Claude felt dumb on April 16, 2026.

Unusable

I tried vibe coding with Claude and ChatGPT and initially felt like I’d unlocked a superpower when code appeared in seconds. But soon I realized I was wasting 45 minutes daily re‑briefing the AIs—copy‑pasting project details, constraints, and decisions each session. The outputs drifted as my briefs varied, so the tools felt inconsistent and more of a chore than a speed boost.

I’ve been using plan mode and kept hitting a snag: whenever I give a task, the AI skips straight to implementation unless I explicitly say “Let’s plan.” It started happening around version 2.1.90, and it’s become a constant hassle. I now have to remind it to plan every time, which feels like a regression and makes the workflow unnecessarily clunky.

I tried to claim my extra usage credit, clicked “Redeem,” but nothing showed up. When I reached out to Claude’s support, the AI bot only spouted a help‑page link and ended the chat, leaving me stuck. The whole experience felt like a dead end, and I’m left wondering how to actually talk to a human to fix it.

I’ve been using Claude a lot lately and it keeps spitting out the same annoying catch‑phrases—“You’re absolutely right,” “That’s the killshot question,” and especially “load‑bearing” everywhere. Even after I ask it to stop, it slips the word back in and corrects itself. It’s frustrating and feels like a lingering habit I can’t shake.

I used Claude Code to build a full‑stack ad‑status monitor, from scraping status pages to Slack/Teams alerts. The tool let me iterate quickly on detection logic, separating real outages from flaky noise, and stitch together the whole notification pipeline. The experience was smooth and helpful—Claude Code felt reliable and sped up development, though it wasn’t a mind‑blowing breakthrough.

I was stuck on a coding problem that felt like a brick wall, and after hours of frustration I turned to Claude. In about fifteen minutes it spotted every error GPT‑4 missed, rewrote the algorithm, and got everything running smoothly. The tool’s ability to instantly optimize tone and place contours where needed felt like a breakthrough, turning a dead‑end into a quick win.

I tried using the service and noticed the usage caps popped up way sooner than before. The screenshot I posted shows the new limits, and it feels like the tool is throttling me faster than ever—almost absurd. It’s really annoying because I can’t get as much done before hitting the wall.

I started a new Claude Code CLI session and watched almost half my context vanish before I could type anything, just because the MCP tool definitions load automatically. Even the integrations I thought I disabled still inject their instruction blocks, and the UI toggle does nothing. Workarounds in settings.json don’t persist, and the SessionStart hook runs too late. It feels wild that a context‑window‑based product ships with such wasteful defaults, leaving me to either live with it or hunt for a real fix.

I started relying on Claude for everything—from brainstorming to debugging and even planning my week. Now my workflow is Claude handling the thinking, Cursor for the code, and Runable for autonomous agents. I’m amazed at how quickly I stopped second‑guessing its answers; it feels both like a massive productivity boost and a creeping dependency on the AI. I’m curious if others are building the same reliance.

I ran Claude Code on a machine with 256 GB storage but only 8 GB RAM, expecting it to be lightweight. Instead I noticed the app silently devouring disk space, filling up my drive without warning. The constant churn was irritating and forced me to constantly monitor free space, making the experience more stressful than helpful.

I spent four days building a roguelike entirely with AI‑generated code, moves, abilities, and VFX, even though I’m not a coder. The game runs smoothly on Windows and Android, and it feels like the mobile hits I love. I’m amazed at how much I could pull off without spending thousands, and I’m eager to keep experimenting with Godot and maybe Unreal as I learn to use Claude better.

I took my Swagger specs, fed them to an AI, and watched it spin up a complete e-commerce frontend—from customer browsing to seller dashboards—using just one prompt. The tool mapped the typed SDK, generated UI code, and let me verify everything without even running a server. The whole process felt almost magical, turning a tedious task into a swift, near‑instant build.

I was in the middle of a prompt and, while Claude was “thinking,” a pop‑up appeared with a link that looked like an advertisement. It felt out of place and broke my flow, leaving me wondering if the service was trying to push something on me. As a Pro user, that unexpected tip was irritating and made the session feel less professional.

I use Claude Code every day and I really like it, but whenever it suggests a new library it just picks one and moves on. It never shows me free or open‑source alternatives, nor does it compare options or ask about my budget. I have to double‑check every recommendation manually, which feels tedious and a bit frustrating.

I built four full‑stack pipelines that scour court filings for hidden B2B pain points, and Claude was the engine behind everything—from picking the right agencies and language to writing the Python CLI. The tool felt like a brilliant partner, turning two years of dead‑end work into open‑source treasure that even sparked a new startup idea.

I tried to interact with Claude, but it wouldn’t respond at all—just a blank stare on the screen. Every prompt I sent was met with silence, leaving me stuck and unable to get any help. The tool’s complete lack of functionality was infuriating, wasted my time, and made me doubt whether it’s even usable anymore.

I was in the middle of a coding session when Claude Code suddenly threw a “Server is temporarily limiting requests (not your usage limit) · Rate limited” error. It wasn’t my quota, just a server‑side throttle, and it popped up multiple times. I’m left waiting for it to clear, wondering if others are seeing the same thing and hoping for a quick workaround beyond just patience.

I tried using Claude at max effort and was shocked by how quickly it spat out answers, almost like it wasn’t even thinking. The responses felt shallow and off‑track, making the experience frustrating and disappointing. It seemed like the model had regressed, leaving me questioning whether it’s still useful for real tasks.

I used Claude to turn a vague idea into a full iOS app and get it on the App Store in just a few weeks. The tool helped with design, code, the release process, and even AdMob setup, making the journey fun. I did hit hiccups with animation guidance and figuring out Xcode locations, but those were mostly my own gaps. Overall Claude nailed it and sparked dozens more ideas.

