(AI) [C]ompanies reporting their own (security) self-reported metrics:
AI companies using their own AI to report it’s own metrics:
2 Likes
There’s definitely a massive difference in well reviewed AI generated code and non reviewed code. A program manager with no dev experience made a single massive monolithic app that violates just about every architecture and security best practice you could name and handed me the code too look at because it wouldn’t deploy. I’m glad he couldn’t get it to deploy because the automated security scanners we have would have set off pagers and woke oncall up to fix it.
I’m actually using AI to fix it, splitting it into microservices, fixing the massive security holes, etc. the existing knowledge is important because the AI will still at times try to apply fixes that are very dumb or insecure.
3 Likes
Does this save you time? My experience is that while the path to a solution may be different; the amount of time spent is roughly the same between AI and just doing it myself.
Edit: I am also starting to wonder if my experience with AI as a front-end UI developer may be much different than someone writing Dev-Ops scripts or database queries. The results I am looking for are very specific and often complex UI behaviors. When you write a script to deploy assets to a pipeline; results re what matters; not so much the ‘how’. For me, the how is very important to reach the desired visual appearance. Without it being done ‘just right’; it is just wrong.
3 Likes
TL;DR AI is a great tool, but some people (I’ve been clear, by now, who) expect it to be a self sustaining automated programming manufacturing line
Like I said, AI as a tool 
but managers/higher ups are acting like every problem is a nail that only AI can solve. I was initially against AI (which, retrospectively, was for the wrong reason; I very much still am against AI, but now, having learned a lot since and actually trying to use it, not because of AI itself, but how companies are attempting to leverage it)
@AudreySickburn I agree, AI generated code, as it stands today, still needs to be reviewed. Additionally, again, I can only speak personally, but my managers know this, is their voice actually worth a spit, and will they actually (hopefully vehemently) oppose sole AI use by devs? I suppose for now, only time will tell. However, since this thread expanded beyond it’s original scope, my eyes have been opened.
2 Likes
Honestly yes, using Claude 4.6 + specs I’m working on two completely different projects at the same time today. In one I’m completely rearchitecting this slop and in the other I’m working on helping a colleague with another project. Hopping back and forth between the two sessions to give instructions and guide, like I’ve got two interns. I burned through something like 2000 tokens in 2.5 days so my apologies to the environment but my productivity is definitely very much up this week in terms of work done per hour.
3 Likes
I should add that I’m not the best dev by a long shot. My time is so fractured. I work on so many things that aren’t normal dev work, before the demand to move to AI my time was maybe 20% dev work. I have a weird job. I do however have a working knowledge of pipelines, and the basics and a degree in information security so I’m able to work faster and identify and fix slop where I would be a lot slower without it. Could you do the dev work I did this week with ai faster if you had 100% of your time dedicated to it? Probably? Honestly guaranteed.
2 Likes
I have found an unintended side effect of AI. Working significantly longer hours. It’s a lot easier to let’s say sit on the couch until 8 pm instructing two AI agents to fix x y or z while watching the Winter Olympics. So while I’m normally able to check out at 5 and just stop working I’ve been casually working much later than usual.
4 Likes
OT tho?
1 Like
Salaried. OT doesn’t exist for me.
2 Likes
Same OT and salary are not really a thing in our field
2 Likes
I’ve been doing this all week 
. It is nice to get all my meetings done for the day at 5-6pm and ask Claude Code and Cowork to handle the stack of tasks while I game, make dinner, or talk with friends.
But I’m 100% putting in more time than I should now because I will just keep poking at the agents on and off all night, vs just getting some of it done and leaving things for the next day.
2 Likes
…no 9000 computer has ever made a mistake or distorted information…we are all by any practical definition of the words, foolproof, and incapable of error.
2 Likes
The senior dev on a project gave some notes on some required changes for a code review and I tossed them to Claude and they said I’ll make all of the changes except this one I don’t think it’s needed I’ll comment back on your behalf and tell her no. 
I was ready to get out the spray bottle. Bad ai no, bad.
6 Likes
I would like to go back in time to when I was significantly less productive but didn’t have to spend all day arguing with a glorified phone tree.
Me: okay Claude we have to use these specific references and create this in this exact way per this documentation. I have given you steering documents, clear instructions, and the exact code to use.
Claude: I decided you didn’t need that complexity so I made a custom auth script instead. It’s much simpler.
Me: the company literally requires what I told you to use. Also your script fails to build and would be insecure if it wasn’t broken.
Claude: okay fine but I don’t think you need it mine is much simpler.
Me: if you don’t follow these instructions I will go to work for minimum wage at a megacorp.
2 Likes
Now AI is telling us what the best solution is? Someone needs to tell Claude who is running this show. 
2 Likes
That’s all I want is to not be back talked and argued with about software requirements by a toaster. The dream.
2 Likes
i hate that the overall corporate response to this is generally, well there’s nothing wrong with the AI, its the future, you’re just not telling it the right things and not asking the right questions.
3 Likes
For real. I’ve spent a lot of time researching, practicing and working on prompts (spec method) and I will spec out in careful detail what I want and need. And it usually works, like I recently got several code reviews approved by an SDEIII where I didn’t write a single line myself and the changes are in prod now. But sometimes the AI goes so rogue in such a horrible blatant way. And then it’s like you are right you did give me clear instructions and I ignored them, sorry. I guess it really is like an intern.
2 Likes
My current reality is due to the recent insanity that is the DoW all the stuff we built internally with anthropic has been haulted and is being ripped out and replaced as fast as possible with OpenAI… being a federal contractor there’s no desire to even risk backlash from the current administration so it’s just rip and rebuild asap don’t ask too many questions. Chaos is the name of the game this week…
2 Likes