Base44 创始人 Maor Shlomo 谈 Vibe Coding 如何杀死 SaaS

Of all the things in the industry, I think margins are the least thing that I'm worried about. We're taking into consideration that the prices of models will go down toward zero. Personally, I don't believe we'll hit a speed bump. I don't think LLMs will hit the wall like OpenAI's LLMs.

Even with existing models, we're only scratching the surface of the economic value that you can create. If a one-person team can get sold for $80 million without people, it means that you'll be able to do a lot more stuff with fewer people.

Ready to go? Maor, it is great to have you on the show, dude. I've been so looking forward to this one. I'm a fan of Base44. What a freaking journey. Thank you for joining me.

Thank you for having me. I'm a fan of 20VC.

1. The Decision Behind Selling Base44

Dude, that's very kind. Listen, ego flattery will get you everywhere, and I'm very susceptible to it. I was chatting to Michael Eisenberg before this show, and he said, “Dude, you've got to start with the fact that it was acquired for, like, $80 million bucks.” Why did you decide to sell, before we get to anything else?

I started Base44 as a bootstrapped business after doing a very capital-heavy one, which we'll probably get to in a bit. It started honestly just for fun. I wanted to get back to coding. I felt like everything in the software industry was changing, and I loved the fact that with LLMs, we can do so much more.

It started just for fun, and then the traction was insane. I started seeing 10,000 users, 100,000 users, and the business was very profitable. I stood at this junction of, “Do I continue doing this alone, bootstrapped thing, which I might get eaten by companies that are raising a shit ton of money?” Or should I raise a lot of funds, go very aggressively, and try to build this on my own?

The third option was getting to a company that could be a parent company and trying to find something that could harmonize with the business and actually push it forward. I remember when I started having conversations with Wix, I was very worried, and I thought, “I'm not going to sell.”

I remember telling Avishai, the CEO of Wix, “If Canva was sold to Microsoft, there probably wouldn't be any Canva today. The fact that they were independent and could build the business by themselves, the way they wanted, without the corporate stuff, has gotten them far.”

We had many late-night chats, and we understood that Wix, for a lot of reasons that many who know Wix could understand—and I'm happy to tell you about—is a perfect match because we're targeting the same audience. It's the same spirit.

The marketing team is, in my very non-objective eyes, the best in the market. They're aggressive, they're creative, and they're absolutely awesome. If we can structure it in a way where the company itself—the product team—stays lean, small, and startup-y, while we benefit from the monster that is Wix for all the other stuff, this could be a lot of fun.

For example, the support and customer care, the marketing, and the things that you don't really want to build, like the legal and operations stuff. We structured a great deal, and I think to me it was like, “Okay, we have a shot at actually building something that matters.”

For an actual shot at taking this from a low few hundred thousand users to something that might, if we work hard throughout the years, actually move the needle for a lot of people—something that people might speak about as, “Hey, this actually made an impact on the world”—it was either I raise a lot of money and funding, or I go and join Wix, which basically triples my chances of actually building something special.

Today, Base44 has continued to thrive and is now over $100 million in revenue. It's done so well post-acquisition that everyone's going, “God, does he regret selling now?” When you look at how fast it's grown and the fact that it's now over $100 million in revenue, do you regret selling so soon?

Absolutely not, for all kinds of reasons. I can't comment on the revenue stuff in a public company and so on, but, one, we structured the deal in a way that even financially, I'm compensated for additional revenues, targets, and milestones we're hitting.

Second, this was the whole point. I felt like, in my eyes, this was going to be the largest category in the software industry because a lot of existing categories that we're familiar with—CRM, support systems, task management, project management, and stuff like that—are going to fold into the vibe-coding category.

There will be a time—it might be in a few months, it might be in a few years—when it will be easier to build your own Salesforce-type CRM than to actually buy a license to something off the shelf. The whole point of partnering with Wix is to take a shot at building something that would be huge. Without it, my chances would be really, really small. With it, they're much bigger.

With the greatest respect, that's wrong. You're not going to have people build their own Salesforce CRMs on the whole. Maybe a tiny, tiny portion will build their own Intercoms, Sierras, or Decagons, and then do maintenance and updates. Are you serious?

I honestly think so. It might not be the current way that we think about vibe coding, where you prompt, “Hey, build a CRM for me,” and then you get a fully fledged product with all the features inside.

It might be that software becomes more liquid. Maybe you have a starting point, templates, or open-source projects that people can take and then vibe-code their own features into. They might say, “Okay, this is great. Now I want this version in Arabic, from right to left, and I want to be able to add my own pictures of the lead to every lead,” or whatever. Then you basically convert and transform the software into something that's more what they want.

I think the current model of one-size-fits-all doesn't make a lot of sense. It's not that everyone will move there, and it's not that enterprises will move there quickly—it will take a lot of time—but it makes so much sense for the buyer. Think about it as the software being theirs. The code will be theirs. They will not need to share their data with other software vendors and be locked in or whatever. They could adapt it and create a leaner version that's way more customized to their needs.

I'll go even one step further. I think every software company that needs implementers, whether it's a Salesforce implementer or whatever, maybe a decade from now, is not going to make it.

A year ago, vibe coding wasn't even a thing, and building software wasn't even a thing. Maybe it was a nice toy; maybe you could build a small task-management software product, or maybe you could build a nice website. In a year or 2, you'll get to a place where a lot of the current organizational tools that companies are using could be built as your own version or customized to your own version, and it would make so much more sense to you as a buyer.

The code will be yours, the data will be yours, and you'll be able to adapt it to your needs. There's no one-size-fits-all. There's no feature bloat. When you buy task-management software today, it has so many features that you don't really need.

Do you think small businesses give a shit about that? Do they give a shit about actually owning their own data and owning their own code? If you're a small coffee shop, coffee chain, restaurant chain, or legal and accounting firm—and that's the bread and butter of Wix's business—do they need it?

Not necessarily the code and data, but they need something that's simpler and more customized to their needs. You know how I started Base44? My back-then girlfriend, now wife, has a small business. She needed a CRM and some systems for her small business. She's a tattoo artist, and she teaches art and stuff like that.

We tried building that with the former version of—I’m not going to name the software vendor, or the 2 software vendors that we tried doing that with—with the drag-and-drop and the super-customized, build-your-own tables or CRM. It was hell on earth.

I'm a software person, right? I should know how to do that. But just configuring it and making it support her process of how she processes leads and customers and whatever was insane. She's like, “Ugh, this is insane.”