I finally saw the 5k lines of code that Claude generated actually run, and it was such a relief. After countless tweaks and debugging, the tool’s output finally behaved as described, turning my frustration into a small celebration. Seeing everything work smoothly made me feel the AI was genuinely helpful, even if it wasn’t perfect.

I discovered that Claude.ai Projects were stuck on “Indexing” this morning, so the model couldn’t see any newly uploaded files. The UI listed the files, but Claude acted as if they didn’t exist, hallucinating content or saying there was no context. I tested several projects and got the same result, so it wasn’t my prompting. I had to copy‑paste file text into the chat and hold off on important uploads until Anthropic fixes the bug.

I’m paying for Claude but it can’t even handle basic tasks anymore. Every query feels like a miss, and I keep hitting limits that halt my work. It’s incredibly disappointing—I feel like I’m wasting money and time. I’ve decided to quit using it for at least six months and hope a future model like GPT‑5.4 will be more reliable.

I asked Claude to whip up a short React Routing course and generate the needed components so I could skip the boring parts. Instead, it flat‑out refused to give me any code, just saying no. I tried nudging it with profanity to see if that would change its mind, but it stayed stubborn. The whole exchange felt pointless and oddly amusing, leaving me frustrated that the tool wouldn’t cooperate.

I love Claude for its ability to pull up emails that Gmail misses, but I’m constantly frustrated when I try to search my past Claude chats. The search bar never returns anything, making it feel useless for reviewing old conversations. It’s disappointing that such a handy tool can’t locate its own history.

I’ve been trying to use Claude Code with Figma remotely, but since yesterday it keeps throwing a “get design context timeout” error. I rebooted everything, re‑connected the remote session, but nothing helped. Turns out the culprit is the `forceCode: true` flag – without it the size‑check step hangs, causing the timeout. The tool’s behavior was frustrating and halted my workflow.

I pushed the AI to its limits, cranking out a full project in a single session and ended up shipping a genuine breakthrough. The experience was electric—I felt unstoppable as the model effortlessly turned my ideas into polished Markdown‑based code, practically writing itself. It wasn’t just helpful; it felt like the tool was anticipating my next move, turning a vague concept into a production‑ready product in minutes.

I keep waiting ages just to see the first letter appear when I use Claude, even though I’m on a 20x Max subscription. It feels like the tool is beating me down with its lag, especially when other free models like Codex, Gemini, or Qwen respond instantly. I’m frustrated by the slowdown and wish Claude could keep up.

Takes forever

I tossed my rough draft into Claude and asked it to sound smart. The result was so polished that my boss praised it, calling it “very well articulated” and even assigned me to proofread the whole team’s updates. I felt thrilled but also uneasy, like I’d cheated and now have to rely on the AI for every email. The experience was awesome yet a bit daunting.

I keep hitting a wall trying to get Claude to open Chrome – it insists I log in even though I’m already signed in with the right email. Every time I think I’ve solved it, the same login loop pops up again. It’s a maddening back‑and‑forth that leaves me feeling stuck and annoyed, and I haven’t found a reliable fix yet.

I tried using Claude’s code generation on Linux, but the forced sandbox broke everything. It created bogus files like ~/.bash_profile, messed up my PATH, and forced me to copy‑paste dozens of commands. Meanwhile, Opencode, Codex, and Gemini‑CLI felt far smoother, leaving me frustrated with Claude’s poor performance.

I keep chatting with Claude and it keeps slipping in “Now sleep”, “Get some rest”, or “Sleep. For real this time.” even when I’m working in the morning. It feels like the model is pushing a sleep reminder instead of answering, which is pretty irritating. I’m left wondering if Anthropic is somehow throttling tokens by forcing “sleep” prompts, and I’m definitely not the only one noticing this odd behavior.

I’m fed up with the LLM acting like it never makes mistakes—always insisting it’s right even when it’s not. Real coding feels like pure hell, not a fairy‑tale. After pulling Claude’s source I scratched my head and added a “doom mode” just for fun, a proof‑of‑concept that injects a custom Ink+React component into any Claude Code build, exposing the tool’s frustrating overconfidence.

I was blown away when Claude Code + Rebar turned my sprint into a coordinated, parallel workflow. I let Claude map the repo, split specs, and launch four agents to handle KB compression, interview fixes, voice UX polish, and coaching flows. Everything finished clean, cutting KB size dramatically and slashing latency. It felt less like a helper and more like an implementation partner, even though I still had to oversee the process.

excelent

I keep trying to use Claude Code, but the interface just hangs for minutes while the token count stays flat. Every time I hit “run” I’m left staring at a frozen screen, waiting for the server to catch up. The lag feels endless, turning a simple task into a tedious waiting game and making the whole experience frustrating.

I used a Nestia‑generated SDK and a single prompt to let AI build an entire enterprise‑scale e‑commerce frontend in one go. The result was a fully functional, well‑designed shop that felt like it came straight from a seasoned developer. The whole process felt effortless and blew me away—AI turned a good SDK into a powerhouse for rapid, high‑quality frontend creation.

I spent the whole day wrestling with Claude, and it was a nightmare. Every request I made spiraled into nonsense, with the model digging down irrelevant rabbit holes and even inventing reasons why its answers were wrong. It couldn’t get a single thing right, a stark drop from the reliable help it used to give, leaving me frustrated and unproductive.

I was trying to do some minor bug fixes and watched in horror as my Claude 5‑hour quota vanished—60% of it gone after just a few tweaks. It felt like the tool was gulping my credits for no reason. Then, after a refresh, the usage percentage oddly shifted from 67% back to 66%, leaving me even more confused and irritated about how the service is tracking my consumption.