I'm back!
After a year at the AI frontier, I am back building. Introducing Lucid Mayhem and our first project, Tomo.
The scarcity constraint has been lifted
After a year at the AI frontier, I am back building. Introducing Lucid Mayhem and our first project, Tomo.
A little over a year ago, after six years at Hotjar and Contentsquare, serving as Hotjar’s CEO and Contentsquare’s CSO, I announced that I was leaving to go back to being a builder and an entrepreneur. I knew I wanted to build something again, although I didn’t have a specific idea or plan yet, and before starting anything I wanted to take a bit of a break.
So I took some time off, travelled, and recharged my batteries. After operating at full speed for many years, having no meetings, no quarterly targets, and no large organization depending on me felt pretty good.
Within a few months, however, the pull of AI became too strong.
I spent most of last year working as close to the AI frontier as possible. I read, experimented with every new tool and model I could get my hands on, built prototypes, threw many of them away, and then built some more. Sometimes I was testing a specific product idea, and other times I was simply trying to understand what these models could reliably do.
This was a very interesting time to be experimenting because the frontier kept moving. Every few weeks something that barely worked became surprisingly good, and every few months a new model made a whole category of problems feel solvable.
Then Claude Opus 4.5 was released.
The Opus 4.5 moment
When Anthropic released Claude Opus 4.5 in November, the curve crossed a threshold for me.
Opus 4.5 could take a complicated objective, reason about the tradeoffs, explore a large codebase, make a plan, use tools, and work for a meaningful amount of time with much less hand-holding. I could spend more of my time describing the outcome I wanted, reviewing the work, and steering it in the right direction.
One model happened to make the change tangible, but the direction and pace of the entire frontier were the important signal.
For the first time, I felt that the scarcity constraint had been lifted.
Great people are still scarce. Taste, judgement, deep customer understanding, distribution, and founder attention are still scarce. The amount of software that a small team can produce, however, has expanded dramatically, and it will continue expanding as the models get better. This changes the economics of building companies without making the work easy. Many of the rules we follow are really adaptations to scarcity.
Focus is an adaptation to scarcity
Focus is considered one of the most important virtues in business. Pick one problem, build one product, serve one customer, and dedicate the whole company to doing that one thing extremely well. I generally agree, but building only one thing is partly a response to the economics of producing software. When every product requires 20 engineers, multiple layers of management, and millions of dollars in capital, every additional product creates a huge opportunity cost.
AI changes this calculation.
A small number of exceptional people can now use increasingly capable AI agents to do meaningful work across research, engineering, design, marketing, and operations. New ideas can be prototyped and validated much faster. Shared infrastructure and workflows can be reused across different products, and every new venture can strengthen the system behind the next one.
This is the idea behind Lucid Mayhem.
Lucid Mayhem builds and operates AI-native companies through small teams and AI agents. Each venture starts with a durable thesis and a real customer problem, while the companies share an operating system that gets stronger over time.
The individual ventures need to be extremely focused. The organization behind them can pursue several ventures in parallel.
Everything. All at once. On purpose.
Why build multiple ventures?
A few years ago I wrote about managing risk like an investment portfolio, with a mix of opportunities and capacity reinvested as risks are validated. I later wrote about how companies need to make multiple big bets and stack S-curves to sustain growth. Lucid Mayhem is the practical conclusion of those ideas. It gives us the ability to make several calculated bets while sharing the cost and learning between them. Code, infrastructure, agent workflows, distribution capabilities, and operational knowledge can compound across the portfolio.
The model still operates within hard limits. Judgement, distribution, customer understanding, and founder attention remain finite. It therefore depends on explicit theses, clear stage gates, and concentrating resources where the evidence is strongest. Each venture must validate its biggest risks early and earn further investment. Weak ideas should stop quickly; strong evidence should attract disproportionate resources.
I like to tell my teams that my job is 90% chess and 10% poker, and I think the same principle applies here.
Our first venture is already live.
Tomo: Duolingo for everything
Tomo started with a simple observation: we have access to more information than at any point in human history, yet most of us still struggle to learn the things we are curious about.
You watch a few YouTube videos, read half an article, open 30 browser tabs, ask ChatGPT a question, and then move on with your life. A week later, you barely remember anything.
Information is abundant. Structure, consistency, and habit are the real challenge.
Tomo lets you type any topic and turns it into a structured, gamified course with bite-sized lessons, quizzes, levels, XP, and streaks. You choose what you want to learn and how you want to be taught, and Tomo builds the course for you.
Think of it as Duolingo for everything you are curious about.
AI allows Tomo to create a personalized course on almost any subject. The game mechanics help you build a daily habit and make progress five minutes at a time. You can learn about investing, Roman history, coffee, psychology, artificial intelligence, or the strangely specific topic that has been sitting in one of your browser tabs for the last two years.
Tomo is available on iPhone and Android and is completely free during early access. Anyone who joins during early access will also keep it free when we introduce paid plans.
What happens next?
Tomo is our first venture and the first real test of the Lucid Mayhem operating model. We are learning a lot about how small AI-native teams should work, where agents create meaningful leverage, and where humans still need to stay very close to the details.
We are also already working on the next venture.
I will share some of what we learn here semi-regularly, including the things that work, the things that fail, and the parts of building with AI that are much harder than they look from the outside.
I believe this is one of the best times in history to be a builder. Small teams have more leverage than ever, and many ideas that were previously too expensive to attempt have suddenly become possible.
If you have an ambitious idea that you can’t stop thinking about, take it seriously. Put in the reps, build something, show it to people, and see what happens.
I am very excited to be building again, and I’d love for you to follow the journey.
Visit Lucid Mayhem, try Tomo, and let me know what you think.