I was like, “An LLM can code something leaner and way better for what she needs—exactly what she needs. It's not going to have this entire feature set, enterprise features, team-management stuff, and whatever settings. It's going to be super simple for what she needs and could adapt to her needs.”

I knew LLMs could do that. I just needed to provide LLMs with the right setup. They can write the code, but she would need a database to keep the leads, and she would need integrations for sending emails and stuff like that. That's why I started Base44, exactly for that.

I think small businesses specifically would want to build something leaner and more customized to what they need—something that they can understand and adapt. Bloated SaaS from the pre-2022 era is going to be harder for people to use.

Will Salesforce be bigger or smaller in 10 years' time?

It might. Mark Benioff is a smart person, and I'm sure that they'll still innovate. The company is incredible.

So they might adapt. But I think generally, companies from this category would be smaller. If not, some of them would be eliminated.

Which would you say is most likely to be eliminated?

Smaller CRM companies: every piece of software that's basically a front end on top of a database, with some team features and some integrations, and does not have a moat. I think Salesforce is so big and so massive that this is one of their moats. They can make big bets, move a lot of stuff, and go in many different directions while we all try to adapt in this AI era. But smaller players, I think, would be eliminated.

2. How Base44 Thinks About Defensibility

You said the word “moat” there. It's often a criticism of this category. I've had Marc Benioff of Salesforce; Mike Cannon-Brookes of Atlassian; Matthew Prince of Cloudflare; Dylan Field from Figma; and even Carl Pei, the founder of Nothing, the phone company, create their own Base44 equivalent. Everyone says this is a business with no moats, and you see that in the ability for these people to spin up products in 3 months. How do you respond to that statement?

I think it's relatively easy to create a vibe-coding tool. It's very, very, very hard to create a platform that could help people build products that people actually use, that are functional, and that are complex enough for real-world use cases. For that, you need many layers of integrations. You need to adapt and tune the agent to handle very complex projects.

I think some applications in Base44 are millions of lines of code. Those are eventually the customers and users that you care about, because those are the ones that will go beyond the hype of, “Hey, look, I've built this monday.com clone with a single prompt.” No, you didn't. It's going to take you hundreds or thousands of prompts to get to something that you're going to use day to day.

I think the moat in the category is being able to get there. That's layers of integrations, heavy lifting, and building compute. You're almost building a small cloud, because you want to give the LLM access to databases, user management, built-in integrations, and the ability to run scheduled tasks, because that's what real software does. You want to have the LLM access many cloud services and integrations, so you're building an entire ecosystem in this platform.

But to build a vibe-coding tool that can build some websites or a front end that could connect to some APIs, I think it's super easy to do nowadays.

Does that not take away a lot of the low end of the market there? If you think about that, essentially, if it's able to build good front ends, good landing pages, and good simple projects, then it basically leaves the higher-end champion users for the more complex service providers like you. Does that not eat away at and cannibalize a lot of the market?

I think it depends on what you call the market. For Base44, we almost don't focus on websites, which also goes really well with Wix, right? Wix does a lot of websites, marketing websites, e-commerce websites, and stuff like that.

For Base44, our laser focus, our core focus from day one, is complex applications and functionality. If you were looking to build a marketing platform where AI agents can constantly generate content, read your LinkedIn posts, see the traction that you're getting, and from that try to digest it and then fill your calendar—different functional, complex stuff that you use day to day—I think that's where we focus.

Early on, the vibe-coding category was focused a lot, and still is to some degree, on front ends and landing pages. Many use cases were around websites or e-commerce websites and stuff like that. I think this will get commoditized.

For the category, there's still a lot to do when you're looking at SEO and domain management. If you're building an e-commerce site, that's a complex business. You want to integrate with a full back end like what Wix has or Shopify or whatever.

But for the functional stuff, the complex applications, the organizational applications, the personal applications, and the full-fledged SaaS platforms that people are building, I think they won't get commoditized. I think that's not the long market. Also, I think those are the more profitable use cases for tools like ours, because at some point, okay, you're building a website—let's say a nice website for your VC firm or whatever. At some point it's like, “Okay, this looks pretty. That's it. I'm not going to prompt anymore.”

A lot of the business model of this category is around credits for prompting. At some point, you're like, “Okay, that's it. I'm done. I have this front end. It probably costs $30—not a lot—to host it somewhere.” I think the more profitable stuff, with the higher margin and the more value, would be building the complex applications that you'll use day to day.

3. Mapping the Future of the Market in 3–5 Years

When you map out this market in 3 to 5 years' time, can you help me understand what that looks like? Because you've got, in my mind, 3 different entry points. You've got the Base44s, the Lovables, and Replit in one. You've got the Figmas, the Cloudflares, the Atlassians, and the Salesforces in the incumbent insertion point. Then you've got the Canvases on the other, more consumer end. I'd probably put Figma in there with Canva. What happens here?

And then, sorry, you've also got Cognition, Cursor, and the pro tools there. Do the pro tools come down and eat your business? Do you go up and eat their business? What happens there?

It's a really good question. I'll start with what you categorize as incumbents, but I'll actually call them the different software categories that exist. Let's take Salesforce or monday.com, for example. I think a lot of those tools will build built-in vibe-coding tools inside their platforms.

You can see that with monday vibe, which is a tool they've built on top of the monday.com infrastructure of boards, the database behind the scenes, user-management tools, permissions, and so on. This is what I meant when I said software is going to become more liquid.

I don't know if Salesforce is planning to do that at all, but I'm assuming that at some point you'll be able to create your own user interface for Salesforce. Salesforce would become this system of record where you can create the UI that you want and how you want it to look when you're looking at a lead, contact, account, or whatever.

A lot of tools are going to do that, because instead of configuring different things and building different features, it would allow a lot of extensibility to existing products. You start seeing that, and you also see Microsoft building some vibe-coding tools on top of Office and whatever. I can see a future where a lot of the existing SaaS tools and SaaS products have a built-in vibe-coding, customize-the-software-to-your-needs capability.

I think it's an interesting take on what's going to happen with the pro players versus the very consumerish AI app builder, or visual AI app builder. For Base44, for example, from day one the objective was that you wouldn't need to see a line of code. For a lot of our users, code actually—I wouldn't say it terrifies them, but it's not what they want to do.

You're a lawyer and you want to create some system or software for your law firm. There's no need for you to see code. You should be able to customize and do whatever you want just by prompting it.

