If you're a tech junkie like myself, you may have scene the sharp rise in startups in the last 3 years. With tools like Claude code, replit, codex, and numerous other artificial intelligence tools, going from idea to MVP has never been easier. It's changed the tech conversation completely- whereas vying for a Google role just 10 years ago was all the rage, now building something yourself solving a problem you believe consumers face is where the craze is. And this is not at all a diss at the startup scene, I myself am particularly drawn to startup culture, but rather an analysis of where I think this sector is headed and why.

So lets talk about how this startup scene even came about.

how we got here

Honestly it wasn't one thing, it was like three things stacking on top of each other at the same time.

First, entrepreneurial incubators and pitch competitions are basically everywhere now, even in high school. Stuff that used to feel exclusive to MBA students or the Y Combinator crowd is now getting marketed to fifteen year olds. So the "I should be a founder" seed gets planted way earlier than it used to.

Second, LinkedIn stopped being the app you opened once a year before job season and turned into a full on personal brand machine. "Building in public" became its own content genre, and every tiny milestone, every fake funding round, every janky landing page turned into a post. The algorithm just kept rewarding it, so people started chasing the clout of looking like a founder more than actually solving anything.

And then the tools caught up to all of it. Vibecoding, basically describing what you want instead of writing every line yourself, took the biggest barrier (actually knowing how to code) and made it almost optional. Now anyone with an idea and a weekend can have something that at least looks like a real product.

Stack those three together and yeah, of course you get a flood of founders, a flood of MVPs, and a flood of "revolutionary" ideas solving problems that may or may not even be real.

but is it actually oversaturated?

This is where I push back a little on the doom takes. Yes there are more startups than ever, and yes a lot of them are just the same AI wrapper with a different color palette slapped on. But being crowded on the surface doesn't automatically mean it's crowded where it actually counts.

What's flooded is the idea and MVP stage. That part got commoditized, anyone can spit out an idea and ship a rough version of it now. What didn't get easier, and honestly might've gotten harder with all this noise, is distribution, retention, and actually getting a problem deeply enough to solve it well. The tools sped up execution, they didn't hand anyone good judgment.

So it's not that there are too many good companies. It's that there are too many attempts. Those are not the same thing.

what I think comes next

If literally anyone can build something now, "I can build it" stops being the flex. What actually matters is going to be the stuff that's way harder to shortcut:

**actually knowing your space. ** someone who deeply gets a niche industry or user base is going to beat someone with a shinier MVP and a surface level take on the problem

taste. when everyone's using the same AI tools, the thing that stands out is going to be actual design sense and product intuition, not just features

having distribution before the product's even done. the founders who win early are the ones with an audience or a wedge into a market already, not just whoever shipped first sticking around. most of these vibecoded startups are going to quietly die in a year because there was never a real problem behind them. whoever's still standing in 3-5 years is going to be the person who actually cared about what they were solving, not the person chasing whatever was trending

I don't think the startup scene is dying or slowing down at all honestly. I think it's just sorting itself out. The easy part got automated away, so now the only thing left to actually compete on is the hard part, understanding people, building trust, and sticking with something long enough to see if it actually works.

okay but zoom out, what's the next barrier that falls

Software was just the first domino though. If I had to guess what's next, it's whatever domain still requires a ton of specialized knowledge or expensive infrastructure to even get started in. that's basically the pattern every time, some field is gatekept by cost or expertise, then a tool comes along and knocks the floor out from under it.

A few candidates I keep thinking about:

hardware and physical prototyping. right now if you want to build a device you need CAD skills, know how to spec components, understand manufacturing tolerances, all of it. but AI assisted design tools are already getting good at generating schematics, simulating stress and thermals, and picking parts for you. pair that with cheaper 3D printing and small batch manufacturing and you get "vibecoding" but for physical objects. someone with zero EE background could plausibly spec a working circuit board the same way they vibecode an app today.

wet lab science. biology and chemistry research is still bottlenecked by needing years of training just to run an experiment correctly, plus access to expensive lab equipment. but AI is already speeding up literature review, hypothesis generation, and even experiment design. add in things like cloud labs where you send a protocol and a robot runs it for you, and you get a version of "idea to MVP" for actual scientific experiments. that's a much bigger deal than another SaaS wrapper because it touches drug discovery, materials science, all of it.

legal and regulatory work. a shocking amount of what blocks people from starting things (contracts, compliance, filings) is really just knowing the right process and the right language. that's exactly the kind of thing AI is good at compressing.

content and media production. filmmaking, animation, even game development used to require teams and studios. that barrier is already collapsing fast, one person can now do what used to take ten.

the common thread is this: whatever domain still requires years of specialized training AND expensive infrastructure just to take a first swing at it, that's the next thing that's going to get "vibecoded." and honestly the same thing that's about to happen to startups is going to happen there too. a flood of low effort attempts, followed by the real differentiator becoming taste, domain depth, and follow through instead of raw access.

so if you're trying to figure out where to plant a flag early, I'd look less at "what's hot right now" and more at "what field still has a dumb, artificial barrier to entry that has nothing to do with the actual hard part of the problem." that gap is where the next wave is going to come from.

that's the part AI can't vibecode for you.