Over the past few weeks I’ve been thinking a lot about an essay by Andrew Chen, one of the most respected investors in Silicon Valley. His argument is simple, but powerful: AI doesn’t just change what startups build — it changes how they are built.

When I started working with founders, one of the toughest hurdles was always the first product. Getting from idea to prototype required a team of engineers, months of effort, and a lot of capital. Today, a single founder with a handful of prompts and low-code tools can build something in weeks that used to take an entire team half a year. That speed fundamentally shifts the logic of the early stage.

But here’s the catch: when everyone is equally fast, speed stops being a moat. Launching quickly is now the baseline, not the differentiator. What really matters is whether you are building the right thing — something that delivers lasting value, earns trust, and keeps users coming back.

This is where Chen’s warning about “AI wrappers” comes in. A thin layer on top of an API might impress for a moment, but it’s easy to copy and just as easy to forget. Real defensibility comes from proprietary data, deep integration into workflows, or solving a problem so painful that your solution becomes indispensable.

Teams will look different too. Instead of ten developers grinding out code, you might see two or three people orchestrating AI-driven workflows. New roles are emerging — prompt engineers, AI product managers — while others are under pressure, especially junior coding positions. Leading these lean teams will require a culture that is both experimental and disciplined, with a sharp eye on data and feedback.

And then there’s retention. Many AI apps dazzle users once or twice, only to be abandoned. Retention isn’t a by-product; it has to be designed in from day one. Ironically, AI can also help here — through personalization, instant feedback loops, and real-time testing.

Finally, we can’t ignore ethics. Bias, transparency, data privacy — these aren’t abstract academic debates. They’re make-or-break issues. Startups that ignore them in the rush to ship will find themselves facing much bigger problems later. It’s far easier to build fairness and trust into your DNA early than to retrofit them after the fact.

So, what will startups look like in five to ten years? I expect the biggest shifts in industries with massive data and broken processes: healthcare, education, finance. AI won’t solve everything — not every problem is an AI problem. But the playing field is changing. Success will belong not to the fastest builders, but to those who combine speed with clarity, trust, and execution.

And that, I think, is both the challenge and the opportunity: AI lowers the cost of building, but raises the bar for what’s worth building.