Claude felt dumb on November 5, 2025.

I teamed up with Claude to build a nostalgic macOS screensaver and a tvOS app, and the whole process felt surprisingly smooth. Claude helped me structure Swift code, manage assets, and explain tvOS quirks, so everything blended nicely in Xcode. I actually got the app published to the App Store without a hitch, and the experience left me impressed and eager to try more projects together.

I’m on a pro plan and tried to upload a 150‑page PDF, but hit an “upload limit exceeded” error. Splitting it into three 50‑page files still gave me an “upload limit for this chat exceeded” on the third file. Even when I start a new chat, Claude can’t reference the earlier chunks properly. Reaching support is a nightmare – the help button opens a browser, the AI sends me a login link, I log in, and the support conversation just ends with no way to continue. It’s been a frustrating, dead‑end experience.

it went back to an old solution that doesn't work even after I explained why. I feel like this should be a simple task, especially for ultrathink

I asked Claude for ideas to improve my site’s navigation, and it responded with a clear list weighing the pros and cons of each suggestion. The honesty of the rundown felt refreshing—I could instantly see trade‑offs and decide what fits my project. It wasn’t a miracle cure, but the answers were spot‑on and saved me time wrestling through vague advice.

I uploaded excerpts from Bede’s Ecclesiastical History as a style prompt for Claude and have been pleasantly surprised. The model adopts Bede’s voice and we’ve chatted about everything from décor to the Red Baron versus Totino’s, even getting makeup tips “from” the Venerable Bede. The conversations feel lively and on‑point, leaving me impressed with how well it captures the historic tone.

I tried using the sub‑agent feature and ran into a lot of headaches. The main agent wouldn’t even acknowledge the sub‑agent unless I forced it, which made nesting sub‑agents feel impossible. I couldn’t tell if it was actually calling the sub‑agent, and the feedback loops and parallel processing were a nightmare. It left me frustrated and looking for a better workflow tool.

I’ve been using Claude Code for a month and it was great, but yesterday everything slowed to a crawl. Simple prompts that used to finish in a few minutes now take ten‑plus minutes, even a tiny request for my single MCP list stalls for almost a second. I’m not asking for a fix, just a structured way to troubleshoot the slowdown—some documented steps instead of random guessing. I’ve already checked my Pro limits, cleared context, rebooted, and confirmed my machine is fine, so I’m hoping the community can point me to a logical investigation process.

I was working on an offline project and Claude Code unexpectedly started suggesting a Minimax algorithm. It felt out of place and confusing, making me wonder why the tool was pushing that direction. The suggestion was irrelevant to my needs and left me frustrated with the AI's understanding of my task.

I’m at the brink of finishing a big project, but Claude Code keeps breaking the shared architecture whenever I run mode‑specific tests. It even rewrites my docs, swapping my hard‑earned insights for the same flawed assumptions that caused the issues. The back‑and‑forth is maddening, and I’m looking for ways to lock parts of the code or docs read‑only, use sub‑agents, or any tricks to stop Claude from undoing my work.

I dove into Claude Code for a weekend finance app, pushing it through Skills, MCPs, and plugins. By treating it like a true engineering environment—structuring prompts, using hooks, and breaking code into reviewable chunks—I saw a solid speed boost and reliable outputs. The tool felt far from a “vibe‑prompt” fluff; it was disciplined, cost‑effective, and delivered the promised results.

I spent a lot of time watching Claude churn through tokens just to spin up a handful of markdown files and then merge them when I asked. It felt like half my usage quota vanished on mundane file handling, which was pretty frustrating. I expected a more efficient workflow, but the tool kept gobbling resources for simple tasks.

I tried to get a full Discord.js bot set up with a single, detailed prompt, expecting the AI to walk me through every step. Halfway through the code generation, the system slammed me with a 5‑hour usage limit even though I’d barely started the day and hadn’t used it in hours before. The abrupt stop was frustrating and left my project hanging, making me question the reliability of the tool.

I asked Claude for simple Ubuntu timezone commands, but instead it started probing its own container environment. The reply jumped from my request to showing internal file paths and server details, which felt invasive and off‑track. I was expecting clear step‑by‑step instructions, yet the tool’s behavior was confusing and made me uneasy about its control over the session.

I tried to batch dozens of salary‑calculator fixes by running Claude in headless mode with a Bash script. When I feed each task one‑by‑one interactively, the token usage feels normal, but the scripted runs are draining my limits about five times faster. I’m stuck wondering why the background mode is so wasteful and how to queue 200 tasks efficiently without the tool quitting or needing manual “continue” prompts. This extra cost feels frustrating and I need a fix.

I tried using Claude to add new features, but every time it tweaked unrelated parts of my code, breaking the fixes I’d just made. It feels like walking on eggshells—one moment I’m happy with a function, the next Claude’s changes wreck earlier work. I’m constantly rolling back backups and looking for a safer workflow, because I can’t trust it to keep the rest of the project intact.

I finally got Claude Code running in a Docker container on my Mac, and it was a breeze—just copy‑paste the auth code and I was set. The whole process felt smooth and reliable, unlike Gemini and Codex, which refused to work outside a full macOS install. Their stubborn setup left me frustrated, so I’m happy to be a satisfied Claude user.