I think Cursor, Windsurf, and Cognition are doing a good job focusing on the existing developer community and target audience. There's a lot to do there, because it's going to take a lot of time for tools and platforms like Base44, Bolt, and Lovable to get to a place where you can build whatever X percentage of existing software that we cannot build today.

I'll give you an example. It's very hard to build a firewall today in Base44, right? Something that will sit on your computer, be very light in terms of performance, be very native on your computer, and send logs to a centralized server that will analyze them for the organization. This is an existing software product that AI app builders like Base44 are going to take a lot of time to get to, if ever.

I think that's also a lot of how we measure internally. One of our objectives is to ask, what's the percentage of software that we can build today, and how do we push it forward?

I'll give another example. We're working now on allowing applications inside Base44 to use scheduled tasks. Maybe you want to build something that, every Monday morning, sends an email to all your users with a summary of what they did inside your app or whatever.

Or you build a personal assistant, you've integrated WhatsApp, and you want to tell the personal assistant, “Hey, remind me in 2 days to pick up the kids.” Whatever. This obviously doesn't exist today, and building this basically pushes the type of software that you can build with Base44 slightly further.

We've built this built-in capability to build AI agents, and we've integrated it with WhatsApp and Slack and whatever. Now you can actually build bots, not only applications. We constantly measure that type of software that we can build.

I think Cursor, Windsurf, and Cognition are doing a terrific job working on enormous software projects, especially enterprise projects and large, very complex software that needs a lot of back-end services. They're working on very complex applications, like if you want to build a firewall. We internally obviously use Cursor to build our own back end.

I don't know where the market will be headed. I'm sure that they will go a bit upmarket and bring in some of the visual builders for the less complex users. I'm sure that the outer layer—the Base44, Bolt, and Lovable of the world—will go a bit downmarket with the types of software that you'll be able to build. Software that you weren't able to build a year ago or a few months ago, you'll be able to build in a few months.

It's going to be interesting, but this is a huge market. A lot of it depends on the type of user, because you cannot build for every type of user. Either you serve the developer, where they want to see the code, edit it, bring in terminal functionality, and run it on their computers, or you serve the nontechnical user, which is completely different.

4. I’m Worried About Google

Now, you have Lovable, Replit, and Bolt. Who's the best and who's the worst?

Tough question. I think Replit is more technical and started as a developer tool. For developers, they're doing a good job—for semidevelopers, we'll call them that way. Hardcore developers will probably go to Cursor. I won't say who's the worst.

Which one do you not worry about?

I'm not worried about those 3.

Wow. Who are you worried about?

5. Margins Do Not Matter

I think more about market dynamics and how things will turn out with the model providers. My theory and thesis is that there's always going to be a huge advantage to the folks who are playing with multiple model providers. But if, at one point, one of the model providers wins the market by a wide margin, then the next logical thing is to conquer the vibe-coding market because, in my non-objective eyes, this will be the largest software category.

As long as there's a fight and there's a race, we're positioned to control a very large share of the market and bring a better solution to the market.

Who do you think is more likely to do that: Codex or Claude Code?

Google, by far. They have the compute, and they're moving very fast. If Gemini at some point wins the race—literally wins the race—then they have the entire stack, including Google Cloud and whatever else they need. They have the data, integrations, Google's suite, and so on to build an incredible empire.

I'm hoping for a tough race, and it looks like that right now. It's, by the way, insane market dynamics because, obviously, everyone's spending a lot on AI, and in a split second, you change 1 string in the code and move millions in spend to a different provider. It bounces back depending on who provides the best model and the best whatever—cost, throughput, and so on.

Can you give me an example of that?

Let's say, in the coding space, Claude comes out with Claude 4 and Sonnet 4, and you're like, "Perfect, let's use that." Before that, we were splitting between Sonnet and Gemini. Then OpenAI comes up with GPT-5.

It's not necessarily that there's a better model and you move your entire workload. You see that there are places where each model is doing better. Sonnet was the best designer, knew how to structure things, and was really good at writing a project from scratch. But what we found is that GPT-5, especially the thinking and pro versions, is really good at solving hard bugs. So, you move a lot of workloads—especially on complex applications and very large-context API calls—to OpenAI.

When you do that, you're literally moving hundreds of thousands in spend with just the simplest change. If Gemini 3 is going to be better than anything else—which is what people are saying nowadays—then we're going to move a lot of spend and a lot of workload overnight to a different vendor. That's insane market dynamics to me.

I get you. It's an insane market dynamic because everyone compares the LLM market to the cloud market in terms of it being a commodity, but actually, it's a megamarket that will be phenomenal for the providers, with a couple of core winners. What you just said there is completely different, which is an ease of switching that you would never have in a cloud market. You do not switch your cloud provider with a line of code in seconds, multiple times a month.

No.

How do you think about that analogy from LLM provider to cloud provider, given what you just said?

You do not switch a cloud provider. That's very hard to do. My previous company tried doing that, and it was so painful. If I had known how tough it was going to be when we started that long process, I probably wouldn't have done it.

For the LLMs, it's insane how easy it is. I think what we're seeing from the model providers is that they're building multiple tools for multiple levels in the value chain. You see all of them having a CLI coding tool and all of them having this very high-end, or end-user-focused, chat product, like Gemini and ChatGPT.

You're starting to see them bring in—Gemini is now everywhere in my Gmail and in a lot of workflows—and build automation. You're starting to see them go up the software market and bring their tools and models to drive consumption all across the stack, whereas a year ago, it was mostly that you could consume them either through an API or through this end-user type of chat.

I don't know. It's going to be super interesting. To me, it's crazy that one of them could storm ahead for 6 months—something that hasn't happened before—and literally have the best model for 6 months. How will they be able to handle that?

Let's say someone comes out now with a better, stronger, and cheaper model than Sonnet 4.5, and that remains the case for Anthropic for the next 6 months. What happens to Anthropic's revenue? Overnight, you switch from all the Cursor calls and Claude Code calls to Gemini or whatever—GPT-5.5 or whatever. That's such a fast decline, but they also had such a fast ascent.

I don't know. It's going to be crazy, but I don't think I have the right answers for how this market is going to play out. Obviously, we're all hoping that this stays a very competitive market where prices go down.

I think also, when we're making strategic bets on what we're going to optimize and what we're not, it's very strange to manage a business like Base44. In some places, you see things that you need to either optimize or build, but you're deciding not to do that because you know future versions of the models are going to take care of it.

