I've had ChatGPT open in a browser tab more or less constantly for the last three weeks now, which is longer than I usually give any new toy before the novelty wears off. Normally by week two I'm back to doing things the old way. Not this time. So I figured it was worth writing up what it's actually been useful for, as opposed to the "look what I made it say" screenshots that are clogging up every timeline right now.
Quick context if you've been under a rock: this is OpenAI's chatbot thing, free to use (for now, no waitlist drama that I've hit), and it writes weirdly confident prose about almost anything you ask it. Everyone's using it to write wedding speeches or fake homework essays. I've mostly been using it for scripts.
Where it actually earns its keep
Regex. I will die on this hill. I have written regular expressions for probably fifteen years and I still have to sit and squint at them like they're a foreign language, every single time. Last Tuesday I needed something to pull timestamps and status codes out of a pile of nginx logs, roughly 40,000 lines of them, and instead of doing my usual thing of copy-pasting from an old script and fiddling for twenty minutes, I just typed out what I wanted in plain English. Got a working pattern back in about ten seconds, tested it against three sample lines, and it worked first try. That alone has probably saved me a couple hours this month.
Same goes for one-off bash and Python scripts — rename these 200 files based on a date embedded in the filename, dedupe this CSV, that kind of grunt work. It's genuinely good at the boring plumbing tasks that used to eat up a chunk of my evening.
Where it lies to you with total confidence
Here's my actual complaint, and it's a real one, not just a "well nothing's perfect" hedge. It will absolutely invent things. I asked it about a flag for the find command the other day and it gave me -newestmod, described exactly what it would do, formatted nicely with an example. Doesn't exist. Never has. I only caught it because I happened to check the man page out of habit, and I'm now paranoid about literally everything it tells me involving an actual API or command flag. If you don't already know enough to catch it when it's wrong, this thing will send you down a wrong path with total, unwavering confidence, and it will sound exactly as sure of itself when it's right as when it's completely making things up. That's the part that worries me more than anything else about it, honestly — not that it's wrong sometimes, but that there's no tell.
The mood right now
It's a strange week to be writing about a shiny AI toy, timing-wise. Half of tech Twitter is doom-scrolling through layoff announcements, feels like every big company is trimming staff this month, and the other half is breathlessly demoing chatbots. Those two things are happening at the same time and nobody's really talking about how weird that juxtaposition is. I don't have a grand theory about it. I just noticed it while doomscrolling at 11pm instead of sleeping like a normal person.
My actual verdict
If you're doing any kind of scripting or text wrangling day to day, it's worth having open in a tab. Use it the way you'd use a very fast, very confident junior dev who has read everything but verified none of it: great for a first draft, terrible as a final answer. I wouldn't trust it near anything involving money, dates, or version-specific library behavior without double-checking every single claim it makes.
For everything else, the poems, the "write this in the style of Shakespeare" stuff, the pretending-to-be-a-Linux-terminal party tricks, that's fun for about ten minutes and then it's just a novelty. The regex thing, though. That's sticking around on my desktop for a while yet.
One last small thing: I tried getting it to write this blog post for me, just to see. It was fine. It was also completely lifeless and used the word "seamless" twice, so, no thanks. Back to typing it myself.