Lessons

The vibe coding tools made building so easy, and shipping without looking is where people are walking off a cliff. Here's the gap between "it runs" and "it's done," and how to close it.

People are getting too lazy to write their own posts, so more of the internet is AI-generated every year. The next generation of AI trains on that content. So what happens when AI starts learning from itself? It has a name, and real research behind it.

Okay somebody needs to say this out loud because everybody building their little SaaS with Claude Code right now is having a great time and not thinking about the obvious problem sitting right in front of them… The same tool you used to build your SaaS in a weekend? Your customers have access to it too...

Here's a thought experiment worth sitting with for a minute. Imagine you wake up tomorrow and every AI coding tool is gone. Not down for maintenance, not rate limited, not expensive, just gone. Claude Code, Codex, Cursor, all of the LLM-powered tools that a huge number of solo founders now depend on, simply don't exist anymore, and there's no indication they're coming back. What would you actually do?

One of the strange consequences of being able to ship platforms in a week with Claude Code is that you also have to develop a much faster sense of when a platform isn't working, because the old model where you spent months building something and felt obligated to give it a year before declaring it dead doesn't really apply anymore...

Most solo founders start their first platform on a managed hosting service like Vercel, Render, Heroku, Railway, or one of the other platforms that have made deployment dramatically easier over the last few years...

I've shipped more than ten platforms in the past 3 months, and one of the things that's quietly happened over that stretch is that I've developed an actual decision framework for whether a new platform idea is worth building...

Let's get honest about something that doesn't get talked about enough in the vibe coder world, which is what happens when the tool goes down, when the API has issues, when the model gives you something that breaks production at 2am and realizing you don't actually know what any of it means. These moments are going to happen to anyone who builds with AI coding tools, and how prepared you are when they happen is the difference between an inconvenience and a real problem...

Before I get into this, I want to say upfront that I have nothing against dev shops. Some of them are great. They have real talent, they ship real work, and if you have the resources, they are absolutely the right call. This post isn't about whether they should exist. It's about what I learned when I actually went out and got quotes for a few of my ideas, and what happened when I tried to build the same things myself...

I ship a lot of sites. Most of them are built with Claude Code, sometimes in a weekend, sometimes over a few weeks. The build part is fun, but the "now what" part is where I kept getting stuck. How do I get my new site known to others?