Coding is about precision. Natural language is not. Coding assistance from AI platforms is brilliant – an incredible thing. End-to-end project development using AI seems quite another thing. While these platforms are capable of building something “complete,” the amount of English-language definition that goes into getting something out the other and that even approximates your goals starts to look just like a very verbose, English-language version of true coding – only far less precise. Is it worth it? At this moment I’d say, not yet…

Podcast: Play in new window | Download

It seems a little absurd to do a bonus episode when I’ve only done one episode, but I felt like it was important to follow up after the experience that I had, much richer than before, with the idea of developing an entire project using AI. In the first episode on AI coding, I covered the topics of code assistance and full execution of a project, using platforms like Claude, CoPilot and ChatGPT. Last night, my husband, who is a WordPress expert, was nonetheless having a really hard time finding a plug-in that he needed for his purposes.

I don’t have a tremendous amount of experience with PHP development and my WordPress is okay, but I decided this was a perfect time to sit down in front of Claude. I also used ChatGPT, tried the both, and asked it to execute the creation of a plug-in like that from end to end. I’ll start by saying it did it.

It built the thing end to end. Pretty cool. Here’s the problem:

There was so much that it did not do according to our needs and exactly according to what had to be in place for the thing to work properly. So I went back and I edited and altered what I had asked it to do. I had a conversation with the platform where I told it, no, not that way, this way.

Under certain circumstances, I actually had to remind it that I had previously asked it to do a feature or a function in a certain way. And he apologized to me and then corrected what it had not done the first time around as a manager of engineers for many years. That sounded vaguely familiar.

In any case, my point is this. It turns out that the amount of work that I spent editing an English language prompt that defined what it was that I wanted was so much time and so much effort and iteration that I do believe, especially if I had any form of familiarity with developing plug-ins for WordPress, I would have done a better job, and in fact, a faster job coding the thing myself, especially by the way, coding the thing using coding assistance. This is my way of returning to the first episode and pointing out the fact that we are not yet at a place where developing rich and full projects without engineers is going to make a whole lot of sense, because coding effectively is about precision.

It’s about getting exactly what you want to happen to happen. And programming languages are still more accurate ways, more precise ways of commanding the machine than natural English language is. So while there is a definite advantage in the coding assistance world, I am absolutely holding my reservations that we are not yet there when it comes to full project development and execution without software developers in the mix.

All of that may seem very obvious to many of you, but I just wanted to make myself heard on the subject before we move on to AI clones in the next episode.