Kazuo runs on an unusual org chart, with AI making the trend calls and assignment decisions while more than 300 human creators execute. What does managing that structure day to day actually involve, and what breaks first when it scales?
When I founded Kazuo, I wanted to build a system that could scale creator-driven distribution with the same scalability as traditional paid advertising. I designed the system around that goal. Day to day, managing creators is now less about managing individuals directly, and more about building the system that can coordinate and track content across our network. My company has automated content sourcing, most time-consuming parts of management, tracking, and payouts. There are also certain aspects that can’t be fully replaced today, such as content feedback, which is incredibly important for virality. Kazuo itself is only a three-person team: one developer, one UI/UX designer, and me. I personally manage our entire network of more than 300 creators. That would normally require a dedicated content and creator team, but our software automates much of the content sourcing, coordination, tracking, and payouts.
The most difficult part of scaling a creator network has been attribution. Kazuo’s creator network generates over 100 million views for our in-house products every month across thousands of posts, helping our portfolio reach more than 130,000 monthly active users and generate over $115,000 in monthly revenue. With our large volume of content and traffic, it’s very difficult to directly link an individual creator view to a paying customer. Unlike paid ads, there is no direct click-through attribution or the same control over where content is distributed or how it is optimized toward a conversion goal.
We generally have to rely on the audience of a creator to predict how valuable that creator’s views are. Every creator has a different conversion rate, audience demographic, geographic distribution, and level of emotional resonance with their viewer base. For example, on Tethered, our long-distance relationship app, we’ve seen two creators generate similar view counts but dramatically different customer value because one audience was concentrated in the U.S. and much more aligned with the product.
The vast majority of people search for our products directly on Google or the App Store, which breaks the attribution signal. To solve this issue, I developed our approach to probabilistic attribution models that piece together all the information we have, such as automated audience screenshot uploads, demographics and conversion rates of link in bio clicks (although a minority of total traffic), engagement rates, and comment sentiment. That being said, at our scale it’s impossible to get perfect attribution.
Entrepreneur First invested in you before you had an idea, betting on the person rather than the pitch. How did that shape what Kazuo became, and what did the Founder in Residence period change about how you evaluate opportunities?
The biggest shift was learning how to think bigger and develop my own theses about the future. Before Entrepreneur First, many of the ideas I worked on were relatively niche and had a lower ceiling. I was mostly solving problems that existed today rather than asking what the world would look like several years from now and how I could build for that future.
Being forced to think further ahead led me to the core thesis behind founding Kazuo: the cost of developing new software products will eventually approach zero. When anyone can build software, the product itself is no longer the moat. The real moats become knowing what people want and knowing how to generate massive distribution around it. At the same time, this enables an entirely new category of hyper-personalized, niche software products that historically would never have made economic sense because the development cost was too high. That thesis is already playing out across our portfolio, which now reaches more than 130,000 monthly active users on our apps and generates over $115,000 in monthly revenue.
Individually, many of these products are too small to become venture-backed companies. But collectively, they represent a massive market. I want Kazuo to become the system that continuously identifies which of these products should exist, builds them, and distributes them at scale. My role is increasingly about designing the strategy and systems that determine how we make those decisions at scale. That changed how I evaluate opportunities. I think much less about building a single app and much more about how we can develop a system to dominate the new market that will emerge when the cost of building software becomes zero.
Large companies spend months on product validation that your system compresses into days. What are established organizations structurally unable to copy about this model, and what could they adopt tomorrow?
One of the biggest advantages established organizations struggle to replicate is our ability to share both creators and content intelligence across multiple products. A traditional consumer brand usually only has a creator network and content insights around a particular niche. We operate more like a shared distribution network, where the same creator relationships, content learnings, and operational infrastructure can be reused across many apps.
On the creators’ side, this gives us meaningful economies of scale. Instead of hiring a creator on one deal for one product, we can give great creators opportunities across several apps. That allows us to negotiate better rates, reduce the time and cost to source new creators, and double down on creators that are driving traffic. Finding great creators often means testing many that don’t perform. Whenever we find one, we can multiply their traffic many times by putting them behind multiple apps, whereas with one product you can only benefit from their traffic once. We’ve seen this work particularly well between Tethered and Closure, our mental health and journaling app for people going through breakups. The audiences and emotional content styles are closely related, and nearly every creator who performed well on Tethered also performed well on Closure. That gave us a much cheaper and faster way to scale Closure because we already knew which creators could resonate with that type of audience.
Another key advantage is the ability to adapt viral content strategies in one niche to a different niche that hasn’t caught up yet. Short-form content is highly format-driven. The underlying structure of a successful video is often the same across vastly different niches. The hook, storytelling, sounds, pacing and visuals often follow similar patterns. Because we operate multiple products simultaneously, we can identify these patterns earlier than a company focused on a single category.
