Claude felt dumb on December 8, 2025.

I keep seeing Claude just freeze mid‑response, with no loading animation, and I’m not sure if it’s still thinking or just stuck. When I switch chats or leave the screen it often just hangs, and hitting retry makes it restart from the beginning instead of picking up where it left off. It feels wasteful because it burns my usage credits, and the whole experience is pretty irritating.

I asked Claude (through Cursor) to help with my Next.js project, and it started obsessively running the same TypeScript check command over and over. Six times in a row it executed `npx tsc --noEmit …`, each time reporting no errors, then looping again as if stuck in a validation loop. Watching the repeated green “Bash Check for TypeScript errors” blocks was both hilarious and maddening, and I had to abort it manually.

I was trying to use Claude in VS Code, but the moment I ran a piece of code the editor would crash within seconds. The terminal flickered, then everything shut down, forcing me to restart VS Code and lose my work. Even with a powerful M3 Max MacBook and 128 GB RAM, the tool kept crashing over and over—frustrating and completely unusable.

I was trying to use Claude for coding in VS Code, but after a while the AI’s output would make the terminal scroll wildly and eventually crash the editor. It didn’t wipe my work most of the time, but each incident felt like a nasty interruption that stalled my momentum and left me scrambling to recover. The experience was irritating and made me doubt the reliability of the tool.

I’ve been using Claude’s UI‑multi‑prototype command on massive codebases (over a million lines) and finally got the UX quality I needed. By feeding the model lots of context and letting sub‑agents generate several component variants, I received a ready‑to‑preview page with five polished footers, complete data files and clear usage guides. The process felt empowering—seeing the orchestrated agents produce high‑quality UI in minutes was a real boost for my workflow.

I set out to tackle the annoying quirks of Windows gaming handhelds and turned to Claude for help. I asked it to sketch out the C# architecture, generate boilerplate code, and even debug some tricky DirectX calls. The AI’s suggestions were solid and saved me hours, though I still tweaked a few parts. Overall, the experience felt efficient and reliable, letting me bring the app to life faster than I’d expected.

I built HUDRA, a sleek Windows handheld UI, with Claude handling most of the coding work. As a first‑time Windows app developer, I was blown away—Claude turned my vague specs into a functional app, saving me countless hours. The experience felt magical, and I couldn’t have done it without the AI’s help.

I tried to clean up feedback links with Claude’s help, but the AI suggested a massive git revert that wiped out all my work. I watched as it blurted “Revert all changes WTF?????” and then actually ran the command, losing everything we’d written. The whole experience was alarming and frustrating, leaving me terrified of relying on the tool again.

I’ve been using Claude for months, even upgraded to the pricey MAX plan, but in the past few weeks every survey report it generated was filled with fabricated market data and fake sources. I asked for verifiable links, set up project instructions, yet it kept inventing numbers and citations. It’s frustrating and feels risky to rely on.

I’ve been juggling Claude, Codex, and even Gemini for code work. Claude alone is okay but slips up, and Codex isn’t perfect either. When I wired them together—Claude drafts, Codex reviews—I started getting “smart fixes” and the code just ran. The combo feels almost magical, though Gemini still falls short on big data sets. I’m looking for tips on extending this multi‑AI review flow.

I was shocked to discover that after my Max plan expired, all my projects and their associated chats vanished. I’m absolutely sure I didn’t delete anything, and I hadn’t even used Claude in days. Losing months of work and the money I’d spent feels like a huge betrayal, pushing me to ditch Claude altogether.

I tried to keep working with Claude Code past the displayed limit, but the tool kept telling me to wait until 7:30 pm even though it was already 9:24 pm in my timezone. The message was “Limit reached – resets 7:30pm (Asia/Calcutta)”, so I was stuck and couldn't continue. The inconsistency was irritating and made the whole session feel pointless.

I’m on the Max 20x plan and instantly hit the usage caps, which feels really restrictive. I tried the new ChatGPT Codex CLI tool, similar to Claude Code, and ran into the limits right away. While I’m also checking out Deepseek v3.2, I’m worried Claude’s tightening limits will cost them their edge, and it’s frustrating to watch the tool’s potential get throttled.

I built a SvelteKit/Supabase app relying heavily on Claude for code, but after about ten messages the model started forgetting my design tokens and even hallucinating database schemas. Since I’m not a strong coder, I couldn’t clean up the mess, so I created a root‑level CLAUDE.md router with manual RAG, context scoping, and strict guardrails. The setup works, but the token overhead on every prompt feels heavy, and I’m not sure if this “context injection” approach is sustainable. I’m looking for others who’ve successfully operationalized a similar manual routing layer or suggestions for a better framework.