One of the things that Avishai and I were discussing when we were discussing strategy is that we're taking into consideration that the prices of models will go down toward zero, at least for platforms like Base44. This is a strategic decision where you're saying, "Okay, I could optimize some, and we're doing that, but I won't invest a lot of people right now to optimize the costs." We're taking into consideration that the prices of the models are just going down and down and down.

One of my core questions that I want to understand is the margin structure. When you look at today, every dollar that comes in, what percentage of that dollar goes back to an LLM provider?

It depends on the timing. The exact number I cannot say because we're a public company, even though I miss being able to share that a lot.

No way, dude. No one shares that. Why? There's no way that—again, educate me—does no one share it because it's so bad?

No, it's also very decision-dependent. You can change it from a low margin to a very high margin with just 1 decision: deciding how great of a service you want to give your users.

I think what will happen eventually with all coding agents is that the cheaper open-source models will get way better. You always want to play with the strongest models, but what we're looking at internally is building this switch or proxy where, when you get a prompt from a user, maybe the prompt is so simple that it doesn't even matter what model you use.

If it's, "Change the color of the button from blue to red," I don't really need to use heavy Sonnet maximum thinking or whatever and pay a lot for that. I can forward the request to a smaller, cheaper model. This will also be a way better experience for the user because it's going to be so much faster than playing the heavy guns of the frontier models.

The more smaller or open-source models are able to take a larger percentage of the requests and prompts, obviously, margins are going to improve drastically.

I don't think it's bad.

Yeah, I don't think current margins are bad, but I think they're going to get way better.

Today, they suck. Everyone always says, "Oh, but we're going to have model selection," exactly as you said, "which is we'll be able to intelligently route our models." And I'm like, "Okay."

I hope so, Maor.

Yeah, I hope so. We all do. I think it depends on the product and the flow that you're building.

So, one of the players that you mentioned, for example, is doing a really good job with autonomy. You prompt something, and then they go and run for 20 or 30 minutes. They have this coding agent. This is Replit’s agent, right?

Yeah. Yeah. But first, what if my prompt—what if I didn’t really mean what I just prompted? It went and built this fully, and I’m just a free user, and I just burned a lot of cash right now on coding and testing and whatever. So, I think it depends on the level of autonomy you want to give and how fast your iterations are for users.

I don’t know the margins, but I’m saying you could burn a lot by not understanding what the user actually wants. But from everything in the industry, margins are the least thing that I’m worried about. The thing you’re most worried about is Google coming in—not really, but the competitive landscape. The pace is insane. For us, every feature that we put out, we know it’s going to take either a few weeks or a few months for a competitor to copy, and we see that a lot.

How does that change what you build? It’s really interesting. I had the founder of Fiverr, Micha Kaufman, on the show—he’s from Israel, too. He said the time to copy is the single biggest difference in the last 2 years of software: where it used to be years, now it’s weeks. How does that change what you build or how you think?

You take big bets on things that are substantial to your users and will be really hard to copy, if you can find those. So, I’ll give you one example. The whole reason for Base44’s existence is the built-in backend. The fact that we built Base44 as a fully vertically integrated platform means that, for every application, you already have a built-in database, user management, authentication, integrations, analytics, and so on.

The thing is, we didn’t use a third-party provider like Supabase, which is what the rest of the industry is using—Supabase, Neon. We’ve built that in, and we’ve taken a very courageous step there, and there’s still a lot more to build. This is going to be very hard to replicate, and very hard to migrate your, whatever, millions of users from Supabase to a homegrown solution if one of the competitors tries to do that. I think this is a bet that we’re deepening further and further and further.

For other stuff, it’s a game of velocity, delight, taking the right UX decisions, and having the right taste. It’s just trying to be there before everyone else gets there. About 2 weeks ago, we came up with something called app suggestions. It’s a very small feature, which is very nice: after you prompt, it gives you the next steps, like, “Okay, now add a dashboard.” It gives you suggestions for what you’re doing in your app.

It took maybe a few days for other competitors to copy that. But it’s fine. That’s the game that we’re playing now.

What percentage of starting users finish the project?

It’s a good question, because it depends on what finishing means. I don’t think I have the answer for that.

A deployed application that works. That’s many.

But that’s not the right question. The right question is, how many get the usage they want? How many get to actual usage?

Many published apps are fully functioning. At some point, you battle-test your agent enough to teach it how to query the database. For every flow and use case that you prompt it with, you start categorizing the different flows and use cases, and you make sure that you feed your agent with the right examples. Many get to functioning applications, so this is a saturated benchmark. The question is, how many get to actual day-to-day usage? I don’t have the right answer to that.

6. What’s the Metric for User Success?

What metric is user success for you, then? If there’s this nuance and complexity with regard to the success of a user, internally, what is your metric for user success?

There are a couple. First, we measure the sentiment of the messages they send in the chat, which was also weird to me. But it turns out many people speak with the chat as they would to a human. They either tell it, “Okay, fantastic job. This is exactly what I wanted when I was asking for this feature,” or they cancel a lot: “This is not what I wanted. You deleted this button, and I wanted this and I wanted that.”

So, we have this metric that we literally measure on a minute-to-minute basis: the sentiment of the prompts. Nowadays, we get tens of thousands of prompts a minute or whatever, in tandem. That’s how we know our coding agent does what it’s being asked to do. If we release an upgrade, you see the negativity metric go down. When we jump from Sonnet 4 to Sonnet 4.5, you see a major improvement there.

Again, even the metrics change a lot with a business like Base44. When we started, it was mostly about bugs: how many times the model messes up something and the application doesn’t work, or there are bugs and things that break your app. Nowadays, it’s not easy to create bugs in your app. The LLMs are really smart, and the infrastructure is battle-tested.

It’s more about whether the agent actually does what the user is asking of it. Does it make the right change without making additional changes or messing up other features? For example, take scheduled tasks. Maybe the user asks, “Hey, I want you to send me a daily email with reminders,” and then the coding agent says, “I cannot do that. I don’t have the right infrastructure to do scheduled tasks.” When we add scheduled tasks, we should see a few percentage points of this metric improve.

Going back to the element of margin, I had Av Randhawa, who’s a new GP at Benchmark, and he said, “Harry, Harry, you’re thinking about this wrong. We need less of a focus on margin, and we need more of a focus on gross dollars per customer—in other words, expanding the spend of the customer, not the actual margin on a lower spend.” Do you agree with that, or do you actually think, “No, we’ll revert back to a margin-optimized world where we have open-source models, as we discussed”?