As the network grows, these advantages compound. Every new product gives us another surface to test content, every new creator adds additional distribution and data, and every successful format becomes something we can potentially adapt across the rest of the portfolio.
Established companies could adopt similar strategies by building a comparable system around sub-niches that fit their existing audience rather than expanding into completely different niches. Within any content niche, there are generally different content styles, creator archetypes, and audience segments that a single brand could expand into without changing the product. This can create some of the same compounding effects as our creator network, such as sourcing creators more efficiently, negotiating better rates, and doubling down on strong creators. The limitation is that the opportunity remains constrained by a single product and audience. Kazuo can apply the same system across entirely different niches, giving us a much larger surface area for experimentation, creator reuse, and the transfer of successful content strategies.
Your creators supply scripts and on-camera work while software decides what gets made and who makes it. What have you learned about where human judgment still beats the system, and where you were wrong about needing it?
Human judgement is incredibly important when it comes to creative feedback. AI can analyze performance and identify patterns, but it is still limited when it comes to understanding whether a video feels awkward, scripted, or emotionally resonant. These are subtle creative qualities that humans are still much better at interpreting accurately.
Where we were wrong was in thinking creator relationships could become entirely software-driven today. We’ve been able to automate far more of the operational work than a traditional creator team would, but we underestimated how important the human layer would remain. We’ve learned that the creator relationship itself is one of the hardest things to scale. Great creators want to work with people who understand them, respond quickly, and genuinely care about their success. That trust becomes more important over time, and it’s something AI still struggles to replicate.
Revenue above $115K a month across 8 apps means some bets worked and some didn’t. How does the engine decide when to kill an app versus keep feeding it?
There are many industry benchmarks people use to judge whether a mobile app has the potential to scale. Aspects like App Store view-to-install, install-to-trial, trial-to-paid, customer lifetime value, etc. These are all incredibly important metrics we take into account, but we look at the funnel as a whole, including the ability to go viral consistently and sustainably. We’ve had apps with strong metrics relative to industry benchmarks that still underperformed because they had limited viral potential, a high competition creator niche, or a difficult-to-scale content strategy. On the other hand, we’ve had apps with low conversion rates but incredibly cheap virality. We look at the funnel as a whole starting with content. Content replicability is also critical to scaling. Some content strategies are much harder to execute consistently, which makes it more difficult to find creators who would be a good fit. The more creators who can successfully execute a content strategy, the easier and cheaper it is to scale.
A good example is LumeSkin, our first app. The product-level metrics were strong: install-to-paid conversion was well above average. The bottleneck was distribution. In our first creator test, skincare creators were extremely expensive because we were competing with large beauty brands willing to pay sponsorship rates that simply didn’t make sense for our economics.
About two months ago, we revisited LumeSkin with a different approach. We offered some of our strongest Tethered creators additional work at a bundled rate, had them create separate skincare-focused accounts, and used our internal software to identify successful content formats and scripts from the niche.
The experiment taught us something important. While the strategy worked for a few creators, skincare content was significantly harder to execute consistently. Creators who could repeatedly produce high-performing content for Tethered often struggled to do the same for LumeSkin. The problem wasn’t whether one creator could make a viral video; it was whether we could find enough creators who could reproduce the strategy reliably and cheaply.
Compounding effects are also a major consideration. Some apps are great for generating short-term profits but are trend-driven and don’t build a strong long-term position. AscendMax, for example, was one of our first profitable wins early on and benefited heavily from the popularity of the “looksmaxxing” trend. While looksmaxxing is essentially a form of male beauty and AscendMax still operates today, we don’t expect the content trend itself to be as durable as categories like relationships, skincare, or mental health. We still invest in these apps while they are profitable, but we only double down on apps that have inherent network effects or a moat that compounds over time.
Executives reading this are being told AI will transform their businesses. From inside a company that was built AI-native from day one, what do you think that advice gets wrong?
A lot of the advice around AI focuses on making every employee more efficient. I believe the bigger opportunity is to build systems that use AI to amplify the highest-leverage people at a company. This allows a small number of strong decision-makers to operate at a scale that previously required much larger teams. Kazuo is a direct example of this. We operate the entire company with three people: one developer, one UI/UX designer, and me. I personally manage a network of more than 300 creators through the systems we’ve built.
For companies that want to become AI-native, executives should be thinking less about how they can layer AI tools onto existing workflows, and more about how to restructure the organization so that their best people can influence and execute across a much larger portion of the business. If execution becomes dramatically cheaper, then the optimal company structure changes with it. Roles, reporting lines, and entire functions that made sense before AI may no longer be the most efficient way to operate.
