The most dangerous skill for early-stage founders is the ability to build anything.
I learned this by doing it wrong myself, then watching dozens of others do the same. You start a company with a product idea. Let's say it's a semantic search engine. But instead of finding out if anyone wants it, you start building infrastructure. Auth systems. Email pipelines. Deployment automation. Six months later you have an impressive technical achievement that nobody uses.
Here's what you should do instead: buy everything.
Auth? Clerk. Email? Resend. Payments? Stripe. Even your core feature, if someone else built it, use theirs. Vector database, embeddings model, the whole semantic search stack. Rent it all.
This feels like cheating. That's the point.
Your goal pre-PMF isn't to demonstrate technical prowess. It's to test whether anyone wants what you're making, as fast and cheap as possible. Every hour building infrastructure is an hour not learning if your product matters.
The cost curves are perfectly aligned with this. Third-party services are cheap when you're small and expensive when you're big. Which means they're cheap exactly when you can't afford to spend time, and expensive exactly when you can afford to spend money.
When you hit PMF and those services get expensive, that's when you build. Not before.
But what about vendor lock-in? What about domain expertise? What about moats?
None of that matters yet. Vendor lock-in is a problem for successful companies. Domain knowledge matters when you're sure you're in the right domain. Moats matter when you have something worth defending. Pre-PMF, you have one problem: finding out if you should build anything at all.
Here's a test: could you rebuild your product in two weeks using only third-party services and glue code? If not, you've built too much.
The counterintuitive thing is that this doesn't just help you move faster. It helps you think better. When you're not distracted by implementation details, you can focus on the actual problem. What do users want? How do they want to interact with it? Does this solve their problem?
After PMF, everything changes. Now you know people want this. Now you have revenue. That $100/month service now costs $10,000/month at scale. Time to build.
But building after PMF is easier. Not technically, it might be harder technically. But you know it's worth doing. You're not building in the dark. You're building something people already pay for.
The companies that die aren't the ones using Clerk for auth. They're the ones that spent three months building a custom auth system before finding their first real user.
So don't build. Buy everything. Get to market in two weeks, not six months. Learn if anyone wants what you're making.
Then, if they do, build.
[This obviously doesn't apply to hard tech companies where the technology itself is the product -- nuclear reactors, semiconductors, new materials. Though even they might benefit from this thinking when it comes to validating what problem they're actually solving.]