I found myself writing this in my one of my CS Classes, today, and thought I would share it here:
AI Coding is an interesting problem. I am old enough to remember the 1990’s when we tried to write everything down perfectly and outsource the building to other companies and countries. That went INCREDIBLY wrong; it turns out that English is a vague and ambiguous language, and that good software delivery is as much as discovery process as it is a requirements process. So the person doing the work discovers what they are building as they build it. When you don’t have discovery, and you do not have back-and-forth, then you just have order-taking – and vague and ambiguous requirements will result in software that is not fit for use.
The primary mode of thinking of these AI tools for coding seems be very similar to what we tried to do in the 1990’s – just write “plain English” and out pops the software. That isn’t going to work; the results will be laughable. Now, to use AI as a sort of actual co-pilot, to have a discovery process, to go back and forth — that has a better chance of working. The result will be faster to write than a human could do on their own, a good solid first or second base hit to make programming faster. Will it be more than that? Time will tell. As for me, I see exponential increasing costs of data center and electricity resulting in incremental improvements in AI quality. So yeah. My money is on: No, it won’t be more than that, not for a long time.
