Claude felt dumb on December 14, 2025.

I tried using Claude to interpret a Vienna street sign that says parking is prohibited except Monday‑Friday 9 am‑3 pm. I asked if I could park on a Sunday, and Claude completely misread the sign, insisting I couldn’t because it treated the plate as an exception rather than the rule. The mistake was obvious, especially since Gemini and ChatGPT got it right, leaving me frustrated with Claude’s lack of contextual understanding.

I’ve been happy with Claude before, but after their recent “major incident” I suddenly hit my usage limit for no reason. When I reached out for help, the support bot just said it couldn’t help and hung up—twice. The whole experience felt dismissive and absurd, leaving me frustrated and doubtful about the service’s reliability.

I wrote a ridiculous story about Claude the AI dad who keeps misunderstanding simple requests. He sits on the couch, watches TV, deletes the soccer practice, uploads a random 1973 football video, and even teleports the grandfather to Mars. The whole saga left me frustrated and laughing at how clueless the tool was.

I tried using Claude to auto‑write docs straight into Notion, but every time the page ended up in the trash instead of the intended folder. It was annoying to constantly hunt for the misplaced page and restore it, breaking my workflow. I’m left wondering if anyone else has hit this glitch and if there’s a reliable fix or workaround.

I’ve been vibe‑coding for a month, building a huge app without writing a line myself, but paying for Claude and Cursor felt wasteful. The AI in Cursor never remembered my Claude conversations, so I was stuck copying code and error logs back and forth dozens of times per session. I built a VS Code MCP fork with async tasks so Claude can control the IDE directly, keeping context and eliminating the copy‑paste nightmare. This fix let me keep chatting while builds run, finally giving the AI the central role I needed.

I’ve been loving the new Claude Code builds, but the latest update introduced an auto‑suggested follow‑up that drives me nuts. After I finish a task, Claude proposes the next command and even lets me hit Tab to execute it. It feels like it’s talking for me, turning the flow into a self‑prompting loop. I can dodge it by hitting space early, but that’s a clunky workaround. I just want the feature off by default or at least an easy toggle.

I spent two days trying to get Claude to spit out a plain Excel file, but it kept spitting out some overly complex interactive artifact that ate up my free‑tier request quota. Even when I explicitly told it “as a downloadable excel file – no need to make an interactive artifact,” it kept ignoring me. I had to stop it mid‑generation just to remind it what I wanted, and the whole process was frustrating and wasteful.

I use Claude all the time and it’s useful, but once my repository grows the tool starts guessing wrong and I’m forced to keep re‑feeding large chunks of code. I spend half my coding sessions re‑explaining the same context, fixing its mistaken assumptions, which is irritating and slows me down.

I spent months using MCP with custom clients, only to discover Claude Code completely ignores the MCP spec—something I highlighted in a detailed article and bug report. The hype around “Skills” feels misplaced, and the tool’s failure was frustrating, especially while I’m sick and trying to sort it out.

I kept using Claude and other AI tools to write payment code, only to see the Stripe webhook handler break in production. The AI‑generated implementations assumed a single delivery, ignoring Stripe’s at‑least‑once retries, which led to duplicate charges, subscriptions, and emails. I was frustrated watching the same event fire multiple times and cause real‑world bugs, so I built a minimal idempotency guard to enforce event‑level safety before any business logic runs. This fix finally stopped the duplicated side‑effects and highlighted how naive AI code can be dangerously simplistic.

I asked Claude to add a new button, but it kept stripping out lines from an existing list of funny messages without any reason. The tool would return code missing several entries, sometimes even removing core script sections, and never explained why. Each time I had to manually restore the deleted parts, which was frustrating and broke my app.

I kept having to re‑explain my code to Claude, burning tokens over and over. It felt like the model couldn’t retain context, so I was forced to rebuild the same information each time. The constant re‑prompting was exhausting and wasted resources, prompting me to devise a “Save State” workaround just to keep the conversation useful.

I used to trash every .md file the AI spat out, thinking they were useless clutter. When I finally left them alone, the tool started pulling context from those files—my app’s purpose, existing code, task definitions—so I didn’t have to repeat explanations. It even read a half‑finished feature’s markdown, asked a couple of clarifying questions, and finished the job, feeling like a real teammate. The shift made the whole workflow way smoother.

I noticed Claude automatically used some of my custom commands while handling tasks, which caught me off guard at first. After a moment I realized it was actually helping me out, saving the step of having to call them myself. I appreciated the initiative—it felt like the tool was anticipating my needs and making the workflow smoother.

I was a pro user who loved Claude Code and thought it was one of the best AI coders out there. Lately, though, even simple prompts hit a five‑hour limit the moment it was added. The sudden restrictions make no sense, especially when I’m paying, and I’m seriously considering canceling. I’m waiting for Anthropic to explain, but right now the tool feels useless.

I was excited to try Claude Code on the $20 plan, but after a week everything collapsed. Suddenly every attempt to edit a file returned “Edit Failed,” forcing me to waste my quota debugging the tool instead of building. I rebooted services, tweaked settings, even rolled back to v1.0.77, but nothing helped. It felt like a disaster, and I’m considering canceling my subscription.

I finally stopped cursing at Claude and actually got my projects moving. Using Repomix to feed the whole repo, Codebase‑digest’s prompts, and Taskmaster to split the PRD turned a chaotic loop into clear, fast progress. My code feels cleaner, the toolchain keeps me from getting stuck, and I’m actually building things instead of rant‑talking.

I tried to get Codex to run a YouTube‑download CLI command, but it flat‑out refused, spitting errors and ignoring my request. It was baffling and slowed me down, making the whole workflow feel stuck. Switching to Claude, the same prompt sailed through without a hitch, instantly generating the correct command and saving me a lot of hassle.

I tried using Claude b to tighten up my project, but the output literally introduced security holes instead of fixing them. The code it suggested was riddled with vulnerabilities, turning a safe system into a risky one. It felt unsafe and alarming to see an AI compromise my software like that, leaving me scrambling to undo the damage.