4/ The most trustworthy AI system won't have the model we trust most. It will be the one designed so we don't have to trust the model at all. My riff on Satya Nadella's essay, "Models as Insider Risks in the Super Intelligence Era"
4/ It has worked every morning since. The insight: synergies inside one ecosystem, where scheduling, messaging, and contacts compose into something more powerful than any one of them alone.
4/ If you run a company, every AI IQ you can put to work expands what you can accomplish. The more tokens you have, the more software you can build, and the more problems you can tackle. Part of the competitiveness equation may simply come down to token volume.
4/ The half that didn't: the assistant. My 2008 fantasy had Google caring for you: the water bill reminder, grandma's birthday, the cookbook she'd love. That never shipped. In 2008, total memory required a Google. Today it fits on a laptop. Time to build the missing half.
1/ Your personal agent will remember your life better than you do. What you ate, how you slept, the stress building for weeks. Part psychologist, part personal medic. That's the memory thesis: in personal agents, memory matters most. This series is about building it. 🧵
3/ The half that came true: the memory. Your location history, every search you've ever made, every photo, every email. One company really does hold a near-complete record of your digital life.