We're at the moment where every company with a product roadmap is asking the same question: Where do we bolt on AI? The answers have been predictably messy.
This week alone, we've seen announcements about AI-powered everything from streaming devices to delivery logistics. Some of it matters. Most of it doesn't. And in the next 18 months, a lot of it will quietly disappear from marketing materials and feature lists.
Here's my take: The winners in AI-powered products won't be the ones chasing feature parity or the ones shouting loudest about their transformer architectures. They'll be the companies that use AI to subtract, not add. The ones that make products simpler, faster, or less annoying to use.
The current landscape is cluttered. Product teams are layering AI onto existing features like it's a marketing requirement rather than solving an actual problem. Does your smart home system need an AI assistant that can discuss philosophy? Probably not. Does it need one that understands what you actually want when you say "make it cozy"? Maybe.
This is where the real work happens. Not in the model training. Not in the press releases. In the unsexy territory of product design where someone has to decide: What is this feature actually for?
Look at the products that gained genuine adoption in recent years. The good ones typically removed friction. They took something annoying and made it painless. When a new technology becomes the answer to "how do we jam more into this product," it usually fails. When it becomes the answer to "how do we eliminate the frustrating part," it tends to stick around.
The AI product category is currently in the feature-bloat phase. Every company wants to demonstrate they're not behind. So they announce AI capabilities that sound impressive in a product brief but create confusion in actual use. Users don't want to learn new interfaces or understand how the AI works. They want better results with fewer steps.
There's also a durability question that people aren't talking about enough. Products built on complicated AI stacks are fragile. A Framework laptop with a BIOS update that bricks the machine is a straightforward hardware problem. But imagine a product where AI optimization goes wrong or a model needs retraining or an API dependency breaks. The support matrix becomes a nightmare. The companies that win will be the ones that keep their products maintainable.
Simplicity is also an antidote to the hype cycle. Right now, venture capital and media are obsessed with how big the AI opportunity is. But products don't sell on opportunity size. They sell on utility. The companies that resist the urge to claim their AI does everything will age better. They'll iterate more cleanly. They'll build trust instead of managing disappointed expectations.
There's a financial angle here too. Bloated product features are expensive. They require more engineering, more testing, more customer support. Companies that pare back to the genuinely useful AI capabilities will have cost advantages. They'll also have simpler pricing models, which means simpler sales conversations.
The next wave of product companies to watch won't be the ones with the flashiest demos. They'll be the ones shipping something boring that actually works better than it did before. The ones that added one AI feature, made sure it was bulletproof, and stopped there.
That's not a contrarian take in venture capital. It's contrarian in the current moment, which means it's probably right.