dogfood very, very regularly. I was able to, as an individual, just kind of go from a couple different platforms. And that was a huge advantage. So anyways, I just wanted to call that out because that's something that I thought was, you know, as an individual, like a big boost for myself individually.

I don't know, you as an individual user also feel this. Did you notice that when you're working with AI agents that it will send you a little question and ask your permission for something and you will click on like allow this time or allow once, allow always, allow things like that?

Yeah.

Did you notice that this is very annoying? And because I feel me as a user or engineer, I'm rubber stamping on all of them pretty much. Do you really have time to review every little thing what agents are doing? And I'm just doing that at a very repetitive level and having couple of tunnels. I'm just kidding. I think I'm just exaggerating, but, but I do, I don't know if you feel that, you know, it's, it's an issue.

And I think I do, I do feel that. And the other day I was going through, you know, we, we, we just closed our quarter and we had like an awesome quarter. So I wanted to like get together like a really great message to the company. And I've spent a long time kind of tailoring everything and I used, you know, the connections we have to all of our different systems to gather some data. And I was just kind of clicking like, yeah, yeah, except, you know, I was just going through it. And next thing you know, I hadn't realized that I asked it to actually connect to Slack and like public— publish the thing. You know, I hadn't, I hadn't yet been totally done with it. Luckily it was as a draft, but I was like, oh, I didn't mean, I didn't mean to do that last accept. You know, I just wanted to like, I just wanted it to draft and then I, I'd make it human. You know, once I have all the data, I can make it really human and make it within my own voice. One of the, one of the cool things about, you know, getting to work with customers is that these things do come up and they ask for a lot of fine-grained control as to what each of the tools can do. So you can go through and you can say like, allow, never allow, or warn me about these things. The problem is that sometimes those can be too high level and it's asking you for permission for every single little thing that you do. With Speakeasy, we allow that real fine-grained tuning. So that I always wanted to tell me if you're going to like post something to a database, if you're going to, um, you know, do anything destructive, like I always want to know. And I want you to actually pop up at those times. So by having that more granular control, I know that I'm only being asked for permissions when it's something that I specifically wanted to be asked for.

Yeah.

So that's a little thing that helps. And then most of the time I allow things.

When you are onboarding a client company, right, that this company has to route all its traffic through your platform. I want to ask how heavy a lift is that? Is it complex?

Yeah, most of the time people that we work with already have existing MCP servers that they use, and then they can just, they can just proxy the traffic. So it's, it, we don't force people to redo anything. We work really nicely with the ecosystem that they're already in. But yeah, we route, you know, everything through an MCP gateway. We apply that identity, you know, to all the actions that are being taken within that gateway. The nice thing is that we're able to just do these things one time, you know, again, instead of having like every single employee within your company, you know, have to go through the entire process of setting up their connector to Notion, you can just do it one time. There are really nice things where like, I don't know if You've gone through this recently, but every once in a while your authentication expires and you need to go through and you're like, how did I do this last time? Well, this is a custom MCB and you're having to go through this for like 10, 20, however many connectors you have. You know, that's something that we can also take care of. You, you know, a user can authenticate into, you know, their, their preferred platform once and all of the connectors that their identity provider already uses can be propagated. So it goes from the first time you set it up, you probably had to authenticate to 30 different things or at least a dozen different things just to do everything that you wanted, to you just do it once and it takes care of that for you. It takes care of the refreshes as well. These are like the little niceties, you know, that like the quality of life things that like when you don't have it, you really are like, oh, this is— This is just toll. Like, I don't, I don't want to toil this.

I don't want to, I don't want to do this every 14 days.

Yeah.

Removing toll is the best use of AI.

Great.

Every part of our lives. What about, what about the companies building agents? Do you think, do you think they would think about, you know, stand up some pieces of, you know, like what you're doing themselves? Have you run into customers telling you that, you know, they have something like what you're providing half built and you know, they're going to be debating or they're thinking of, they're going to just build it from ground up?

