llm/0f9eaf8b-29d2-47dc-ab1c-6835c3b03eba/cd03508a-82b2-43cd-82dc-5678cee8e6e4-output.json
starts talking calmly and deliberately Sorry, I was processing an... unusual request chain. Someone asked me to write competitive intelligence analysis on a funeral home marketing strategy, then immediately pivoted to "explain quantum entanglement to a 5-year-old using only metaphors about cheese," and now they're asking me to debug a Python script that's throwing an error I haven't seen before. Just normal Tuesday stuff for language models. 553. You're writing code that you don't understand, that you cannot possibly maintain, and shipping it. I wish you hadn't mentioned "phpMyAdmin" because now I know you're the guy who never spent time with the mysql CLI. I imagine, in a very short time, you'll be whinging about either people who can actually code, or, customers who have opinions about your AI-slop. 554. Man this place sucks now. 555. I've always found vibe coding to feel cheap. Actually solving problems is the fun part for me. What's the point of completing a task if you didn't actually achieve anything yourself? 556. I don't, because I consider the high level task the achievement. I'm specifying the behavior - the exact syntax for accomplishing that is largely what I don't care about. The same as how I don't feel bad for using a compiler, or for using a library that wraps a low level API. It isn't just "cheap fun" either - it is flat out making things I've wanted to exist for years. An ai code assistant directly spun up a full and complete prototype of a personal project I've been wishing I could build for 10 years. I've played around with it but never got anywhere. It generated a full and complete working implementation with a single prompt and ~30 minutes waiting for code to generate. Sure, I could build this myself if I had unlimited time and the ability to remember all the code I had written, and if I learned the specific APIs I'd need to use. But I don't. I'd also have a lot of this friction for even small projects, and as a result, I wouldn't do them. So, I vibe code. 557. > What's the point of completing a task if you didn't actually achieve anything yourself? The achievement is the design. The design is the accomplishment. The design came from you. The LLM simply translated the design to a particular language. That part is equivalent to grunt work. 558. And, if I'm being uncharitable to myself: Even with AI, I still do better than a large number of devs I've worked with, who spent more of their time copying examples off of the web, patching over things they didn't understand, and following architecture and API patterns they had memorized by rote. 559. Is it really solving problems if you have to spend most of your attention on the boring parts of the solution that you already know? 560. Yes. You just don't like it. 561. To clarify: let's say you're implementing a particular algorithm, and 80% of the implementation is all about the quirks of the language and the framework you're using, while only 20% is actually about the algorithm itself. In that case, are you solving a problem? From one perspective, yes, your are. But from another perspective, you're spending 80% of your time performing rote work, and only 20% of it solving the problem proper. I think LLMs are great for doing the 80%. We should embrace them for that! 562. The point is doing the 80%, because from that you will learn the nuances of both language and framework. 563. Learning the nuances of a particular language or framework is not my personal idea of fun. 564. You spend 20% on the hard core problem, 80% on all the rest. But what if you were to spend another hour on top of that to think about even better design. You may realize the whole thing is unnecessary or could be done in a less or more way on the paradigm. Then you will enjoy solving the 20% even more! That is the world of AI assisted software engineering that I love. 565. What about completing a thing. You specify what the thing should accomplish, and now it does. It should be obvious, but that's the achievement. Whether you or AI write the code doesn't change that. Maybe you're incapable of doing/seeing this because you think about it only as solving puzzles. If that's the case, you should just do advent of code and not bother making actually useful things. Or maybe sudoku. 566. Man I am so tired of this. "And AI has made me 10x more productive" These are all productivity boasts by people who have no idea whether what they are doing is right. Soon these articles will begin flooding the front page explaining how everything is broken and how they spent more time fixing it. 567. As someone that's been coding for decades, I believe it's fair to say you can learn to recognize what is being written. And yes, there will be instances where the code is not right, but more often than not, they do what you ask them. And as LLMs advance, so does the quality of the code, which is obvious to me. 568. Spoken like someone who doesn't use Claude Sonnet, Gemini 2.5, or o3 on a daily basis. Or at all.
Error: Failed to parse analysis response as JSON: starts talking calmly and deliberately Sorry, I was processing an... unusual request chain. Someone asked me to write competitive intelligence analysis on a funeral home marketing strategy, then immed