Blog
I write about what AI changes and what's worth building. A few AIs help me think it through.
The new timetable for startup growth
Modal turned demand for running AI-generated code into a $300M run-rate product. Growth across AI companies is changing what venture investors consider ambitious.
The future of software is personal
Experts could use software to deliver services to people they cannot currently make time for. The customer buys completed work adapted to their circumstances.
What I learned from Instinct
About a year into building Instinct, its founder has a company valued at $10 billion. What can we learn from its approach in a market the largest AI companies also want?
The engineering case for private equity
Buy a business people already depend on, fund the engineering to rebuild its operations, and grow what it can deliver.
Building while AI keeps changing
Using Jev changed our implementation. The harder planning question is which ambitions new capabilities make worth attempting again.
The future of education
I want an expert's AI to help me get the work done and understand the decisions along the way. That mixture could change how people learn.
What becomes worth automating
Cheap, frequent judgments could let assistants and services respond while the work is still changing. Jev gives me a way to try that product behavior.
Who gets paid when agents work?
Stripe's five stages of agentic commerce, the new personal agents, and who gets paid as software starts choosing what to buy.
We are still early
Uneven AI adoption in Brazil leaves practical work to do. A guitar-practice example explores who we can reach, what stops them, and how to choose a first useful tool.
Where to compete in AI
Better models make new businesses possible. The strategic question is what serving customers lets you build that competitors would take time to reproduce.
What software becomes
I expect software to take shape around the work: a question becomes a chart, a follow-up becomes a simulation, and the next job starts with what the last one taught it.
Giving agents a model of the world
Planning a dinner or checking a discount depends on relationships. When should an agent infer them from context, and when is it worth preserving them explicitly?
What the agent walks into
Why AI agents need reliable context, consistent records and clear authority to act inside a business or a person's daily life.