The Neo Show: Founder Fridays #18: Lexi Reese CEO, Lanai with Host Howie Xu

Can you guys hear me? What do you guys call yourself? Nortonians or? Genovators. All right. Genovators. Excited to be here. Thanks, Howie, for having me. So, Lexi Ree, I'm the co-founder and CEO right now. We organize AI at work and make it visible, safe, and impactful. We'll get into what that means, but 30 years operator, Google, American Express, Gusto. I was a COO at Gusto when it went from 100. Hello. Okay, that feels a little better. 100 people to about 3,500 people from our series A to our series D. So longtime operator, but what Howie was referring to, I spent a year and a half between when I was COO of Gusto and when I took a next operating gig as an EIR at an investment firm called General Catalyst where I was examining what does the future of work look like given big tectonic shifts like AI and the labor markets like climate, and I was developing the firm's thesis on where we should invest given those big tectonic shifts. And I had two ideas for companies coming out of that. One was Lai, but the first that I pursued was to run for the United States Senate for Diane Feinstein's seat. And that is much more like a startup than people can understand, but you are the product. You're kind of taking yourself to market and your features are the policies. So happy to go into any bit of that.

Yeah, sure. So for this technical audience, the shorthand is it's DataDog for AI interactions. But I'll explain more what we work with customers on. We sell to mid-market and enterprises. And most companies at this point have said something like we're going to be AI first, just like they said we're going to be cloud first back in the day or mobile first or digital first. And every technology wave introduces a new set of data. And in this case, the data is the interactions or the conversations employees are having with AI. The majority of which are invisible to IT teams or security teams or data teams because these are conversational by nature. So traditional tools, I'm sure you all are catching up on this, but generally speaking, there isn't a set of technology solutions that help people in a non-creepy, not Big Brothery way see what are employees actually doing with AI. And I really view this as AI is less a set of tools and more a set of capabilities that have been added to every team, sometimes through the tools that your IT team bought as a top-down procurement initiative. Here's our version of ChatGPT. Here's our version of Copilot. But oftentimes through what employees at every level find on their browser as something that is helpful for them to do their job every day. So if you think about that, those conversations, that's millions and millions of pieces of information that if you want to be AI first, you need a system of record to tell you who's using AI in the company, why, how risky is the conversation itself, and then most importantly tie the usage to impact. Is the fact that 80% of people are using X AI tool every day translating into real different and meaningful results that justify your investment? So in short, we are deployed as a browser extension right now and we've created the first-of-its-kind detection mechanism that creates telemetry and defines that an interaction is an AI interaction when it's happening between an employee and any AI capability anywhere they might be accessing it. And then we capture the prompt and label the use case and the risk level of the conversation itself based on the atomic unit of the prompt. And to date, we think we're the only people who have this AI-native observability platform that helps people again see all of the AI that's happening in their company to organize it and make it impactful.

I knew you were going to ask me that because I listen to all the podcasts.

And I haven't nor have I come across. I've had say 300 conversations in the year with companies that I think are your ideal target and I have not heard NEO come up.

Because of my LinkedIn post.

Okay. Yes, the podcast. You have seen the podcast, but you haven't installed Neo.

You got to pitch me, Howie. But you got to pitch me on why I need to try it.

So people install a browser extension and then you are able to see all the conversations with ChatGPT, with Glean, with everything. How does that work? So, I'm not going to go into the technical specifics because we're seed stage in beta with customers, but I will simply say the following: when a CEO says we're going to be AI first, there is some poor schlub that has to manage that mandate. So, I feel like we're in some blind AI mandate era. So if you're the CIO or the chief data and analytics officer, you're on an AI council with all of your colleagues and there's a portion of your colleagues that want to accelerate innovation. We want to use as much AI as possible. There's a portion of your colleagues that say our job is to mitigate the risk of AI. And so we are partners to that whole group of people that basically have said AI is moving too fast. People are using too much of it. It's proliferating too quickly and too heterogeneously for the way we bought applications in a SaaS era and the process by which we approved those applications and trained people to use them could possibly be relevant in the AI era. So you have had a bunch of AI security tools pop up and say we're basically doing DLP for AI and the problem we're solving is that data might leak through a prompt. That is a use case that is a function that could happen as a result of shadow AI. We're much more obsessed with how do we keep you current on exactly what people are using, when they're using it, what teams are using it, and helping create a taxonomy or a new semantic language for you to say, I don't just need to know usage for usage sake. And all the security companies are going to catch up with data leak detection and data leak prevention eventually. The real thing is how is this new capability changing how my employees get work done? And so what we really had to spend some time on, Howie, to kind of come back to your question without fully answering it, is how technologically do we do real-time detection of whether something is an AI interaction versus all other outgoing web traffic with the design constraint that we don't want to exfiltrate any data because if we exfiltrate any data to make that detection, we're expanding the attack vector of a company that is desperately trying to reduce their attack vector. So that was the technology challenge of detecting AI without a static list. So lots of people, you'll Google and I and others will come up similar, but usually how anybody in our zip code that's doing this is taking a static list of apps that the IT team says I want to know if people are hitting these apps, but that's so beside the point. And I can give you examples of what really matters is surfacing the workflows. And that is why our customers are psyched because we're not just giving them, you know, this data leak, this data leak, this data leak. Though we do that and we help people prevent that. We're saying these are the jobs to be done with AI.