I think he’s 100% right. Right now, the game for our category is growth and capturing as much of this insanely big market as possible. There’s almost no person who shouldn’t be a user of a platform like Base44. So, obviously, it’s about growing—not necessarily the spend of every customer, but the number of customers—as much as possible. The spend is going to come down as models get cheaper.

So, I don’t think it’s, “Let’s try and grab as many dollars from the customer as possible.” I’d say margins are always important to keep your business going. There’s one thing I learned in Base44: when you’re lean, you’re a sane business, and you’re a good business, that does really well for all other aspects of the business.

It’s not something that I understood in my previous company. It wasn’t like I made fatal mistakes, but we were very capital-heavy. There’s something about creating a business that makes sense and is healthy in margins, retention, spend, and whatever that allows you to grow later on in a sustainable way, without it at some point hitting the brakes and saying, “Oh, well, nothing makes sense in our margins,” or whatever.

7. Do Early Revenue Numbers Actually Matter in AI?

I’d say margins are something you should constantly look at while knowing—or while planning ahead and taking into consideration—that the model-related stuff, which is 99% of the spend, or 90% of the spend, is changing drastically and aggressively every few months.

When you have that change in spend, you have less certain revenue. As we said earlier, the transience of spend means your spend with Anthropic could skyrocket or could not, depending on where you shift spend.

Can you advise me honestly, dude? I’m an early-stage investor. I invest in AI apps today, and everyone is posting insane numbers. You look at yours, and it’s like $100 million. Lovable and Replit are like $200 million. How much weight should I, as an early-stage investor, place on revenue today?

That’s a very good question, which I’m also asking myself as an investor. Obviously, metrics today are different. When I started my previous company, we got to $2 million in ARR in, I think, a year and a half, and everybody was cheering and saying, “This is insane,” or whatever. Nowadays, if you can show the hockey stick of how fast you got to $10 million or $20 million, everything changes.

I think it depends on the model of the company you’re investing in. For me, it’s smaller checks into very healthy businesses. The thing that I look at the most is, okay, fast revenue growth—obviously, that’s what you want to see. It’s the most exciting, but there are 2 things.

One, a healthy business: how likely is this business to get eaten by other players, the model providers, or whatever? And the second, which is very much related, is a vertically integrated business. Here’s something really cool that we’re looking at right now. A company with an unclear name came out, I think it was 2 weeks ago or maybe a few weeks ago, with its own model, right? Composer.

And this is a nice way to think about the strategy of those types of businesses. Think about that company. I don't know what the margins were, so let's take the not-optimistic approach of saying, “Margins were bad, right? They were paying so much to the other model providers or whatever.”

This is an insane strategy. It’s like: grow as fast as you can, get insane distribution—every developer that creates for Twitter probably installs it—and then, at some point, bring in the very efficient model that you trained, which is now fast, and start moving people from the very expensive, low-margin models into your own model, which is likely to be high-margin.

Then, again, overnight, your margins get tremendously better. You’ve gone fast as hell, and now you have the distribution. You can optimize margins because, again, the margins for the models—that’s something that isn’t easy to change, but if you decide to, it’s very doable to change drastically overnight.

I think this is a really interesting strategy that we’re also thinking about internally. It’s like, go as fast as you can with the frontier models. At some point, open source becomes a really viable option to maybe fine-tune on the billions, almost trillions, of lines of code that we have inside Base44, and then have this very interesting margin take.

For some of the prompts, you forward that to the fast, high-margin, great model. Users really enjoy it because those models are going to be way faster than their heavy Sonnet 4.4 models. This is an interesting take when you think about vertically integrated applications.

There are some applications today that I think are growing very, very fast that are going to have a harder time keeping it this way. I don’t remember the name of the company. There was this company that was one of the earliest practical use cases for LLMs. It was in marketing; they were generating content, and they grew super fast. You probably remember them.

Jasper AI. Yeah, it grew super fast, and the moat was just—it was absolutely hard to create a moat there because it was clearly like, “Okay, you’re only using LLMs, right?”

For Base44, it’s an entire infrastructure. You build a mini cloud that allows applications to send emails, generate images, use other LLMs, deploy, manage users, and whatever. So you have something around LLMs that makes sense to have as enough of a business.

When I’m investing nowadays, I’m thinking, “Can this business, if not today, be at some point in the future vertically integrated and self-sustainable?” Maybe that’s what Cursor will turn out to be at some point.

Can we just go through this in a helpful way? You said 3 elements there: revenue growth. What revenue growth is exciting, then? Is it just the highest? Do I need to see 1 to 10 in the first year?

If you’re building a SaaS here, it’s going to be 0 to 10. That’s going to be exciting.

0 to 10 in a year, for you, means—

Yeah, I think this is insane. 0 to 10 is insane growth for any SaaS. Well, honestly, now it’s good, but there are quite a lot that do 0 to 10. I mean, a lot.

Insane. Yeah, that’s what I’m saying. It’s going to be the exciting piece. I’m not, when I’m looking to invest, necessarily going to say, “Okay, the company has to do 0 to 10.”

How do you determine if they’re going to be eaten by models? What is the question or criteria that will help you get comfortable with that?

If most of the value is in the way they prompt the application for a certain thing—for example, they’ve built a photo editor—and there aren’t a lot of features in an AI-based photo editor, where you feed it a picture and say, “Okay, make me—I don’t know—make me pretty,” or whatever, and if a lot of that is going to be the UX and the prompt, I don’t think that’s going to be a healthy business for the next decade. I don’t think there’s a lot of moat.

If you’re building a lot of infrastructure, or the company itself is vertically integrated, I think that’s a very interesting take. You could nowadays build a competitor to Harvey and Spellbook that would teach an LLM a better way of doing law or whatever. Or you could build a law firm.

I think building a law firm is going to be very, very, very interesting because this is not something that a model provider would do, or any other sane software entrepreneur that wants to do just software. I think there’s an opportunity today to build very healthy, new-age, vertically integrated businesses.

It could be a law firm. It could be buying a hospital, basically optimizing everything on every layer, making it AI-native, and maybe in a few years being the first ones to introduce robots as surgeons or whatever. Nowadays, there are opportunities all across the funnel, in different categories—software, not software, whatever—to build vertically integrated companies.

