I started investing in the pre-generative AI era. Back then, the primary hurdle for zero-to-one founders was speed. For “non-technical” founders almost impossible to build a tech startup.

Building a prototype took weeks, if not months, of gruelling engineering.

Many founders stayed in the “thinking” phase too long simply because the barrier to “doing” was so high. Today, the pendulum hasn’t just swung; it has snapped to the other side.

With LLMs and NLP-based coding tools, there’s much good news. The CTO hurdle is no longer really an issue for basic applications in the zero to one stage. You can now prototype very easily.

A founder can spin up a semi-complex application in a weekend and while this is generally good news, this efficiency has birthed a dangerous paradox:

it has never been easier to build something that nobody wants.

The “Castle in the Sky” Syndrome

I’m seeing a recurring pattern in the zero to one stage: robust, seemingly polished prototypes that look incredible during a demo but suffer from three fatal flaws:

1. The Missing User: The product is often built “head-to-code.” It’s a brilliant reflection of the founder’s vision, but it hasn’t yet been stress-tested by a customer’s messy, real-world workflow.
2. Too Many Features: Because AI makes it so easy to add “just one more thing,” prototypes often try to solve five problems at once rather than mastering one.
3. It’s Actually Expensive: While these tools feel “free” or cheap, the time spent iterating in a vacuum is expensive. It’s easy to get caught in a cycle of building for the sake of building, which can unintentionally delay true value creation.

I call these Castles in the Sky. They are elaborate, aesthetically pleasing, and technically functional, but no one wants to live in them.

The Dopamine Trap

There is a psychological danger in overbuilding.

Every time a coding assistant spits out a working block of code, the founder gets a dopamine hit. It feels like progress. It looks like a startup. You feel like a builder.

However, the more time, effort, and emotional capital you invest into a “perfect” pilot, the less likely you are to listen when a customer tells you it’s useless.

You become protective of your castle. You unintentionally delay customer engagement because you don’t want to hear that your beautiful towers need to be torn down.

The Hard Truth: A clunky, ugly tool that solves a burning pain point for one customer is worth infinitely more than a “perfect” application that sits unused.

Three Considerations for the NLP Coding Era

If you are a founder leveraging AI to build your MVP, you must stay grounded. Here is how to avoid the trap:

• Build WITH, not FOR: Don’t go into a dark room and emerge with a finished product. If a customer hasn’t seen the wireframes or the messy “v0.1,” you are building for them based on a guess, rather than with them based on a need.
• Embrace the “Fail Fast” Ethos: The advice from ten years ago is more relevant than ever. Since building is now "cheap”, your “failed” experiments should be even faster and cheaper. Don’t let the ease of coding lure you into a long development cycle - if it doesn’t work, toss. Try again, and again. Till you find a true problem the market is pulling you towards.
• Prototyping is not Production: NLP coding platforms are world-class for validation. But as you move toward a production-level enterprise solution, you still need a rigorous technical foundation. AI can write code, but in itself without some technical leadership cannot yet architect a scalable, secure company that sells to customers at scale.
• Keep it crazy simple: The tendency to overcomplicate your initial build is real. It will take some real restraint to fight the impulse to “add one more amazing feature”. But at the earliest stages, keep it simple. Find one ideal customer profile, build one feature, let them test it and give you real feedback. Go from there.

Build for the problem, not for the thrill of the build.

At Openseed VC, we back the builders experienced and outlier operators on day zero, and bring an operator network to support from zero to one. Tell us what you’re building on www.openseed.vc/apply

Till next time.

Sincerely,

Maria Rotilu