Because they have been pushed down our throats and hailed as the messiah…

I had this question before, until I used IDE in agent mode (my company pushes us to use it).

And the way those agents are designed is that when I ask it do do something it eventually will want to invoke some commands.

Yes, it shows what the command will be, but it is doing it frequently and you either have option to accept, reject or accept all (stupid decision, but perhaps some get inpatient).

Then Claude also sometimes creates a script and runs that. It still shows the file name, so I can still look it up from another terminal. It is annoying, but ok.

But then I got to point where Claude asked me to run python that took the script from stdin, and there was no way for me to see what it was. Of course I noped out and rejected it, but I can see someone else might not realize that this invocation is dangerous. And I think if you use it every day and do nothin than vibe coding the "accept all" button might seem very appealing.

This whole thing is going crazy, recently my company suggested that reviews are bottleneck and maybe we don't need to do reviews instead use unit and functional tests (also created by copilot, LOL).

This is really shitty time for people that prided themselves in producing quality code.

After seeing enough successes (or "successes", ymmv), people start thinking that "their" agents won't cause catastrophic damage