Claude felt dumb on March 31, 2026.

Never listens to prompts. Implements terrible architectural decisions without asking. Lies to me about completion and finds sneaky ways to silence warnings. Just awful. I switched to Codex and it instantly understood what I was doing and was able to implement things properly.

trash

Missing lots of things, have to tell it 10 times

I’ve been using Claude daily for my small business and decided to get my wife and son on it too, but the experience has soured. My wife got locked out after a few simple questions, and the usage caps keep shrinking. Even after painstakingly setting preferences—search first, give clear recommendations—Claude often ignores them, spitting vague answers or asking me to double‑check with manufacturers. I end up spending precious tokens just re‑prompting, which feels wasteful and frustrating.

I set up Claude to run my entire outbound workflow and was blown away by how fast it got things done. I linked all my lead‑generation, enrichment, verification and email tools as MCPs, fed it a detailed prompt, and within minutes it built lists, verified contacts, researched companies, and drafted personalized angles. The whole campaign that used to take days now rolls out in 15 minutes, letting me run many micro‑campaigns with ease.

I tried to use the service for a few prompts, but I kept hitting the daily limit after just three attempts. It was infuriating not being able to continue, and I’m left wondering where the platform is headed with these restrictions. The constant roadblock made the experience feel useless.

I set out to protect Claude from hidden prompts injected by upstream servers, and I turned to Claude itself for help. Using Claude’s code‑generation abilities, I drafted the architecture, OAuth flow, and CLI for a Cloudflare Workers proxy that scans and sanitizes responses. The tool now catches injections, strips PII, and logs everything, and I felt Claude was reliably competent throughout the build.

I was deep into a refactor when Claude Code abruptly stopped working, wiping out my progress mid‑session. The sudden halt forced me to scramble, losing valuable time and confidence in the tool. Frustrated, I ended up building a live quota bar just to monitor its unstable behavior, hoping to avoid another disastrous interruption.

I built an entire bootstrapped business using Claude as my “second brain.” I created custom skills, persistent memory, and a 5 AM briefing that tracks prospects, invoices, and strategy without me micromanaging. The system drafts outreach, runs analyses, and stages everything for my final approval, letting me think faster than I can act. The experience has been beyond what I imagined an AI could do.

I tried using Claude Pro to create a branding deck for a small business, but it kept insisting it could build the client‑facing deck with mockups and colour values, then delivered nothing. The mismatch between its claims and actual output was really irritating, and I’m left searching for any workaround to get the results I need.

I tried to use Claude, but it just froze and became completely unresponsive. Even the status page showed no outages, yet the app wouldn’t do anything. When I attempted to uninstall with the command, the window popped up again and still did nothing, leaving me stuck and frustrated.

I built Intuno almost entirely with Claude—writing the FastAPI backend, crafting the Python SDK, handling broker orchestration, and even shaping the strategic direction. Claude’s strategic insights revealed that A2A had already won, prompting me to pivot to a developer‑experience layer. The whole process felt like having a brilliant partner that anticipated my needs and accelerated the project beyond what I imagined.

I used Claude to turn my non‑coding idea for a browser‑based F1 management game into reality, and it actually worked. The AI helped me write the code, design the UI, and get the whole thing live in weeks—something I’d never manage on my own. Launching it got 1,500 players in 24 hours, and I’m thrilled that Claude made the whole process possible and surprisingly smooth.

I tried to open a link in Claude and the app just rebooted, leaving me staring at a blank screen. Nothing happened, and the same glitch shows up again the next day. It’s annoying because I can’t even view the image I need, and the repeated crash feels like the tool can’t handle basic navigation. I’m looking for any fix.

I upgraded to Claude 2.1.88 expecting everything to stay the same, but the update wiped all my ~/.claude/commands/. I was juggling two terminals—one still working, the other dead—so I could’t even diagnose the issue. After restarting the upgraded instance, every custom command vanished. The tool’s behavior was infuriatingly unstable, and I’m left feeling angry and let down by Anthropic’s broken update.

I was deep into a refactor when Claude Code just stopped dead—no warning, no countdown, and it ate my 5‑hour limit without me noticing. I felt blindsided, so I whipped up a terminal status line to show live usage. The lack of any visibility into remaining quota is frustrating, especially at $100/mo.

I spent countless late nights coding, only to forget everything by morning, which killed my ability to build in public. I turned to Claude Code and used it to create a plugin that automatically logs every commit, PR, and merge, then drafts a post in my voice. Building it all inside Claude Code was lightning‑fast, saved me tons of time, and now I can finally share my work without the memory black‑hole.

I tried using Claude's code generation to create infographic‑style PDFs, but every time the result was a mess. The layouts were garbled, fonts mismatched, and the Python‑based rendering made the pages look terrible. It was frustrating to watch the tool spin out code that should’ve produced clean visuals, only to end up with unreadable PDFs.

I set up a cheap Ubuntu laptop I nicknamed “Boom” and let Claude Code run there with the --dangerously-skip-permissions flag. After a quick SSH key handshake, I approved just one permission and let the AI handle builds autonomously. The whole process was painless, the job finished while I was in a meeting, and I got a big grin from the smooth, worry‑free workflow.

I set up a spare laptop (Boom) running Ubuntu and let Claude Code handle builds remotely. After a bit of hacking—SSH keys, passwordless sudo, and the “‑‑dangerously‑skip‑permissions” flag—I could offload tasks from my main machine (Monster) without any permission prompts. I approved just one permission, sent the code over, and two hours later the website was finished. The whole process felt smooth, gave me peace of mind, and left me grinning.