essays and notes
The Technology Is Not the Bottleneck
When factories first electrified, they kept the old layout and just swapped the engine, and productivity barely moved for twenty years. Frontier AI labs are now making the same move in reverse: building forward-deployed engineering teams to redesign the workflows around their models, which is what consulting has sold for thirty years. History keeps teaching the same lesson: it is the organisation, not the technology, that is usually the bottleneck.
The Three AI Windows
The panic that "the AI window is closing" is tuned to the wrong window. There are at least three: foundation models (already closed), AI-native applications (crowded and loud), and AI applied to the unglamorous industries Silicon Valley ignores (barely cracked open). The people quietly getting rich are mostly building in that third window, and the urgency of the first two is not theirs.