For the better part of two decades, I've been telling people the same thing: ideas are cheap.
I had a whole framework for it. Of any given idea, at least a thousand people are thinking about the same thing right now. A hundred of them have actually started working on it. Ten of them shipped something. One of them made it a success.
The point was always: stop worrying about someone stealing your idea. Stop being precious about it. The idea is the easy part. Execution is the hard part. Execution is everything.
I believed this deeply, and I was right — for that era.
I'm not sure I'm right anymore.
What changed
After being immersed in the AI revolution for more than three years, I've been watching something happen that makes me revisit this belief at a fundamental level. The cost of building software is collapsing.
Andrej Karpathy coined the term "vibe coding" in February 2025 — describing a way of building software where you talk to an AI in plain English and let it write the code. It sounded like a novelty. Collins Dictionary made it their Word of the Year. That should have been the signal.
The numbers that followed are hard to dismiss. App Store submissions grew 30% in 2025 to nearly 600,000 — the first meaningful increase since 2016. By Q1 2026, submissions surged 84% year-over-year. Apple is now processing over 200,000 app submissions per week. The vibe coding platforms behind this wave — Codex, Claude Code, Cursor, Lovable, Replit, Bolt — are not side projects. Cursor's parent company is valued at $29 billion. Anthropic raised $30 billion in Series G funding at $380 billion post-money valuation early this year.
These aren't toy numbers. And this isn't just about coding. Generative AI has collapsed the barriers across the entire product creation spectrum — design, content, marketing, and growth. Things that used to take a team of ten and six months can now be done by two people in a weekend.
The inversion
Here's what I think is actually happening: for decades, the equation was simple. Ideas were abundant. Execution was scarce and expensive. So naturally, execution was where the value concentrated.
AI is inverting this equation. When execution becomes fast, cheap, and increasingly commoditized, the scarcity shifts. What becomes rare, and therefore valuable, is knowing what to build.
Not just having an idea. Knowing which idea. Knowing who it's for. Knowing what to leave out. Knowing the difference between a feature people say they want and a product people will actually pay for.
This is what some people are calling "sense" or "taste." I think that's the right word. When anyone can build anything, sense and taste become the moat.
The evidence is in the slop
The App Store flood is actually the strongest proof of this thesis. More apps are being built than ever. But developers and users alike are complaining about the quality. Critics call it "app slop": a wave of low-effort, AI-generated applications that technically work but solve no real problem, delight no real user, and add no real value.
When building was hard, the difficulty itself was a filter. You had to care enough about your idea to invest months of effort. That friction forced a certain level of validation before you ever shipped anything. You had to believe in what you were building because building was expensive.
Remove that friction, and you remove the filter. What you get is an ocean of software built because it could be built, not because it should be. The signal-to-noise ratio craters.
This is exactly the pattern I wrote about in my last post — people building things because they can, not because they should. The trap of capability, at a civilizational scale.
Updating the framework
So let me update my old model.
The original: 1,000 people have the idea → 100 start building → 10 ship → 1 succeeds.
The 2026 version: 10,000 people have the idea → 5,000 start building (because now they can) → 2,000 ship something → 10 build something anyone cares about → 1 succeeds.
The funnel didn't just change at the top. It exploded at the top and middle, while the bottom stayed roughly the same - that's the most interesting part. More people are building than ever. The number of things that succeed hasn't changed much. What changed is where the filtering happens. It used to happen at "Can you build it?" Now it happens at "should you build it?" and "did you build the right thing?"
The bottleneck moved from execution to judgment.
What this means in practice
If you're a builder, a founder, or a product person, the implication is significant: spend more time thinking and less time worrying about shipping speed. Speed is now table stakes. Everyone is fast. The advantage goes to the person who is fast and right.
A few things I'm telling myself:
Validate harder before you build. The old advice was "just ship it." The new advice is "make sure the thing you're about to ship in two days is worth shipping at all." When building is cheap, the cost of building the wrong thing isn't the build — it's the opportunity cost and the noise you add to an already crowded market.
Develop taste deliberately. Taste isn't mystical. It's accumulated judgment: built through exposure, critical analysis, and a willingness to be specific about what works and what doesn't. Study the products you love. Understand why they work. Pay attention to what you reach for versus what you scroll past.
Go deep on problems, not solutions. AI makes solutions cheap. The hard part is understanding the problem well enough to know what solution actually matters. Talk to users. Live in their world. The best product insight I've ever had didn't come from a brainstorm; it came from watching someone struggle with something I thought was already solved.
Protect your differentiation. When anyone can build a clone of your product in a weekend, your product isn't your moat. Your understanding of the problem, your point of view, your curation of what matters — that's your moat.
So, are ideas still cheap?
Ideas were never actually cheap. We just couldn't tell, because they were bundled with execution. When building something required so much effort, we attributed all the value to the building. The idea was invisible.
AI has unbundled ideas from execution. And now that we can see them separately, it turns out ideas — real ideas, the ones rooted in genuine insight about what people need and why — were always the scarce resource. We just couldn't see it through the fog of execution costs.
So no. Ideas aren't cheap. They never were.
We just finally have the tools to prove it.