Agents remind me of sculpting

Imagine taking a big rock and slowly removing pieces from it until it becomes something meaningful or useful.

It is one of the new ways of coding that agents make possible. Before AI, producing code was costly, so a workflow like this was almost impossible to use.

These days, I describe what I want to build and dictate a lot of requirements. Agents generate a ton of code and produce a plausibly working solution. Now I have a rough block of code to work with. I start removing everything unnecessary until I reach the essence of the solution.

This doesn’t mean I always work this way. Sometimes, I still build a feature from the bottom up, almost line by line, because I need to understand and control every step.

This is simply one more way to work. Sometimes, generating a few possible solutions, taking them apart, and combining ideas from them gives me a better feel for the problem. To extend the analogy, it is like experimenting with several rocks before understanding what the final sculpture should be.

It doesn’t always work. I might still decide to start from scratch. But even then, I have learned which directions I don’t need to explore.

Whether this approach works depends on the problem I am trying to solve. But I am happy to have this option in my toolbox now.