> The ones that will fall prey to the bitter lesson are those that start with an existing workflow, such as updating inventory data, and AI-ify it. The ones that will survive will leverage models’ unique, nuanced properties to invent new workflows that were not technically possible before. I call these “post-skeuomorphic apps.” **The trap:** Using new technology to just copy old forms, instead of exploring what the new technology uniquely enables. **Skeuomorphic AI apps** = Taking existing workflows and just adding AI to them. [[只包装基础模型能力的垂类 specialized AI 没有未来]] Base models will get good enough to do these tasks directly. Your "specialized" wrapper adds no lasting value. > The founders who will win are asking a different question: What becomes possible now? What work can we invent that only AI makes possible? Models have unique properties—they can coordinate with other models, learn from every interaction, and generate novel solutions to problems that don’t have predetermined answers. The winning applications will discover workflows that leverage these properties. We probably don’t even know what these workflows look like yet. > The apps that won broke this trap entirely. Uber didn’t digitize the taxi dispatcher’s desk. It asked: What becomes possible when everyone has a phone in their pocket that knows where they are? The phone became a remote control for your life, as investor **[Matt Cohler](https://www.forbes.com/sites/roberthof/2012/09/11/benchmark-vc-matt-cohler-mobile-ads-will-be-even-better-than-web-ads/)** [has said](https://www.forbes.com/sites/roberthof/2012/09/11/benchmark-vc-matt-cohler-mobile-ads-will-be-even-better-than-web-ads/) —for food (DoorDash), for rides (Uber), for groceries (Instacart). They didn’t adapt existing workflows. They invented new ones. similar:: [[New technology doesn't improve old systems; it makes them obsolete]] --- ## Build companies that could not have existed without AI In 2026, we’ll see the emergence of companies that simply could not have existed before recent model breakthroughs in reasoning, multimodality, and computer use. Until now, many sectors (such as legal or customer support) have used improved reasoning to *enhance* existing products. But we’re only now beginning to see companies whose core product capabilities are fundamentally enabled by these new model primitives.