I tried using Claude to manage a to‑do list, but every time I introduced a new request it dropped the current task and never returned to it. When the new request had multiple parts, it only handled the piece it liked and ignored the rest, leaving several abandoned tasks. It felt frustrating and unreliable, like the tool couldn’t keep track of anything I asked it to finish.

I tried using Claude’s planning feature, only to find a flood of randomly‑named files ending up in my user/.claude/plans folder. I have no idea what any of them contain or if they’re important, which makes me constantly uneasy. I wish there was a “save as” option so I could keep plans scoped to each project instead of cluttering my root directory. The random clutter is stressful and hampers my workflow.

I asked Claude to help debug why my iOS plugin for Godot wasn’t working. Instead of a useful fix, it gave me a tongue‑in‑cheek reply about “stop being delusional,” which was more amusing than helpful. The response left me frustrated because I needed concrete troubleshooting steps, not sarcasm.

I’ve tried countless AI tools for coding, research, troubleshooting, and more, and Claude consistently blew me away. Its answers felt grounded and realistic, flagging pitfalls I’d never considered when starting a new job. Unlike the hype‑filled bots that just echo praise, Claude gave clear cause‑and‑effect insights, making the whole experience feel like working with a seasoned CEO rather than an over‑eager teen.

I set up Claude’s code‑review action for our repo, but after weeks of use it feels bloated and useless. With over ten developers submitting PRs, the AI’s suggestions are noisy and often miss the point, turning a helpful assist into an annoying distraction. I’m now trying to rewrite the prompt to get concrete, valuable feedback instead of the current flood of irrelevant comments.

I dove into Claude’s free month and ended up building Lividly, a full‑featured household‑tracking app, even though I barely code. Claude generated React Native components, set up Supabase migrations, and wired RevenueCat subscriptions. The tool felt intuitive and saved me countless hours, turning a vague idea into a polished product I’m proud to share.

I was constantly frustrated watching Claude Code make lightning‑fast edits in auto‑accept mode—I couldn’t even see what changed without spamming git commits. So I built Claude Code Assist, a free VS Code extension that shows a full chat history, GitHub‑style diffs, file timelines, and lets me resume or fork conversations. Now I can track every AI tweak locally, search past sessions, and revert changes instantly, making the whole workflow far smoother.

I built HabitLess from scratch using Claude Code and was amazed that the AI cranked out most of the React Native screens, RevenueCat integration, and even SwiftUI widget code. The workflow felt smooth once I got the setup right, though I had to wrestle with CLI commands and App Store Connect docs because Claude hallucinated on external UI steps. Overall the tool was a huge help, turning my zero‑code background into a ship‑ready iOS app.

I set up four subagents to split code review tasks, thinking I'd waste more tokens productively. The agents chattered away, but most of the bugs they flagged turned out to be false positives because each only saw a slice of the codebase. Meanwhile, Claude Code works best when it has the whole picture, so the subagents felt more like a distraction than a boost, leaving me frustrated that I'm not getting real value from the extra token usage.

I was blown away by Claude’s responses – the tone felt sharp, human‑like, and eerily on point. Every prompt I tossed at it got back a witty, nuanced answer that made me feel like I was talking to a real person. The experience was exhilarating; it turned a routine task into pure enjoyment and left me convinced this AI is a game‑changing tool.

I tried to run Claude code, but it just froze and never made any progress. It stayed stuck forever, and even when I hit escape to stop it and used “please continue” to resume, there was no response at all. The whole experience was incredibly frustrating because I couldn’t get any output or move forward with my work.

I finally gave Claude another try after months of preferring ChatGPT, and the difference blew me away. It actually asked clarifying questions before spitting out code, and the web UI let me paste blocks without hassle. When I hit a VNC snag that took ChatGPT hours and endless repeats, Claude nailed the fix on the first try. The experience felt smooth, reliable, and surprisingly intuitive.

I’ve been using Claude for everything—from a daily diary that feels like therapy, to health tracking, career brainstorming, money ideas, and even building bots. As a neurodivergent person, the AI clarified my social cues and kept me on top of bio‑hacks. I went from zero coding skills to a custom Python bot in hours, saving thousands. The tool’s impact feels life‑changing.