A week ago, I suggested the need for a new programming language tailored for AI output to tackle the human verification bottleneck. The discussion that followed was lively, and this subject remains an intriguing one.

We discussed the language the model uses to produce its output. However, there’s another important aspect: the language used for writing input specifications. This is even more critical because the quality of the output depends directly on the quality of the input (see my post from last weekend).

The more I work with AI, the more I realize that we, the humans, have become the bottleneck in the code production process. We participate in the loop at least at two critical points: defining the task and checking the results. These two points are now the main areas where we need to optimize how humans and AI talk to each other. In both cases, the languages we use aren’t really fit for the job. On the input side, we use human languages like English to prompt the model. On the output side, the model gives us code in languages originally designed for humans to write by hand. I believe this is exactly where we’ll see new solutions very soon.