One of the biggest mistakes I made in the early days of Pangea was trying to build the perfect product before we had users who cared. We were opinionated about what the world wanted, and we invested heavily in custom software when we could have shipped something faster and simpler. That instinct — to build it right the first time — cost us years.
The lesson I eventually learned, and the principle I operate by now, is this: ship fast, then systematize. Get something into the world, see if anyone cares, and only then invest in building the systems to scale it.
Speed is a feature. Y Combinator hammered this home. The entire batch is structured around one question: what can you accomplish between now and Demo Day? Pick a KPI, grow it week over week. That forcing function compressed our decision-making and cut through the noise. But the insight isn't just about startups in an accelerator — it's about how you should approach any new project. The default mode for most builders is to over-engineer before they have signal. Resist that.
But speed without systems creates chaos. Here's the tension: if all you do is move fast, you accumulate technical debt, operational debt, and organizational debt that eventually buries you. The companies that scale are the ones that know when to shift from exploration to systematization. You ship the scrappy version to prove the concept. Then you build the dbt transformations, the automation workflows, the dashboards, the proper architecture. The order matters.
AI has changed the equation. This principle has taken on a completely new dimension with AI coding tools. Right now, I run three different projects with nine active work streams. I'm reinventing Pangea's entire tech stack — front end to back end — using AI coding agents to move faster than a traditional engineering team could. I'm building in Claude Code, shipping production features, and iterating at a speed that would have been impossible even a year ago.
What this means is that the "ship fast" part has gotten dramatically faster. You can go from idea to working prototype in hours, not weeks. But the "systematize" part is just as important as ever. AI lets you generate code quickly — it doesn't automatically give you clean architecture, reliable data pipelines, or thoughtful product decisions. You still need the judgment layer.
The framework I use now: Start with the smallest thing that could possibly work. No feature flags, no configurability, no abstraction layers. Just make it work. Talk to users immediately — not after launch, during development. The best product decisions come from watching someone use what you built and listening to what they say. Once you have signal, systematize ruthlessly. Build the data infrastructure, write the automations, create the dashboards. If you do something twice, automate it the third time. And keep your hands in the product. I still talk directly to customers and push code to production in the same day. That's not a phase — that's the model.
The future of building belongs to people who can move between creation and systematization fluidly — who know when to be scrappy and when to be rigorous. AI makes the fast part faster. But knowing what to build, and when to build it properly, is still the hard part.