I just spent a days reviewing and reworking “top priority critical” code that was entirely AI generated, on an AI written PR, and someone else’s AI approved it and said it was great and so clever.

At no point did the PR say what it did at a structural level, but it went on in great details the technical components and a bunch of numbers that all sounded good in isolation.

I had to shut it down because it didn’t consider any impacts of said code, and it was going to fuck the whole system up in incredibly expensive and non obvious ways.

Now I have a meeting with a VP about why this urgent feature isn’t getting shipped, because customers want it…

I’m so tired of this sloppifaction of my job.

Claude/Codex are fundamentally broken, they cannot communicate at all anymore. They’ve spent 10 months reinforcement training it to be “agentic”, which means bad at communicating and solely focused on achieving benchmark coding tasks.

If you paste Claude output in slack, an email, or a PR, I fucking hate you and will enter your dreams and murder you Freddy Krueger style.

Just post the slop into Claude and send them the response

If you ask a colleague to do their fucking job and they hit you with the ol’ executive function theft, you have every right to respond with “Seems about right. Let me know when you’ve had a chance to think about it/test it out.” Then give them a deadline (reiterate when you need their input, when the next status update is, set a meeting to discuss, or something) that makes it clear you regard the question as unanswered.

Nice read.

I also enjoyed reading one of the linked articles: predators

I just ignore it.

/compact