Picture this: You are an ambitious college student in your dorm. It’s 2:00 AM, and you are frantically typing, convinced that you are working on the next big thing. You haven’t seen anyone for days. Your parents think you are the next Mark Zuckerberg. But in reality, you spend your days prompting on Claude Code and pitching your idea to investors, only to learn that your idea isn’t unique, and a hundred other college students also vibecoded the exact app.
Today, anyone can use Claude Code or Cursor to transform an idea into a fully functional prototype in a couple of hours. Coding is no longer a niche skill nerds only know; anyone can create something “impactful.” While it may seem that these tools are flooding the market with “great” startup ideas, the truth is quality has dropped, despite what LinkedIn posts and VCs say.
One key metric in the startup world is a company’s move from Seed to Series A funding. This phase is monumental, as it validates a startup’s idea from feasible to scalable. From Carta's cohort study, the median time from Seed to Series A funding was 18 months in 2021, and is now “roughly 28 months.” Not only has the time to Series A taken longer, but “only 15 to 18 percent of seed companies [today] raise a priced Series A within 24 months, nearly half the graduation rate of the 2018 to 2020 cohorts.” It is no coincidence that this timing lined up with the launch of AI coding tools like Cursor, which launched in 2023, and Claude Code in early 2025.
Consider Y Combinator (YC). While renowned for backing giants like Airbnb, DoorDash, and Stripe through mentorship and finding, the prestigious accelerator is also suffering a quality issue. Masha Bucher, the CEO of Day One Ventures, is a big investor in YC, with her firm investing in 35 companies from YC in the past couple of years, and scaled back during the pandemic “when she saw the quality of YC’s choice in firms go down.” This also coincides with the shared sentiment that YC’s batches are becoming generic, with it accepting overlapping businesses and so many B2B SaaS startups that seem to be “selling stuff to each other.”
The root cause is simple: founders are confusing the ease of building with the necessity of solving. When the biggest technical barriers to shipping software are grueling hours of debugging and learning new languages and architectures, ideas become scarce and heavily scrutinized. This barrier becomes a feasibility filter and forces us to ask a fundamental question before typing any code: Is this problem unique and impactful enough to justify the pain and effort of building this solution?
Today, that filter is gone. Although AI has eliminated the cost of writing code and prototyping, it hasn’t eliminated the other core components of a successful startup. Distribution, unit economics, customer retention, investor relations, and other essential parts of building a startup remain; they matter even more today. Spinning up a sleek dashboard with Claude over the weekend may feel like a milestone, but without customers, it’s just bloatware.
But zero-barrier coding also presents a bigger challenge: defensibility. Technical moat has never been more vulnerable; when everyone has access to the same tools to build, your “unique” idea suddenly becomes noise in a crowd full of prospective builders and problem solvers. Enterprises can replicate your own product using the same tools, rendering countless thin software wrappers obsolete overnight. We have built a culture and ecosystem where thousands of the best builders are building the same automated meeting note apps with minor tweaks and no defensibility.
The path out of this trap is simple: stop building quick software services. If we want the next generation of startups to be impactful and out of a funding slump, we need to stop rewarding AI wrappers and start funding messy ideas that are “moonshots.” VCs should not be spreading their eggs in the same basket of AI wrappers, as it's a zero-sum game. No one wins when everyone builds the same wrapper that can be replaced in one afternoon.
Builders should also rethink building as a whole. If you can build an app over a couple of hours, close it and delete it. Instead, shift your focus to messy, ugly environments, like physical robotics, power grids, and hard sciences, where your value proposition doesn’t depend on LLMs alone.
Close the terminal, shut down the laptop, and touch some grass. Talk to real people, see what the true pain points are, uncover problems that don’t require a new form of automation to solve, and remember that startups are non-linear. Building is no longer the hard part; finding something that actually matters is.