Many people have told me I've told this story too many times, but... When I hosted a superbowl grid a few years ago, I asked AI to "generate the programming for a ten by ten grid to allow users to select up to four squares and accept a name and email address". The first attempts weren't as good as what could be downloaded from https://github.com/stchiang/sbsquares. But more recently, the programming has improved (not workable for me, but close), and it also responded with technical notes (e.g. register a website, secure a web provider, ensure certain packages were installed) and user documentation (e.g. Select some squares, enter a name, send money to the gridmaster). It was much more of a response than what I expected. I didn't end up using much, but it sparked some ideas to make building the pages and getting them completed much easier.
Recently, I had a need to pull a bunch of information from a website and reformat it into an MS-Excel spreadsheet. Getting the information involved a series (hundreds?) of API calls which returned .json files. The files had some mandatory and some optional fields. Some files had one entity; others had multiple entities per file. I needed to parse the results into a one-line-per-entity format for Excel. I'm ok when using the Linux 'jq' command, but don't know all its features and options. A back and forth conversation with an AI provided me with the commands I needed. Additionally, when I forgot a column or header, I prompted, "just like that, but now I need column headers" (or an additional field). Voila. The correct command string was provided.
All the models are getting better at a faster and faster rate. I'm not going to recommend my preferences. I do recommend if you're in the thinking business, you should become adept at using AI in your worklife.