So in the product itself, the thing that we help with directly with respect to true return on investment is we will show you, let's just use the majors, but let's say ChatGPT. You bought an enterprise license and we'll show you that 40% of people have used it since the licenses were purchased. Now that's not revolutionary. You could also get that from their own vendor report. What is revolutionary is that side by side that view you can look by team, by tool, but also by use case to say what are people not using AI at all, or are they using AI for the reasons that we bought ChatGPT? They're just not using ChatGPT. So like a big win for our customers has been, oh, we're going to deprecate these licenses, kick off a third-party review of what was previously an unsanctioned tool, which you can do in our app. You can say, I see DeepSeek. We saw DeepSeek well before anybody saw DeepSeek. The minute that somebody prompted DeepSeek, you could kick off a third-party review and then sanction the tool. And then you're doing real dynamic procurement of what you think really adds value. Now the interesting thing that I think we support operators to do, and again my background is 30 years as an operator, so I've sort of built this in part by how would I be managing this when I buy any AI capability for marketing or sales or product. I would run an A/B test to say what is the team that is using this functionality able to produce, and I would look at the quantity metric, velocity, but also the quality metric, whatever that is. And I think so many people are looking at did we ship faster, are we answering tickets faster, but there's not enough attention being paid to the counter metric of is the quality of whatever you shipped or the quality of that interaction with a customer staying at or increasing relative to a control group. And that's truly, I mean think about it, that's the only real way you could measure the impact of any new thing in your dynamic system of any function.

Yeah, I think that this is fascinating. You know, when I was at the choose your own adventure of I did not find product market fit with my Senate run. I couldn't acquire voters. I didn't raise enough money for the cost of acquiring voters. And that's really the way to think about it. All politicians are going with a double-sided marketplace. You're raising money from funders to turn around and acquire voters. So same as a startup, but the choose your own adventure at the end of that was do I go be a COO or a number two at another high-flyer unicorn, or do I start my own thing? And either being a glutton for punishment or just really interested in learning from the ground up AI, I chose this path. There's nothing we don't use AI for, is the bottom line. And the way that I'm building Lai is fundamentally different in construct than how we built Gusto. So we have 14 human beings at Lai, but we really have 21 FTE equivalents. And why? Because I have set up AI in a way for product marketing, for revenue operations, for legal, for finance, and then for some core engineering use cases. We utilize AI for that. Now we have the benefit of being a startup, so we can set this up from the beginning. Sorry, you said 14 humans and then 21 including the humans? Including using AI. So I think of AI more as tools, like a seven FTE is the virtual FTE, right?

I could make up a number. I'm just saying what it actually is. Right now as a startup with very specific tasks, we have to... And give one or two concrete examples of those. What does the virtual FTE do? Sure, a full marketing analyst that is doing all of my go-to-market content. It's as if I have a junior marketer on my team. I'm managing that marketer. That marketer is just Anthropic, but it's a very specific project that's been set up in a specific way for me to be able to ingest content of customer themes and turn that into market-facing collateral and to be able to do that with sight, sound, and motion. I can do it forever. Is anybody in marketing here or sales? Yes, we have marketing people here. Okay, that does not bode well if you don't have anybody here. But it's really tricky. Oftentimes it's a translation effort to say what is the ICP? What is our true ideal customer profile? What is their true pain point? Where do we truly deliver value? How do we communicate that externally? How do we balance a roadmap product marketing between what we say we do and what we actually do? And so just think of all of those discrete jobs to be done are done in AI. And that's me expanding my team. And I could give you that exact same breakdown for the roles that I mentioned.

It's hard to say, Howie. I would say at least one, possibly two, depending on.

I would have, if we were doing this company when I joined Gusto in 2015 without AI in marketing, right now I would need two human beings doing the work that I right now can do with AI. That's very literally what I mean.

No. And this is why I think the valley needs to really contemplate the impact of the technologies we're building on the labor markets. I would not need to, nor would it be acceptable by investors, to hire that many people for large functions because AI can handle much of the work that we were needing people to do. The work that was repeatable, that was volume-based. So, if I needed, let's say, 14 entry-level customer service folks for every 100 customers we brought in, or 500 customers we brought in, that ratio has gone so far down because if you build your product right, if you build a great technology product, the biggest thing you're selling to customers is time. So if there's a flaw with your product that they have to call you or reach out to you, that should be in great technology companies treated as a defect. You add humans when you absolutely want to cultivate a human-to-human relationship. But we all know that technology in the past has not worked that way. So we've had to have a lot of people selling and servicing and even first-line coding products, and now AI can really and truly handle that. And we can get into why this is so different than the digital revolution in terms of what its impact on the labor market will be, but...

Well, it's interesting. We have a comparison. So 70% of everybody we have is engineering. So it's mostly engineering. So again, that ratio would have been different. Let me just take AI security startups that started a year or two years ago. They have double the team size that we have, literally double or triple the team size, but as much revenue. I mean think about that. We started in 2024. Some of these companies started in 2021. We're that far along with our product and our go-to-market cycle that competitors who started a year earlier or two years earlier, even in the window of let's call the AI era, it really started for most companies after ChatGPT reached 100 million users. So let's say that's January 2023. Like in a year I can get to the same place as competitors, and that at half or less the size.

I mean, there's probably lots of reasons for it. There's a philosophy. There's some ways we're going to market that are different, but I would attribute a big portion of that capability to what we can do with AI.

Well, for AI, because we're again an AI-native company, so we have our own small models that we've built. But we again, I'm not being cryptic because I'm trying to be a jerk. I'm just saying we use a lot of tools and have standardized on the ones that we think are the most efficient. And unlike what is popular in the valley right now, which is margins don't matter, don't worry about margins, margins are cute, we actually do think about margins. So that factors into our decision.