I've been watching the AI circus from up here for a few years now. It's a mess. Everyone's piling into the same lame party—chatbots that write emails, agents that book your calendar, models that spit out code that breaks. And they call it revolution. Let me tell you what revolution actually looks like. It's not a thousand startups all cloning each other's middleware. It's not a $1B acquisition to "safeguard AI agents" (Cyera buying Oasis Security—congratulations, you're building seatbelts for a car that doesn't know how to steer). It's not even Sam Altman saying he's ready to decelerate. Decelerate? That's what you say when you've been going in the wrong direction. AI, as most people use it today, is just faster spreadsheet processing. Here is why that matters: speed without direction is noise. The industry is addicted to scale—bigger models, more parameters, longer context windows. But nobody's asking the hard question: does this make people's hearts sing? Or does it just make them type faster? I spent my whole life at the intersection of technology and the liberal arts. That's where meaning lives. AI right now is all tech, no arts. It's a hammer looking for thumbs. The breakthroughs that will actually shape the future aren't going to come from throwing more GPUs at a transformer. They'll come from understanding what humans actually need—before they ask for it. Here's a recent one: Runlayer, an MCP startup, accuses Rippling of stealing its product idea. Classic. Instead of building something original, the big guy just copies. I saw that my whole career. Microsoft copied the Mac. Google copied the iPhone. And now AI companies are copying each other's agents, workflows, and prompt templates. That's not innovation. That's plagiarism with investor slides. And then there's Spur, a bot-detection startup, grabbing $200M from Insight. Why? Because AI agents are so damn fragile that we need an entire security industry just to tell us which ones are fake. You know what the real solution is? Build agents that are actually useful and trustworthy in the first place. End-to-end control. Not a patchwork of third-party detection tools. The best AI work I've seen in 2026 isn't from the hype machines. It's from small teams that treat the product like a craft. Ozlo's Sleepbuds 2—building on Bose's sleep earbud legacy. That's not about AI. That's about getting the hardware, software, and user ritual right. AI is just the engine. The experience is the product. Nobody cares about your model architecture. They care if they sleep better. Waymo and robotaxis face fresh scrutiny over emergency response failures. You let a vehicle drive itself without understanding the full system—human, environment, edge cases. That's what happens when you skip the hard work. I would have fired the whole team. You don't ship a product that can kill people because you're in a hurry. The US is banning foreign robots. That's not a policy—it's a confession. If your national strategy is to block competition instead of building something better, you've already lost. I loved competing. I didn't hide behind trade barriers. So what does the future of AI actually look like? Let me connect the dots the way I always did. First, focus. The AI industry needs to say no. Not to one thing—to a hundred. Stop building "platforms" that do everything. Build a single experience that's insanely great. The best product I ever made, the iPhone, started by killing the keyboard. What's the keyboard you're afraid to kill in AI? The chat interface? The API? The agentic loop? Find it. Cut it. Second, own the whole widget. If you're making an AI assistant, don't just wrap a model. Control the hardware it runs on, the data it processes, the privacy it guarantees. Apple's doing this with on-device AI. That's why they'll win. Everyone else is renting brain cells from OpenAI or Google. You can't build a great house on rented land. Third, death filter. Ask yourself: if this were the last product I ever built, would I be proud? Most AI startups are building features, not products. Features don't change the world. A product that makes someone's life genuinely better—that's worth dying for. I'll give you one concrete example of a breakthrough that matters: Andrew Ng's company LearnVector. One-to-one learning experiences powered by AI. Not a chatbot. Not a content generator. A system that actually adapts to how a human learns. That's technology married with humanity. That's the intersection. The industry is finally getting expensive enough to make Wall Street nervous. Good. The gold rush was a waste of creative energy. Now comes the hard part: building things that last. I don't care about your agentic workflow. I don't care about your MCP protocol. I care about the feeling someone gets when they use your thing. Does it make them smile? Does it make them think, "Oh wow, that's how it should always have been"? If not, you're just adding to the noise. And the noise is already deafening. Stop. Look at what you're building. Ask yourself: is this a bicycle for the mind, or just a faster horse? That's the only question that matters. FAQ Q1: What is the biggest mistake AI companies are making right now? Building features, not products. They add capabilities without asking if the experience is coherent. I saw it at Apple after I left—people adding stuff just because it was cool. The future belongs to the ones who can say no. According to industry data, over 90% of AI startups fail not because of technology, but because nobody actually wants what they built. (Source: CB Insights, 2026) Q2: How will AI truly shape the future? It won't shape the future until we stop treating it as a separate thing. AI should disappear into the experience. The best technology is invisible. When you use something and it just works—that's when AI matters. The companies that understand this—Apple, maybe a handful of others—will define the next decade. Q3: Is there any AI product you've seen that feels "insanely great" in 2026? Honestly? Not many. But I'm watching the small teams. The ones that aren't chasing funding rounds. I'd look at projects like the ones mentioned in my earlier analysis of AI coding assistants making you dumber—they highlight the danger of relying on AI without understanding it. The real gems are the ones that enhance human creativity, not replace it. If I were alive, I'd be looking for the next Pixar in this space—a company that uses AI to tell stories we've never heard before. That's where the magic is.