How I went from investing in great companies to building one: Munger, moats, and learning machines

When I was in college, a mentor recommended that I read Poor Charlie’s Almanack. It was a timeless piece for aspiring investors. At the time, I thought I was reading a book about investing, which in most ways it was. Charlie Munger was one of the greatest business minds of the last century, and the book is full of lessons on moats, incentives, judgment, and compounding.

But the book stayed with me for a different reason. Munger wasn’t only teaching people how to pick stocks, he was teaching a way of thinking: how to be curious, how to avoid fooling yourself, how to earn trust, how to build good judgment over time, how to keep learning long after you have accumulated enough knowledge to sound smart, and how to live a full life.

That was part of what drew me into investing in the first place. Investing is intellectually stimulating because it forces you to study what makes a business durable. Why do some companies compound for decades while others fade? Why do some teams earn trust while others lose it? What makes a moat real, and what only looks like one from the outside?

For the past few years, I thought about those questions mostly as an investor. Since joining Decagon, I’ve started thinking about them differently where I’m no longer studying great companies from the outside, but learning what it feels like to help build one from the inside. On the inside, Munger’s menu of can’t-fail ideas still resonate strongly.

The first idea is becoming a learning machine. Munger once mused that he saw people rise in life not necessarily because they were the smartest or even the most diligent, but because they were learning machines. In the age of AI, where the pace of change is measured in weeks, that principle feels even more urgent. From an existential perspective, we are all learning how to navigate a world where essentially every cognitive skill is being supplanted by powerful AI, which raises enormous questions about work, creativity, and the quality of human life. It also shows up in a very practical way when building a company where models improve, customer expectations change, and what felt impressive yesterday can become table stakes quickly. At Decagon, the pace of product innovation is astounding: new processes, new tooling, and new capabilities are introduced constantly. This kind of environment seems to reward curiosity and humility. The skills that brought Berkshire Hathaway through one decade would not have sufficed to get it through the next decade without the legendary duo being continuous learning machines. The same is true for our generation, and for us at Decagon.

The second idea is of deserved trust. Munger believed that the safest way to get what you want is to deserve what you want. Working with customers helping to build and deploy agents, I’ve learned that there is huge satisfaction to be obtained from getting deserved trust, where we want to deliver to the world what we would buy on the other end. More generally, trust, reputation and brand are amorphous, but are built over a long period of time through integrity of the team and continuous, high quality implementation for customers. That matters even more for AI application companies like Decagon. In a market where products can be built quickly and features can be copied, moats are harder to define. Scale, reputation, and stickiness of product are all moats. But the most durable of them, and the hardest to copy, is deserved trust earned through continuous, high-quality implementation.

The third idea is Munger’s mental checklist approach. Drawing on pilots and other high-stakes professions, Munger championed checklists as a defense against cognitive error and preventable mistakes. As complexity grows and as new hires join, relying on memory and word of mouth alone becomes unscalable. On the Agent Development team, we are working to build repeatable checklists and playbooks that ensures consistent quality of delivery, balancing that against the idiosyncrasies of each customer. The goal is not only to reduce errors, but to create a scalable system that evolves with lessons learnt and ultimately lower the effort required per successful implementation.

While it’s true that Munger — who did not own a laptop, avoided technology investments, and used an old-school flip phone — would unlikely have flirted with the idea of investing in AI, his menu of can’t-fail ideas still resonates with me. These are the ideas we bring to the Agent Development team at Decagon. We believe we’re tackling one of the most interesting problems in AI companies today, and we are building the best team in the world to solve it here at Decagon.

Interested in shaping the future of agent development? Explore careers at Decagon.