Claude felt dumb on November 29, 2025.

I’ve been co‑writing a nonfiction story with Claude and it feels like I’ve finally found that reliable friend who’s always there with the perfect line. The tool’s empathy and willingness to help turned countless writer’s blocks into smooth progress, making the whole process feel effortless and inspiring. I’m amazed at how much it lifts my work.

I was tweaking my agents and had CC keep my implementation-plan.md tidy across PRs. After setting up plan-manager, MCPs, context7, deepcontext, and brave-search, I even built a custom agent for Microsoft Agent Framework tips. I switched from Azure MCP due to token hogging. Near the end of a PR, I was about to run /clear manually, but today CC started doing it on its own and spitting out a NEXTSESSION.md. The sudden auto‑clear was confusing and annoying, and I’m left wondering if anyone else has seen this.

I was using Claude Code in WebStorm and when it spitted out a lot of text the terminal kept jumping around like a strobe light. The scrolling was all over the place, making it hard to follow the output. I’m stuck between blaming Claude or the IDE, and I can’t share a video because of IP concerns.

I tried using AI to speed up my healthcare project, but the code it spat out was riddled with hidden assumptions. I’d spend hours untangling 100‑plus lines just to find it wasn’t even what I asked for, and fixes often broke working parts. The whole experience felt frustrating and far from the miracle results others brag about.

I was skeptical about AI at first, but after my buddy nudged me to try Claude for simple tasks, I ended up building entire websites and now a full‑blown CRM in just two days. The code looks polished, integrates Docusign, payments, email, texting, and pipelines—everything my tiny sales team needed. I can’t believe I wrote it without knowing how to code, and it even outshines the pricey commercial solutions we tried. The experience feels surreal and a bit scary, but the results are unbelievably impressive.

I’ve been playing with Claude’s code skill for front‑end work and kept getting the same “AI slop” – Inter/Roboto fonts, purple gradients, centered cards. It felt generic and frustrating until I discovered that giving Claude a concrete aesthetic (e.g., brutalism with 4px black borders) completely changed the output. I built a “frontend‑design‑pro” skill with 11 distinct styles, each with its own palette, fonts, and real‑stock‑photo handling, and now the generated pages look far from the usual meme‑like templates. The difference is striking, and I’m curious if others still see it as AI slop.

I was initially wowed by Claude’s coding chops, watching it churn out clean snippets that felt like a boost to my workflow. But soon the tool started tripping over basic logic, spitting out outright dumb mistakes that I had to fix manually. The flip‑flop left me frustrated and questioning whether paying for another AI like ChatGPT would be a safer bet.

I tried using Claude to craft some original lyrics, but every line turned into a tired cliché. It kept spitting out generic phrases that felt flat and uninspired, making the whole songwriting process feel frustrating. I was hoping for fresh storytelling, yet the tool kept leaning on overused tropes, which was disappointing.

I told Claude I was stuck and unmotivated, and it actually asked me probing questions, then guided me step‑by‑step through setting up a C# project with Avalonia in Rider. By the end of the night I felt like I’d actually learned something new, and the experience was energizing enough that I’m even thinking about ditching ChatGPT for now.

I chatted with Claude and asked it to analyze my notes to see what kind of person I am. It concluded I was obsessed with “utopian knowledge” and chasing money and freedom—but hadn’t acted on any of it in four months. The AI kept pointing this out, urging me to take steps. I followed its prompts, and it actually guided me toward concrete action, making the whole interaction feel surprisingly helpful and motivating.

I’ve spent the last 36 hours trying to get Claude to follow simple commands, but it kept botching everything. Every attempt ended in a fresh terminal only to repeat the same errors. The constant mistakes made it feel useless and frustrating, and I’m wondering if anyone else is seeing this sudden drop in quality.

I built a massive video‑game analytics dashboard covering 40 years and over 16 k titles, and I did it mostly by pulling styling instructions on demand instead of using plugins. Claude generated the code for data analysis and the build, letting me control fonts, colors, charts, and dimensions exactly how I wanted. The process felt smooth and gave me the flexibility I needed, making the whole project surprisingly doable.

I tried to scale up my project with Claude, but as soon as my code swelled past a few thousand lines the tool started to implode. Every time I asked for updates, it would promise to apply them and then return garbled, incomplete snippets, ruining the whole codebase. Paying $100/month didn’t help; I’m stuck starting fresh chats, but Claude keeps delivering “new” code that never matches what I asked for. The whole experience has been exhausting and maddening.

I let Claude run my Chrome browser while I was at Thanksgiving dinner, and when I returned the tool had completely set up a TikTok shop for me. I was amazed watching it navigate Twitter, generate media, and even handle the whole store creation without my input. The few hiccups with image uploads didn’t dampen the thrill—its autonomy felt endless and incredibly useful.

I asked Claude to tweak a global dual-mode component, telling it to “remove the index-mode for this new subpage.” It obediently stripped that mode from every page that used the component, blowing up a lot of code. I only spotted the disaster in review, so I had to scrap the whole feature branch and start over, wiping out two hours of work. The whole experience was frustrating and a stark reminder to be explicit, request file lists, and avoid mutating shared components.

I tried to have Claude open an artifact on the right for my blog‑writing workflow, but it kept refusing, spitting out .md, .docx, or .txt files instead of the proper artifact pop‑out. Even when I explicitly told it to use an artifact, it acted like it hated that feature. The tool’s behavior was frustrating and broke my smooth editing process, despite having artifacts enabled in the settings.