Using ChatGPT for anything more than repetitive or random generation tasks is a bad idea, and its usefulness becomes even more limited when you're working with proprietary code that you can't directly send to ChatGPT. Even with personal projects, I still try to avoid ChatGPT as much as possible for the simple reason that I'll be forced to pay for it if it becomes an essential part of my workflow when it leaves this free beta testing phase.

One thing I used ChatGPT for recently was generating test data.

Hey ChatGPT, I use SQL Server and here is my table structure, please generate an insert query with 10 rows of fake test data.

It wasn't perfect, but honestly nor is the test data I would have written. It was a great starting point and saved me a lot of time since this is a legacy app with some wide tables (30+ columns).

I've been using ChatGPT at work quite a bit now. Some of the things I've used it for are:

1. Writing a shell script that scrapes some information about code modules and shows them neatly
2. Minor automation scripts that setup and make my day to day docker workflow easy
3. Writing random regex, sql, lua pattern matching functions
4. It turned out to be surprisingly good at creating code examples for certain undocumented APIs (kong.cache, kong.worker_events, kong.cluster_events) in Kong API Gateway.
5. Copy pasting a rough python automation script, converting it into Go, and adding it in the application itself.

I still don't feel comfortable using it for anything big.

I haven't found a good use case for it yet personally, but I am excited to hopefully get it to start helping with the boring boilerplate and similar things.

I am a little afraid that it'll be the first bit of common tech I am too old to really grok and I'll just be that old dude in the corner that insists on doing it how everyone did "back in my day."

I, personally don't trust ChatGPT. So whenever I use it I always look at code it generated until I understand what it does. And the truth is that it takes me more time to understand it than if I would've wrote code myself.
It does wonders for repetitive tasks or data generation, though.
I believe in any place where security is critical, developers should not use AI-generated code irresponsibly.

I tried to use it, but it have some big issues in reliability, because at the end of the day, despise the dataset it's trained on, it's still something I describes as a "language interpolation."

It sometime make TERRIBLE recommendations for which tools/libraries I should explore, because it assumes that those libraries might have support. Those libraries never does and so I wasted weeks on it. (It doesn't help that both code and project are undocumented.)

So after that experience, I demote ChatGPT usefulness to just "cleaning up pre-written documentation so it sounds better." That's it.

I primarily use ChatGPT to ask about how to use X, Y, Z software libraries together, and other general programming questions. Unfortunately, the project that I work on is proprietary, so I'm not allowed to just feed ChatGPT my code and tell it to correct/refactor it. But that's fine, because ChatGPT still provides a lot of value to me for free.

I usually use it more to help me write documentation and add comments on some functions. It helps explaining what a function does.

To write code I usually just use it to write simple functions or a template code for me to start from somewhere. I avoid using it with external Libraries as in my experience, it likes to "invent" functions and methods that are not implemented.

This is a paywalled URL, but at least private windows work on Medium.