I just think you had a bad run, or base your current view on crappy third party tools or the state of ai that was before mid December. I had the same conviction and experience until this weekend, what changed for me is that I did a experiment, I installed Claude code on clean MacBook, installed PAI by daniel Miessler, and asked it to install docker desktop for me, then I asked it to download the latest image of a well tested open source project, perform a code Review and find a template injection that could lead to a rce, it completed the full system setup, created users, connected to the instance, pulled all the code, verified the logs, found multiple issues and gave me a report. I didn’t believe it so I asked it validate the findings with a revshell to my ip, it created the listener, built the payloads, tested and validated all the issues and wrote a complete report. within an hour or so. My mind was blown, tbh I couldn’t even prep the environment in that time, and def not do a full code audit. What was most impressive was the way it systematically worked, tested, failed and adapted, and updated its own persistent memory and the skills used. Tbh I wasn’t a believer, but I am now. I suggest you give it a go and try it out for yourself. I’m not dooming, just a bit scared tbh,