I wanted to create a simple text animation, with words zooming from the center of the screen. Overall, it took 14 tries, and it felt like I was talking to a slightly dimwitted genie, asking it to fulfill my wishes!
You can see the final result here.
Each step involved me cutting and pasting the files to a local server to see what happened, then I would ask ChatGPT to fix what was missing or not working.
I had to be careful when interpreting the results, because each new iteration didn’t necessarily retain what was working the previous time (for example, when I asked for the text to be in color and all-caps, it “forgot” some of the previous required code and CSS syntax).
But that doesn’t take away from the remarkable result: I was eventually able to coax out usable, fully documented and commented code (so I can extend it later if I want) without writing a single line of it myself — and creating it by hand from scratch would have taken many, many, many hours of frustration.
One of the big themes over the last year has been the rise of “business technologists” who are empowered to do more innovation themselves, thanks to the rise of new no-code and low-code tools.
Ultimately, this will is clearly going to be a huge step forward in the continuum of no-code/low-code/business-user enablement—people like me who know “just enough code to be dangerous” are suddenly empowered to do a lot more, a lot quicker.
And it turns out that ChatGPT is already pretty good at generating business code such as SQL and SAP ABAP—and presumably large-language models trained exclusively on this type of content would be even better.
For now, there are obviously limitations: you clearly have to be able to at least understand the code generated in order to spot the potential problems. And presumably there’s a danger of potential bugs that people like me aren’t knowledgeable enough to spot.
But it really does feel like “magic” is becoming part of business!