Before You Build
Find out if someone wants what you are building before you write another line of code.
The situation
More than one in three new startups is now solo-founded. That number doubled in five years. AI collapsed the cost of building software to near zero — a solo founder can ship a full-stack product on a hundred dollars a month. The bottleneck moved. It is no longer "can I build this?" It is "should I build this, and for whom?"
I researched this. I pulled the data from Carta, the Census Bureau, Stack Overflow, and Anthropic's own reports. The numbers are real: 36.3% solo-founded startups, $1.8 trillion in solo business receipts, coding 60–70% faster with AI assistance. Building has never been cheaper or faster. And the failure rate has not changed. Because the failure was never about building.
AI scales faithfully. It will amplify whatever you give it with total fidelity and zero judgment. The skill that matters now is not the building — that got cheap. It is the intentionality about what gets built and for whom.
The honest question
Answer this before anything else: what do you actually want from this project?
There are only three honest answers:
- I want to build it because building is fun. Great. Keep building. You do not need this section. Skip to Go Deeper and sharpen your craft.
- I want people to use it. Then someone besides you needs to see it and tell you whether it solves a problem they actually have. That is what the rest of this section is about.
- I want people to pay for it. Then you need everything in answer two, plus evidence that the problem is painful enough to pay to solve. The bar is higher and the questions are harder.
Most builders never make this distinction explicit. They drift between "fun project" and "startup" without ever committing to either, which means they never do the work that either path requires. Pick one. You can change your mind later, but you cannot validate a product you have not admitted you want to be a product.
Who is this for?
Not "who could use this?" — that question is too generous. Everyone could use most things. The question is: who has this problem badly enough that they are already doing something about it?
Signs you have a real audience:
- They are already paying for a worse version of what you are building
- They are cobbling together spreadsheets, manual processes, or duct-tape workflows to solve this
- They complain about this problem unprompted — in forums, in communities, in conversations you did not start
- They have budget allocated to solving this category of problem
Signs you are imagining an audience:
- You can describe the problem but cannot name five real people who have it
- The people you have told about it say "that is cool" but do not ask when they can use it
- You are building features before you have talked to a single potential user
- Your best evidence is "I have this problem myself" and you have not checked whether anyone else does
What are you actually selling?
One more question before you validate: what role does AI play in what you are building? This shapes your pricing, your risk, and what you need to validate.
AI as tool. You used Cursor or Claude Code to build your app. Your user never sees AI. They are paying for the outcome: a scheduling app, a CRM, a marketplace. The AI was your power tool, not theirs.
AI as feature. Your product does something bigger, and AI handles one piece. Smart search in an e-commerce app. Auto-categorization in a bookkeeping tool. Remove the AI and the product still works, it is just less good.
AI as core. Remove the AI and there is no product. A legal document analyzer. A personalized tutoring system. Kill the model, kill the product.
If AI is your tool, you are selling software. Validate the way any software product validates: does someone want this outcome? If AI is your core, you have an additional constraint: per-request costs. Every API call costs money. A typical interaction runs about a penny, but at 1,000 uses per user per month, that is $10 you need to cover. When AI is your tool, a hallucination is your problem during development. When AI is your product, a hallucination is your customer's problem in production. Know which game you are playing. For the full cost breakdown, read The Math Never Works.
What kills you even if people want it
The wrapper trap. What happens to your product if Anthropic or OpenAI ships the exact feature you are selling? A wrapper is a product where the entire value creation happens inside someone else's model, and your contribution is a UI layer and maybe a system prompt.
The question is not whether you are using third-party AI. Almost everyone is. The question is whether you have built anything on top of it that the provider cannot trivially replicate.
Four defenses against being a wrapper:
- Proprietary data. Your users' data makes the product better over time, and that data cannot be replicated. After 1,000 users for six months, is your product meaningfully better than day one? If not, no data defense.
- Workflow integration. A chatbot is easy to replace. A tool embedded in someone's CRM, triggering on pipeline stages, feeding their reporting dashboard? That is much harder to rip out.
- Domain-specific engineering. Anyone can write a prompt. Not anyone has spent 200 hours refining prompts for a specific domain with eval sets covering dozens of edge cases.
- Multi-model orchestration. If your product only works with one model from one provider, you are dependent on their pricing, capabilities, and strategic decisions.
You do not need all four. But you need at least one, and two is better. For a deeper framework on choosing your differentiation, see Pick Two.