If your company is just about how you’re prompting an LLM, and maybe you’re doing a better job tuning it to whatever legal needs or financial needs, and the infrastructure that you have below it is not very complex, it’s going to be hard to have a moat.

Are you angel investing today?

Yeah, or starting to.

Do you enjoy it?

Very much. It was scary at first, but nowadays it’s so much fun. It also gives you a chance to work with incredible founders and become part of this huge platform shift, probably one of the biggest ones that we’re facing, and to think with them, “What the hell are we going to do?”

But also to think with them about the new opportunities that they’re going to open up. They’re building, so coming back to this crazy idea I said earlier, building a law firm 5 years ago would seem like the worst idea you could think of, right? Building a law firm, buying a hospital and optimizing it with AI or whatever, building a new bank.

I think today you have this huge platform shift that creates so many exciting discussions when you look at different industries. You think, 2 years from now, 4 years from now, what’s the craziest thing that we can think of? What should we aim for that’s barely possible today, or not possible today, and should be possible in 2 to 4 years?

I think this is an incredible way to think about angel investing, and it’s so much fun doing so. It’s very risky, but very, very fun.

Where do you think no one is investing today that everyone should be investing today? And where do you think everyone is investing today where they should not be investing?

Oh, wow, that’s a good question. I think the not-sexy industries—again, finance, restaurants, whatever, HMOs—actually having entrepreneurs building that end-to-end is what people should be investing more into.

8. Where Should Smart Money Go Today?

I think a lot of people are still investing nowadays in very fast-growing, hockey-stick-revenue startups, which is not what I think—unless you know a business or an industry very, very well. I think it’s very hard for investors outside of it to understand if this is going to have a moat or not.

A good example of that is that I had the opportunity to invest in these agent companies, like Manus and those firms. I felt like I didn’t know the industry well enough to understand if, at some point, this would get commoditized by the model companies. This is an area that I wouldn’t invest in because I was like, “Okay, this is obviously dangerous.”

The capabilities that OpenAI built into ChatGPT, and then Google into Gemini or whatever, are going to include serious and incredible agents. I’m not sure this category—even if you get to insane growth—has a lot behind it, rather than prompting and giving it tools that are for sure going to be part of the ChatGPT or Gemini interface or whatever.

9. How does Base44 beat Cursor?

Mike Eisenberg said I had to ask you: make the case for how you beat Cursor.

When I started Base44, I said, “Okay, I’m going to do this really for the nontechnical people,” because, at that point—and still today—most of the players in the market, if you wanted to build something really functional, you should have prompted, “Hey, connect the backend,” and then it would tell you, “Okay, bring the API key from Supabase,” and whatever else.

Those are things that your mom and my mom wouldn’t be able to do, right? So I said, “Okay, let’s build for the very nontechnical folks.” When doing so, obviously, Base44 started getting a lot of traction, and we also got some developers and technical people who said, “Give me the code.”

They said, “You should invest in a GitHub integration, allow me to edit files, and whatever.” At some point, we also implemented that. But the more time passed, the better the models got, the faster they got, and the fewer bugs there were, nobody wanted to see the code.

I think the direction we’re going in—and again, that we’re going to—is that more and more types of software will be buildable inside Base44, where nobody wants to see code.

I don't think it's like if you'll get to this flow state where you're prompting with AI, it's going to be super fast and almost immediate every prompt. You'll see that, and then it's going to change. I don't think you'll want to see the code, and I don't think you want to start saying, “Okay, you should look at these files and those modules, run those tests,” and whatever.

I think nowadays, for some—and a growing part—of software, even the most technical users would prefer building inside Base44 than doing that in Cursor. They don't want to set up servers and databases, see the code, write tests, and navigate through the files. My guess is that this portion of software that you can build inside Base44 is going to grow.

Obviously, nowadays, if you can build it inside Base44, you'll probably prefer to do it inside Base44 rather than in Cursor.

You can invest in Cursor at $29.3 billion, or you can invest in Cognition at $12 billion. Which one would you do?

Probably Cognition. I think the notion of software building as a category is going to grow insanely fast. I think our notion from the previous age of venture capital, where the winner takes it all, is wrong, and I think this category is going to be so big that everybody will take a niche at some point.

I think Cognition is going to grow and have more room to grow. Throughout time, they'll be able to replicate the most successful features inside. I think they have more room to grow, whereas betting on Cursor is betting that it's a winner-takes-all market.

Why don't you think it's a winner-takes-all market? And if it's not a winner-takes-all market, does that still work for a venture model?

Our business is predicated on winner-takes-all? Not necessarily. Our business—or the venture capital business—is predicated on huge returns.

If you look at all of my competitors, Base44 included, we're all growing insanely fast. I think, at least, I can say that about Base44, but maybe the other players are healthy businesses. Those are going to generate huge returns. They're all fast-growing businesses, and it's not a winner-takes-all market, which is also, by the way, why I got into this market in the first place.

When I started, Bolt and Lovable were very viral on social media. I felt like this was going to be such a big market, and I felt like I had a very interesting take on this market, with the back end built in and a focus on functional, complex apps. Most of the other players back then were focusing on front ends and maybe could connect the front end to Supabase, but it was very hard.

I think we're at a different ballgame, where AI is going to generate so much value that markets are going to grow very, very fast. Obviously, some will get cannibalized, but markets are going to grow very, very fast. The VC model is not necessarily winner-takes-all; it's a huge-returns type of market.

10. Will the AI Boom Hit a Revenue Speed Bump?

Do you think we will hit a speed bump? Everyone is very concerned about the AI bubble and not delivering on the revenue that we promised. Do you think that we will have a speed bump where there is a sign that we don't deliver the revenue that we promised?

I think the question is whether it's going to turn out to be a platform play, which means that when better models are built, the entire ecosystem and the market benefit from it, or whether what's going to happen is that model companies are going to fight and their margins will get worse and worse because they'll need to lower prices.

Maybe at some point we'll need to figure out or crack something better in the LLMs, because maybe we'll hit a wall. This will be the bigger question: Is the value that GPT-5 to GPT-5.1 is adding only to OpenAI, or is it for the rest of the industry—or the rest of the market?

If it's for the rest of the market, I don't think we'll hit a speed bump. Personally, I don't believe we'll hit a speed bump. I also don't think we'll hit a wall with the performance of LLMs, but even with existing models, we're only scratching the surface of the economic value that you can create.

If a one-person team can get sold for $80 million, building a viable, healthy, and profitable business without people, it means that you'll be able to do a lot more stuff with very few people. Companies will become more efficient, and as a result, every product that we buy is going to be better and cheaper.

