Short notes on business and AI.

How Lego grew up: the power of experimenting when you're already winning.

Lego posted record sales last year with net profit up 21%, driven by strong growth across Europe and the Americas and emerging gains in the Middle East and Africa. The company kept pouring capital into infrastructure — new facilities in Vietnam, Virginia, Hungary, Mexico, and China. By every metric, Lego is in a strong position.

The interesting question isn't whether Lego is winning. It's how a company already dominant in its core markets unlocked another lane of growth. The Icons and Botanical collections are a large part of the answer: both have been remarkably successful in reaching adult consumers — premium Icons sets, gift-ready Botanical sets — positioning Lego against flowers and plants in the gifting category.

There's precedent. Call of Duty found massive commercial success despite an M rating; Catan helped trigger an adult board-game renaissance. Lego joins a list of brands that expanded their ceiling by rethinking who their audience actually was.

The lesson I keep coming back to: Lego didn't need Icons or Botanical to survive. They were already dominant. But the willingness to experiment from a position of strength — not just when struggling — unlocked growth that playing it safe never would have. The most useful version of experimentation isn't desperation. It's discipline practiced when nothing is forcing your hand.

Read the WSJ article →

AI is reshaping education, and the deals schools are making.

At UVA there are still plenty of professors who disagree with any use of AI in learning. But across higher education, universities are starting to embrace the changes AI is bringing.

Anthropic's partnership with CodePath is a clear example: the maker of Claude is integrating its products into the AI courses CodePath runs across more than 1,000 institutions. People keep asking what effect AI will have on education. It's a fair question. I think these institutional partnerships can be a good thing, and here's why.

Students already had access to paid AI tools before these deals existed, and many were using them. My peers build study guides from PowerPoint decks, upload pictures of incorrect answers to understand why, and use AI across humanities and STEM courses alike. When a paid tool becomes widely used and the school doesn't provide it, an equity issue shows up. Some students can't afford a premium subscription, so they're stuck with the limits of the free plan. That gap matters.

This is similar to tutoring. Not every student can pay for private tutors, which is one reason UVA's engineering school provides free tutoring to all students. If AI tools are becoming part of how students learn, then access to those tools should be considered the same way. AI companies offer these tools to universities because they want future users — but institutions can treat the partnerships as an opportunity to build a more equitable campus. That's why these deals can be a win-win.

Read the WSJ article →

OpenAI shuts down 4o: what it means for human–AI relationships.

OpenAI recently retired its 4o model, much to the dismay of many users. 4o had carved out a niche through its persistent agreeableness — something many users became overly attached to. I'm not sure exactly how human-AI relationships will play out over the next decade, but this decision raises a few important questions:

1. Will chatbot companies be held liable for harm caused by poorly aligned models?
2. What responsibilities — and what policies — will AI companies adopt regarding the relationships their technology fosters?
3. Is there a future where this becomes commonplace as AI moves toward more human-like emotional response?

This is the first time human-AI relationships have become a mainstream point of discussion. It won't be the last. These questions will only get more important — and more nuanced — as the technology evolves.

Read the WSJ article →