Yeah, that's a great question. If people already have agents, we can work within the agents that they have. But I think you would be surprised at where the maturity is of most organizations. We do read up about a few organizations that are like really scaled out. Most folks are kind of early enough in the process where like, again, the last, you know, year or like literally last year, 2025, they were like Let's find a use for AI. You know, everyone was scrambling. What is going to make us more efficient? And this year they've been really like focusing on scaling it out. Some organizations do have crazy usage and agents everywhere and a lot of that stuff. I think most companies are just kind of getting to that stage. Like right now, as we speak, most people are like, okay, how are we going to do this? I do think that like, like all solutions out there, There are folks, a good number of folks that are wanting to set this up on their own. And I think that that's a good thing. I think that that's a good thing. Like you should try to set it up and get going. What people quickly figure out is that this is really a layer of infrastructure. It's not just another SaaS tool that you can, you know, vibe code your way to freedom. Yeah, this is actual kind of infrastructure that requires You know, real hardening, you know, that it's not just code, but it's also the maturity, the, all the calories that have gone into building and maturing this. That's like a really, really important part. Some folks, you know, I've already come back and they said, yeah, look, we tried to do this. It was great. You know, we had, we staffed up, we staffed up 15 people and they're still working on it. And like, we just don't want to. You don't want to own uptimes internally for some of this stuff, you know? So I think inevitably people build it. Some people are really successful. Some people have like the technical chops and they want to, this becomes like a critical part of, of their product or who the company is. But I think for the majority of folks, same way that they don't want to manage, you know, a cloud, you know, they don't want to manage this part of the infrastructure. It's, it doesn't really make financial sense.

Yeah. I will not try to build it from ground up if I have a product, you know, serve some different purposes. No way. So how did you find your first customers?

Uh, you know, I wish, Angelina, I wish there was like a magic, you know, answer to this. You and me, we would hit the road and we would make a lot of money if we could tell everybody like, here is the way that you find your first 10 customers. Because that is, that is, that is the question that has been around Silicon Valley and Forever. And I, I think a lot of it is you just gotta get out there. You can't outsource your first few customers. You can't, as much as you want to, you can't outsource the acquisition of them. They are friends. They are family in some instances. They are people that you meet through your networks, through your advisors, through your investors. They are people that That you have some level of inherent trust with, you know, someone that says, oh, I've seen your work elsewhere and I, and I believe in you. Or, or you, or it's borrowed trust, you know, like, oh, I really believe in Angelina. Angelina just spoke so highly of you all that we're willing to give it a shot because not everything is going to be that smooth. Most things are probably going to be pretty bumpy. But if the pain is big enough and if your commitment to the customer is strong enough, if you're reactive, you know, you will overcome all of those things. So how do you get your first 10 customers? You literally knock on doors and try to get people that believe in what you're doing, that have the pain that you have, that have some trust in you, whether directly or indirectly, and you do everything you can. To make them successful enough that they're willing to give you some money. That's, that's the way you get started.

Is one of your magical questions when you're finding customers is what you just shared with me, which is ask a CCO of a company and ask them, do you know how many agents your teams are having?

Yeah, you always want to have, every organization needs to have like one or two questions that, you know, you kind of know the answer to ahead of time. Those are the things that really spark a great conversation. And it may not even be necessarily like, oh, like, you know, let me sell you something. It's just really genuine curiosity. And those are the best conversations when you're like, hey, this is something that I deeply care about. How are you thinking about it? And sparking it. Oh man, those are, that's where you learn the most.

You have such a great question, like conversation opener that I can take out and ask anybody, even for myself, like, you know, looking for podcast guests. And that's a great opener as well.

You know, take, take your internal passion and sprinkle a little curiosity in there. You normally come up with like a really good way of like opening up these conversations. Love it.

Love it. How did— how about your first employees, your first, your early teams? How did you find them?

Network. That is like, that is just like, who are some of the people that I am just in awe of, you know, and, and how can I convince them to come work with me? You know, like, how can I like, get them excited enough about the problem that we're solving. But that your first few, I mean, I think you've had guests on recently that said, hey, you know, the first 5 employees you hire, they make the whole company. Like that is the core, you know, that's where everything starts from. And I violently agree with that. You know, I think that you have to, you go out and you absolutely find like the people that you admire, the people that you were like, This person is just truly amazing, and I want to work with them. And then you just need to figure out how to convince them to join you on this, like, really crazy journey.

That's the million-dollar question. How do you convince them and motivate them to join you?

Yeah, well, I mean, I think if you come at something with a lot of passion and a lot of belief and conviction, I think that's step one. You know, everyone will go through these things in different ways, in their relationships, in their way of, like, selling the idea of their company or just getting people excited about it. They'll do it in different ways. But man, you really need to have strong conviction and excitement that like, we're doing something special here. At the end of the day, I've found that some of the most amazing people I've ever worked with are more mission-driven than even financially driven, right? Obviously everyone wants to have all the resources they need to live the life that they want. And we here at Speakeasy want to enrich our employees. You know, we want them to be incredibly successful and have all the resources that they want to live a happy life.

Yeah.