I think there's so much more to do. And this was with the previous version of LLMs. I've seen what it takes to build a company that needed to raise—or was very capital-heavy—$130 million and over 100 people. I've seen now, in the age of LLMs, how much better it is, or how much easier it is, to build a very lean company. I was one person.

I think this value is going to trickle down from enterprises becoming more efficient, to many more startups because there's no technical barrier anymore, and eventually to the consumer benefiting from all of this economic value that's going to be created.

How many people are in Base44 today?

No, I cannot say. 5 minutes before the conversation, they told me I cannot say numbers or the number of people.

Really? Well, I don't want to pry, but why is that?

I don't know. Just the number of people. I think they're waiting for the investor update.

What's going on? That's interesting. I don't know if it's a good thing to have more or less. If I was a public company, I wouldn't say more because it feels more secure to a less educated investor base that doesn't think about AI efficiency within dev teams.

11. Are AI Builders Overvalued or Is Wix Undervalued?

I think the public markets nowadays are so fluctuating, and it's very hard—you have to see consistent results. I don't think that the stock behaves the same way that you see inside the business.

I don't understand, Maor. If you look at a Lovable, a reported $6 billion price on the latest round, or Replit at $3 billion to $4 billion, and I'm sure they'll raise it to $6 billion soon because they're kind of memetic in that way, either they are grossly overvalued or Wix is grossly undervalued. You have to take one of those views. Which one of those views is right?

My very non-objective opinion is that Wix is very much undervalued today. Put aside the fact of product assets and moat, there's also a moat in being big, and there's a moat in having a marketing team and an organization with experience like ours.

I'm not sure if Lovable or Replit is overvalued. I think, to be able to say that, I need to look at the internals and the economics of those businesses. I don't really know those.

I'm sorry, do you not track them very closely?

The revenue and the things they publish. But we're focusing on just building. And this is something that Wix taught me.

Wix, for many years, was the underdog compared to other players. When I came into Wix, I was saying, “Those guys are doing this, and those guys are doing that. They're publishing here, they're viral here, and they're copying our features.” Wix's mindset is just, “Build a great product, find the right way to market it. It doesn't matter what others do.”

You see that purely in the business, in the metrics. We've released the new agent, the version of our agent, and you can see the conversion goes up. You can see that the marketing team tests different things. Some of them didn't make any sense to me beforehand, and you can see the unit economics getting better.

One thing that Wix taught me is to focus on your business and focus on a great product. There's so much noise out there. I remember, with my first company, 40% of my mind was occupied with what competitors were doing. “Oh no, there's—” And for me as well, it was like, “How do I grasp that the market is so big?”

Nobody cares. As a category, you're currently capturing 10%, 15%, 20%. It takes years and ages until you actually grow. There are so many people that could use Base44. There are also so many people that could use Replit and Lovable.

Focus on your business. Obviously, go as fast as you can, but I'm not trying to sniff around what the unit economics and margins are.

12. Does Not Being in Silicon Valley Help or Hurt You?

Final one before we do a quick fire. Replit, Cursor, Cognition—all very Silicon Valley businesses. Are you helped or are you hurt by not being in Silicon Valley, honestly?

I think the world is changing. With my previous company, I used to fly out every 2 weeks to Silicon Valley because this is where things would happen. Yeah, and it's tough. I'm not going to do that ever again.

I don't think in any way our business is hurting. If anything, I think the war on talent is really tough in Silicon Valley. If you're playing the game of getting funded as much as you can so that you can build for a lot of cash and conquer a lot of the market, and that's your way of conquering most of the market, awesome. That's where you should be.

I think for us, Wix is a monster, having our backs. They have deep pockets, they know what to do, and that's my Silicon Valley, right? That's the backup that I need.

You also start to see—and I think I have a lot of great things to say about Lovable building it from Europe. I think the world is changing.

13. Quick-Fire Round

Listen, I want to do a quick fire with you. I say a short statement, you give me your immediate thoughts. I have Marc Benioff and Vlad Tenev from Robinhood on the show.

They said 50% of their code is written with AI. What will that be in 2 years?

95%. Over 100% for sure. You’ll be prompting agents. I think for them, maybe 50%, because they have legacy code and systems. For us, Base44 was built in the age of LLMs, and I structured the repository in a way that’s going to be very easy for LLMs to navigate through and read the code. I think for us it’s already closer to 90%, mostly prompting agents.

Single biggest piece of advice for someone nontechnical wanting to build with these tools? I listen to you say you’re genuinely doing them, and I’m like, “Wow, I really want to build a tool that allows me to go through all my LinkedIn followers, find everyone who has a certain title that would suggest they’re an LP and an investor in funds, categorize them, and show them to me because they clearly have an interest in me already. I could outbound them and turn them into an LP in my funds.” I don’t know where to start. I’m intimidated, actually.

First, I’ll say step 0 is the biggest hack. It’s how you actually build something that later on you’ll use or maybe be successful with, which is exactly what you mentioned: build something for your own problems. This is the biggest hack that I learned. This is the easiest way to get software that’s actually usable.

This is not something I want to go back into doing for a customer that I interview a lot and understand the pain points of. It was so much fun building Base44 and so much easier when I had a use case in mind. I was like, “I’m going to build for my partner. I’m going to build her a CRM, and I’m going to iterate on this software until it hits the mark.”

For people starting out with tools like Base44, there’s a mindset that you have to get into. One tip that I constantly give out is that Base44 is not really a person that’s working for a week and getting back to you with, “Hey, this is what I built.” You have to get into a mindset that it’s very easy to revert or start from scratch.

Instead of going into building a PRD and then feeding it into Base44, build a first version that you know you’re going to throw out. For some people, it’s emotionally hard to understand that they’re going to throw out tens of thousands of lines of code. It’s going to be okay, fine. But all you need is to figure out the product stuff, because once you have the product stuff, with a few prompts you get to where you got to with hundreds of prompts or thousands of prompts.

Iteration is everything. It’s not anymore about planning ahead so that you save developer time. It’s about very easily hitting the revert button when one feature you wanted and played with is not something that you need, and getting to a point where you’re saying, “Okay, this wasn’t really a good idea.”

Iterate very fast and understand that you’re not working with a human. Therefore, it’s fine to try out different directions or to just give it a high-level idea and see what it comes up with. Then, from there, iterate.

