Edge Cases

I recently saw a VC touting his new AI gateway on LinkedIn. His hope was to establish himself as someone who wiped out this new category with a vibe-coded project.

I usually ignore these kinds of LinkedIn posts, but this time I decided to review it. It was full of slop-mines and surprises: the AI gateway had no rate limiting, the token authentication could be bypassed via brute force, and so on. It reminded me of when I was getting started in the industry, when I would consider something done if it worked. Over the years, I realized that’s not how building software works.

With the advent of agents, this is on steroids now. You can literally take a landing page, prompt an agent to build the product behind it, and it will. What you don’t see are the battle scars the software went through, the experimentation, and how it arrived at its architecture and why. Without those answers, concluding that the product is competitive in the market is like owning the liability without understanding the risk that comes with it. It is easy to build 80% of a product now. The problem everyone is going to run into is edge cases.