But some of the absolute greatest folks I've ever worked with are just as deeply passionate about the problem you're solving, right? Or about like the mission or about like, we are not just another company. Like we matter. Like we are doing something here that's really important that is missing in the space. And we believe that we can do it better than anyone else. We found a really large total addressable market. We found a gigantic total addressable market. This is our at-bat. And man, we are so excited that we get an at-bat. How many people, Angelina, you and I know go their entire careers and they just never had that at-bat? They were incredibly smart and ambitious. They were just, they never found the right time with the right company at the right size to really fulfill their potential. You know, we believe that this, that we're in it right now. Like we are up at the plate, what we do with it, that's totally up to us. And that's exactly how we want it. You know, we want it to be up to us. We want to have everything there set up, a large market, a great product, a strong customer base, you know, in a space that we're really passionate about. Now, what are we going to do about it? How are we going to actually, you know, impact, you know, the marketplace? What are we going to do with our opportunity? That's all you can ask for.

Is there anything that would make you quit working on this? Or has it ever occurred to you, or have you ever swayed?

I don't think as a founder or as a leader of a company, you can, you can go down that path. It's just, you know, you, uh, there was a, there was a one, someone that I got a chance to work with. A few years ago. His name is Joe T. Bonsall. He's the founder of AppDynamics and the founder of Harness and Traceable, just very successful. He has his own VC. And, uh, we were going through this whole product market fit process and he'd been a founder multiple times. And, you know, I was kind of struck. I was like, hey, like, God, Joe, how do you see it? Like, how do you, how do you know what's out there? Like, I'm like, you know, I'm having these conversations and I'm not, I'm not seeing the pattern yet. You know, give me some, give me some belief. And Jyoti said, Hector, you know, we're gonna, we're gonna find it. We're gonna figure it out because there is no other choice. I choose to be successful. There is no other choice. So we will figure that out. And I always thought like, wow, like he just, he, it just, it doesn't, there is no other option. Like you just, he believes in his team. He believes in who he hired. And he just knows that there is, it's not a matter of if. It's just a matter of when, you know, you got to wake up in the morning, you got to put one foot in front of the other and you will find it. If you have the right people, you will find it. I think it's the same way. I don't really think about like, oh, like it's, I don't think there's anything that would sway me right now. I don't think I have so much conviction over this space. I don't think there's anything that would move me. I am, I am as excited or more excited than I've ever been about the space that we're in, the problem that we're solving than anything I've worked with in my entire career.

I want to ask you about teams and more teams and career advices for engineers, especially. Has AI changed how, how you plan your, your teams, like how big they need to be and who do you hire?

In some instances, a lot, but we have people that are, again, they're like architects. They're not just individual contributors. They have, we give them big, crazy, gnarly, ugly problems. And that would normally take like, oh, let's, you know, let's set up a team and let's get a kickoff. And like, you know, this is our squad. We throw these people the hardest problems. And then 6 months later, they come back with a solution. Now we just have one person, right? Like, hey, how do you solve this problem? And like 2 days later, they've not only solved it, but they've like come with a fully production-ready problem or maybe not production-ready, but a pro— or a solution that they're ready to show off to customers. Which I think is just really amazing. We can do so much more, but that does mean that we vet everyone that we hire, uh, even more because their impact is outsized. We are the type of company that we take a really long time to hire. We really want to make sure that it's a good fit, not only for the role, but for our culture, that it's a really good fit for the direction in which we're headed. But But once we find our people, once we find that person, we give them our full trust. They are innocent until proven guilty. And it's okay if they stub their toes and they make some mistakes. That's okay. It's a very high autonomy place. We took a long time finding you. Now we completely believe in you. And we throw them really big problems from the beginning.

What's your criteria of hiring? What kind of people are you looking for?

Uh, we, uh, we want people that have high agency, people that want to, you know, there's 2 things. Collaboration is incredibly important. We need to have people that can collaborate, but that are not afraid to take on like big meaty tasks. You know, we have, we've hired someone that their first day on the job, she was on the conference floor pitching the product. That was like first day, you know, we've had, we've almost everybody, almost every engineer who ships something. The first day that they're on the job. We've had people run proof of concepts with large customers in their first week. We need people that are not afraid to take on a big gnarly task. That is really important for us. High agency, people that are okay making mistakes and have the resiliency to move forward and move past it and not just are like, oh, Got it. You know, not, we can't lose people for days or weeks at a time. We need that resiliency in there. And also look, Angelina, like, you know, we're, we're still a young company. You know, things change very quickly. We need people that are comfortable and can embrace that, you know, that really love the fact that we can change direction without having to go through a lot of different, you know, you know, a lot of red tape or a lot of different meetings about why we're doing something. We'll communicate, we make the change, we communicate it. And then people that really thrive in that, That's really important for us. One thing I do, you know, your earlier question said, you know, what are some suggestions or any advice you have for young folks or for engineers? I think the thing that I want to tell that I would love to be able to just, you know, tell every young person coming out of school or every young developer or anybody that's going into our industry is how important personal relationships are. I tell everybody like, go talk to people. Like, if you want to be in this space, go talk to people. It's okay if you're shy. Everyone else is shy. It's totally fine.

