AI

 I have described myself as AI skeptical. That was because I knew and felt the same frustration most people have had

 with AI, and yet I also see the potential.

I have long thought that computers and humans were... complementary. They're better at certain things than we are, we're better at other things, and I frankly look forward to dumping all the tedious, mind-numbing and boring work on artificial intelligence. 

At the same time, the term 'meatspace' has become common for good reason.

If you're using AI to read e-mails and articles, and then using AI to write a response, and all you're really doing is sitting at the keyboard as a meatsuit for the AI... what are you even doing here?

And don't get me started how tech companies have pushed AI on us. Foisted it. It's one thing to have a nice little icon on the menu bar letting you know it's an option. It's another thing to rearrange your entire website so that the only think you see is an AI prompt window. 

But that's all casual observation, as a customer or user and not as a coder or programmer. 

Ever since my company purchased a subscription to one of the major AI coding platforms, and I realized that (as we lost a team member, and there's only one other member of my team stateside at the moment) I needed someone or something to answer questions and help me resolve issues while I learn my job, I have been using it at work.

And...

It's been a bit like the Gartner Hype cycle, all happening in about a month.

It's a powerful tool. A really powerful tool. But it will also go off on wild goose chases and tangents unless I stop it.

I've also worked hard at creating a 'coding harness', though I didn't even really understand that term when I started. I just knew that I wanted it to remember important parts about how our self-service application worked, and that I didn't want it to keep all that information in it's context window if it wasn't relevant. So I had it create a 'lessons learned' document, and reference documents, and also added in some hooks and workflows to try to make it consistently perform well and not hallucinate or write some crappy slop.

I... am not sure about that last part. I am still learning our application myself, and it's been great at helping explain what the code is doing and how it's doing it, but when you have multiple steps in a flow where one function is calling another function which is calling yet another, I am not entirely sure the AI understands the full flow properly. But that's what code reviews are for, right?

Anyways, in terms of troubleshooting, it's really, really good. Almost scary good. I had it create a skill so it knew what logs to check to see how a request was handled (this took some work too, because I gave it about four examples of different types of requests and had it look for the log entries related to it... and at first it was like "I didn't find any results, expanding the search window from the last 30 days to the last 60 days" and I had to stop it very fast and point out that the examples I gave were all from the last couple of days. Basically if it's not finding results with it's query, the issue is with the query and not with the date-range.)

Anyways, once I got it all configured it proved it's worth. It actually proved it's worth while I was configuring it, as a help request came in while I was training it and I decided to use that as an example. It very quickly noticed that the reason our company's customer support person wasn't able to get an account on their client's instance for troubleshooting purposes was because the group we were trying to create their account in had another group with the exact same name in it, and the duplicate names meant it was erroring out.

The skill has been... a life-saver, tbh. It's helped identify root causes and find workarounds more than I want to admit.

However - it still needs human judgment.  Human involvement. If it starts going down the wrong track it just keeps on going. 

It reminds me a bit of Mickey Mouse as the Sorceror's Apprentice in Fantasia, when he got the broom to start carrying the buckets of water. Except that things soon got out of hand, the brooms multiplied and kept carrying water and kept carrying water and kept carrying water...

Because here's the thing.

A human would eventually notice that the water was overflowing, that continuing to bring buckets of water didn't make sense, and they would probably stop before things got out of hand. They'd at least ask you if you really wanted them to keep doing it.

That broom, however, did not.

And an AI will not. Not unless on some level it's programmed to.

Anyways, as I see the discussions of AI going around and look at what tech companies are doing and consider it's usefulness and it's challenges, I think the heart of the problem is this:

Some people are using AI as a replacement for real people. 

And that, at least at our current level of technology, is going to fail.