I've done it on a lot of fintech production stuff over the last few years. I won't link them as it's a little misleading because they're not exclusively coded in these tools. Rather they're used as parts and helpful for grepping and running tests after the work is done.
They're something like electric vehicles - faster, cheaper, safer, with some new unknowns to be wary of. The bad stuff comes from expecting 2x the work done with half the people. If the employer gives you time to code properly, read, revise, it's all great. You'd be writing code faster, reading more code, understanding more code, having many eyes reviewing the code.
You will get to 10k LOC in a day, which means that you have to set architecture up right away for any new project. You'll need to do best practices, things like XP which people tend to skip over because they don't have time.
It is also very bad with things that it's not trained on, like custom SDKs and integrating GPT-5. This is much like trying to drive your EV through a river.
Thanks for the detailed answer, you hit the nail on the head in terms of giving me the information I was looking for
> The bad stuff comes from expecting 2x the work done with half the people. If the employer gives you time to code properly, read, revise, it's all great.
They made it clear they're expecting former than the latter. In one interview they stated they're building a lean team of elite devs who'll be able to work 12hrs shifts to deliver stuff asap using these Tools, which is why I wanted to see people's experiences of building projects solely with agentic AIs and how they manage to maintain them in the long run
They're something like electric vehicles - faster, cheaper, safer, with some new unknowns to be wary of. The bad stuff comes from expecting 2x the work done with half the people. If the employer gives you time to code properly, read, revise, it's all great. You'd be writing code faster, reading more code, understanding more code, having many eyes reviewing the code.
You will get to 10k LOC in a day, which means that you have to set architecture up right away for any new project. You'll need to do best practices, things like XP which people tend to skip over because they don't have time.
It is also very bad with things that it's not trained on, like custom SDKs and integrating GPT-5. This is much like trying to drive your EV through a river.