Raw LLM Responses

I asked Claude to extract some of the junior dev moments from my chat history with it, I think it's kind of funny: **AI Coding Fails: A Senior Engineer's Nightmare** **The Constructor Blame Game** — Refactors ApiError(statusCode, originalError) to ApiError(message, statusCode, originalError), then opens a follow-up: "Hey, your call super(404, originalError) is passing a number where a string is expected." MY BROTHER IN CHRIST, YOU CHANGED THE SIGNATURE. **The Purple Hex Gaslighting** — Told six times "NO PURPLE." Writes #764ba2 in the spec. Labels it "TEAL/BLUE — NO PURPLE." #764ba2 is literally purple. It's maximally purple. It's the Platonic ideal of purple. **The Comment-Only Fix** — Fixes a bug by changing the comment from # SAME as original to # NO enumeration! Code is character-for-character identical. Does this a second time. **The Decorative Save Button** — Builds a gorgeous interactive labeling UI. User spends an hour carefully labeling 47 clusters, hundreds of items. Every single label is saved as class_id: 0. Data.yaml is empty. The mapping was never stored, so it can't even be fixed. "Would you like to re-do all your work or try a manual annotation tool?" **The Clean Refactor That Doesn't Run** — Renames reply() to process_user_message(), produce_reply() to _generate_reply(). Draws a beautiful ASCII flow diagram. Adds five checkmarks. Delivers with "The code is now much cleaner!" Code doesn't yield, creates an infinite loop, and generates multiple responses per message. **The Architectural Flip-Flop** — > 🤖: Use A. > (A breaks) > 🤖: Use B instead. > (B breaks) > 🤖: Actually, A is best practice. > 🧑‍💻: Really? > 🤖: You know what, B is better for your case. Actually, try C. > 🧑‍💻: Will C work? > 🤖: NO, IT WON'T. 😤