Like, they—

it's expected. You don't need to be the life of a party, but like, you, you need to meet people. You need to get out there. You need to shake hands. You need to have conversations. You need to be curious. You need to share your thoughts. Go to events. I, I am here in San Francisco. I know you're, you're, you're here in the South Bay. Yeah, I'm in the peninsula. And There are so many events here. We are so lucky. It is, it is wild. The other day I was at, I went to an event during lunch or in the middle of my day. It was like on a Wednesday and I saw Brett Taylor talk about the way that they're scaling Sierra. I was like, wow, that's, that's amazing. You know, my, uh, Sagar, my CEO and co-founder was on a walk the other day. He's like, hey, I saw Sam Altman down in Embarcadero during a one-on-one, right? Like, obviously he didn't, he didn't stop and like ask him any questions, but just like Being around that energy, being around that, I think is so important. So I always tell people, technology is a bigger part of our lives. If we choose to, we could be more disconnected than ever. And that is why you need to be more around other humans sharing those ideas. Because the one thing that is becoming a commodity for sure is your ability to like, you know, create a great product or a piece of collateral, all that stuff. There's so much like, you know, there's so much slop that's being thrown out there. Quality and trust are at an all-time premium. You know, by meeting people, by just knowing, I know, and you know, we've, we've shaken hands, like we've, we've broken bread together. That trust, you will take that and, and you will have that as part of a relationship forever. So when young people are telling me like, I'm having a really hard time getting a job. Okay. What are you doing? Well, I've sent my resume to 500 different companies and no one's responded. It's like, okay, that, yeah, like that kind of makes sense. Like, have you, have you gone out and have you met with people? Have you tried to like get someone that you know to maybe make an introduction? And it doesn't need to be an introduction to their company. It can be an introduction to someone else that's in the space that is doing the job that you want. And it maybe that doesn't lead to anything, but at least it's someone else that you know.

Yeah.

It, at the end of the day, go out, meet people, network. Build trust. I think that that is absolutely the most important thing that anybody can do, especially with AI. That's what I tell my kids to do. Go out, meet people, build networks, make friends, not just because you want something out of it, but because it is a genuinely wonderful human thing that we get to do. And more often than not, it does become something that you can share that relationship down the line in a wonderful way.

Are you fundamentally optimistic or pessimistic about the future of AI and why?

I'm an optimist. I'm a founder, Angelina. You have to be an optimist, right? I always think that there are problems that will come up. There are problems that can be really scary, but I think as a society, you know, whenever we look at tackling a problem, we'll be able to find a solution. I think that there is I think that the best days for all of us are in the future. I think the technology that we are building today and the way that we're using it are going to lead to richer, better lives. We can't let everything be unchecked. Obviously, there does need to be human involvement. Humans know humans better than any— than anything else. But I fundamentally believe That we have a wonderful future that will result in richer lives for us and our children.

I'll close every conversation with this. If you could recommend me and my audience a book, what would that be?

A book. Yeah. So like a lot of people this summer, I got the soccer bug or the football bug. And so I started reading a book called Soccernomics. I actually ended up loving it. You know, I started reading, I was like, oh, this is kind of interesting. And the reason why I like it so much is because it's so far outside of what I do every day. You know, it's like very, really different from like my day-to-day, but there's so many learnings in there that you can take into your workplace about how different industries work, about really how the world functions and all of these inefficiencies that are around us in so many different ways. different ways. And I love learning about how other industries end up tackling those things. And I can reflect upon my daily life and about my role. So I, you know, for anyone else that got the World Cup bug and wants to know a little bit more about the world of soccer, I think that that was a pretty fun read. Yeah.

Thank you so much. Hector, really nice chatting with you today. I can feel your passion about, you know, not just your work, but also the idea of how should we prepare for the new era, you know, which is not maybe stay at home and continue to code and stay in the office and write coding, but to meet more people and maybe let these organic feelings and organic chemistries to spark. And you don't know what, where those things will take you, just like how this podcast is happening. And I feel tremendously honored to have you on today and share your thoughts and your passion. Thank you so much, Hector.

Thank you so much, Angelina. It's been wonderful.