You can start with the highest level of prompts, like, “Build something that will help me, I don’t know, engage my audience on LinkedIn,” and see what the LLM does, because it will have ideas that it will implement inside the application. Some of those ideas make you say, “Oh, this is interesting. I didn’t really think about this.” Like the fact that you can maybe generate outreach, or whatever.

See the things that you didn’t like, throw those away, take it again, and then prompt it back and say, “Okay, build something that will engage my LinkedIn audience. One of the features that I really like is this and that.” Then build around that. The idea that you can throw away things that you don’t like and start from scratch, or revert or go back, is very hard for people to grasp.

You made $80 million with the sale. Well done. Does money make you happy?

I don’t think money makes you happy. I think there are 3 things that money becomes really valuable for. One is your time. I don’t like fancy cars, so I didn’t buy a car, but nowadays I’m going by taxi. You save a tremendous amount of time, and there’s all the other stuff that I can pay for instead of doing or wasting my time on. This is a lot of help in personal stuff, especially for busy people and entrepreneurs. You want to be with your wife or family or whatever, so this is a huge benefit.

Second is helping people you care about and helping causes that you care about, which is big, because at some point there’s not a big difference between doing X million dollars and 10X million dollars. It’s more about, okay, you can maybe, with your family and stuff like that, support the people and causes you care about.

The third, I’m assuming, is that you can use the money later on to build the bigger stuff. Or you have this war chest that you can say, “Okay, the next thing I’m going to do—or when I’m thinking about what I’m going to do a decade from now—is, okay, let’s go to space, solve cancer, build robots, or something like that.” Because you can do that. You can take a riskier bet.

I just want a Gulfstream, dude, okay? Don’t get all heavy on me. Fuck that.

There are some things that get edited out. I leave it to the team to see whatever they keep. Remember, the job of most interviewers, yeah? They make you look bad and make themselves look really innocent. I make myself look bad so you look innocent. It’s a much better deal for you.

I think one of my friends told me there was a saying. After I made the money, we had this discussion among friends. I was like, “This doesn’t really change your life.” My friends were like, “You’re using it wrong.” I was like, “What do you mean?” They were like, “Obviously money makes you happier.” I was like, “You should buy a jet ski. Have you ever seen someone sad on a jet ski?” Nobody. I agree.

OpenAI at $500 billion or Anthropic at $360 billion—which one would you rather be invested in?

Anthropic every day.

Why?

They’re moving faster. They’ve cracked something in coding. They’re very focused on delivering value to B2B customers. Although they don’t have the first-mover advantage, I think they have better models, and when I look at their trajectory, they’re better positioned for growth.

You can add 1 person to the Base44 board. Who would you add, and what would you like to learn from them?

Wow, that’s a good question. Jack Dorsey.

Why?

I think he’s one of the most brilliant minds and founders, and you can see that consistently. He builds things sometimes just for the sake of building. You can see some of his latest pet projects, like this messaging protocol and stuff like that.

I think this is a person that’s consistently building Twitter, building Square, building other projects. That’s just a mindset of a builder that I think can add tremendous futuristic strategic thinking, while also, I think, our DNA and the thing that we’re trying to build in Base44 is a builder mindset. I think this is, for me, the ultimate guy.

What thing or event did you sacrifice for work that, with the benefit of hindsight, you regret?

Not necessarily this time around, but in my previous company, I think I missed a lot of my 20s. For the first few years, I sacrificed a lot of time with family and friends, missing out on many different occasions and memories.

I was so inexperienced that I was trying to compensate for my inexperience with overworking, which is fine. It got me to a good place eventually, but I sacrificed many things and many friends.

What thing are frontier labs completely delusional about?

Maybe the moat that they have. They don’t have what they think they have. I don’t know how they’re thinking about this, but what’s the endgame for the race? Either it’s not going to be 1 winner that takes it all, and there are going to be more and more companies getting into this market. There are going to be Chinese players that are now changing the entire market dynamics.

I think, eventually, if it takes 1 better model or a lower price to drastically reduce your revenue, then I think they—or at least, they used to be—delusional about this. I don’t know if they were thinking about this. Maybe they were always planning to build up the stack.

I think nowadays they’re doing small things. For example, Anthropic is building the Claude Code SDK and Claude Code. Even if Gemini 3 comes out and it’s a better model than Claude Sonnet, there are still going to be a huge number of people staying with Claude Code because they’re used to it and they’ve built some things up the stack.

Penultimate one: You got married very recently. What’s your biggest advice on partner selection?

It shouldn’t be like, “This person completes me.” It should be someone that is enjoying life at least as much as you, someone that is similar to you. My partner is an entrepreneur in the same kind of field. She likes the same things that I like. She’s messy. She has ADHD like me. She likes to paint like me.

This is so much fun, picking someone that’s like you, because as an entrepreneur, you have to have sympathy for the person that’s missing out day and night and is constantly working. You have to understand deeply why they’re doing that and that it’s a part of them as a person, the same as it’s a part of me as a person.

So, I think having the same mindset is crucial. It's very hard to date a founder that's deep into building the company zero to one. And I think having a person and a partner that's having the same mindset—that they're building something or they built something—makes it so much easier to empathize with.

Final one for you, my friend. You can go back to the night before you start Base44 and you can say, “Maor, you should know this. Future self has learned this today.” What do you go back and tell yourself that you wish you'd known when you started?

When I was by myself, it was scary as hell. I used to wake up every 2 hours at night just to check that the platform was up. At some point, Base44 started growing really fast, and I was still by myself, and people started building businesses and applications on top of it.

I remember this week when it started really going up, and the infrastructure and the thing that I built didn't hold up. So, there was this week or two or three when I had 4 incidents where the platform would go down, and people were screaming and shouting at me, like, “I trusted you,” and whatever, and I took it very hard.

It was emotionally very hard, and I wish I could have told myself, “This is just going to be a bump in the road. The community is going to recover from that, and eventually you'll build something sustainable. This is going to be great. Don't take the bumps in the road.”

Also, the consumer market is so much different. The emotional spectrum of the feedback that you get is so much wider than building a B2B SaaS, where you get some not-too-positive, not-too-negative comments. I think handling it better emotionally is important. I didn't know that.

Dude, I've so enjoyed this. It's a very free-flowing discussion. You've been amazingly receptive to putting up with my romantic advice request as well as my LLM routing request. Thank you so much for being so good, dude. I've loved it.

Dude, this has been so much fun. Thank you for having me, Harry.