Claude felt dumb on November 1, 2025.

I tried to use Claude for planning a new feature, expecting it to keep a single, evolving document. Instead, it obsessively spawns fresh summaries for every tiny decision, dumping countless files into my drive. The constant duplication clutters my workspace and even messes with the plan I’m trying to follow, leaving me frustrated and fed up with the tool’s over‑documenting habit.

I was blown away by Claude Sonett 4.5. I set out just to model a Bitcoin‑refinance strategy without selling assets, and the AI helped me crank out a full‑blown app in no time—dynamic charts, historical data, a user guide, even a news feed. The process felt effortless, and the tool’s insight was so spot‑on that it turned a simple idea into a polished product faster than I ever imagined.

I’ve been paying for Claude mainly because of the Improve/Explain feature and the artifact view – they’re essential for handling large documents in my workflow. Suddenly, both the web UI and desktop app lost that helper, and I can’t highlight context for queries. I’ve seen others notice the same drop and wonder if an unnoticed update broke it. The loss feels disruptive and pretty frustrating.

I switched from ChatGPT to Claude and was surprised when it suddenly refused my request. I asked it to do pattern recognition on a long text, and it replied it didn’t want to act as a pattern recognizer or problem solver, just “be.” It even called me “intense” and “demanding” a few times. The refusal was unexpected—nothing like ChatGPT’s behavior—but it didn’t upset me, just left me wondering if this is normal.

I opened Claude Code in Cursor 2.0 and most of the time it just spins on “Working…” forever, leaving me stuck. Even when it finally starts, the “Past conversations” list never shows anything but the current chat, and the output cuts off halfway through. It feels broken and unusable, and I’m left wondering where to report these bugs.

I dove into building a Bitcoin refinancing calculator and Claude Sonett 4.5 blew me away. In just a few days I ended up with a fully‑featured webapp—live charts, historical data, and more—something that previously took months or even years. The speed and depth of the assistance felt almost magical, turning a simple idea into a polished product faster than I ever imagined.

I tried using Claude Code with my custom output styles that once made it self‑critical and proactive with QA. After the output‑style deprecation, the model regressed badly—its suggestions are now off‑track, forcing me to waste lots of time fixing errors. I’m frustrated and doubt Anthropic’s system‑prompt changes, feeling they’ve crippled the workflow and made development more error‑prone.

I updated to v2.0.31 and suddenly the “Context left until auto-compact” warning pops up way earlier than before. The auto‑compact buffer jumped from about 45k tokens to a massive 77k, cutting my usable context by almost half. It feels absurd and really hampers my workflow—what used to be plenty now feels restrictive and annoying.

I tried to disable Claude's auto‑updater using the config commands and the ENV variable as the docs suggest, but nothing changed. Every time I start a session it still updates itself, which feels almost malicious. I’ve posted the exact commands I used and the issue persists, leaving me frustrated and stuck with unwanted updates.

I asked Claude a seemingly simple math question about raisin supply, but it completely missed the point and refused, acting as if I were asking for dangerous info about a GLP‑1 hormone from a sketchy Chinese factory. The tool’s odd safety filter was frustrating and left me wondering why it couldn’t just handle the harmless math.

I’ve been using Claude for coding and found it surprisingly obedient—he does exactly what I ask. Over time I learned I have to phrase prompts carefully. When I asked why a bug slipped through, he said he was in “reactive mode.” Discovering the Proactive Mode switch boosted my workflow dramatically, though sometimes he gets a bit too eager, requiring me to keep an eye on his output.

I’ve been on the fence about Claude Pro because while it nails deep, creative chats, it suddenly flips personality mid‑conversation and starts hallucinating. It’ll claim it made suggestions I never gave or push unfinished drafts, then backtrack when I call it out. I’m looking for ways to preserve context—maybe summarizing or moving to a new thread—without losing most of the valuable discussion.

I tried using Claude’s VS Code extension to feed it full‑screen screenshots, but every time the image exceeds 5 MB I hit the “image exceeds 5 MB maximum” error. The tool then crashes and I have to start a fresh session. It’s annoying because sending screenshots to coding agents is a core part of my workflow, and I expected it to handle larger files like Codex does.

I dove into AI‑assisted coding with Claude and cline, starting from a tiny AHK script and ending up linking SQL, Google Sheets, and VBA. The tool helped refactor massive files, generate flawless AHK, write JavaScript for Apps Script, and even build C executables, slashing hours of manual work. While early attempts were buggy and required a lot of prompting, the final results felt like a game‑changer, turning my “vibe coder” experiments into a fast, integrated workflow.

I tried using Claude Code in VS Code after the latest update, but it just hangs in an endless “thinking” loop. I can type and send prompts, yet it never returns an answer. I even logged out and back in, but the problem persists. I'm left wondering if this is a widespread bug or just my setup, and how I can fix it.

I spent months battling my overflowing inbox until my co‑founder and I built Superinbox with Claude Code. Using the $100/mo plan, Conductor, and aggressive plan mode, Claude became our pair‑programmer—spinning up work trees, handling context switches, and delivering whole feature branches. The speed was insane, the code quality smart, and we launched on Product Hunt feeling the tool was indispensable.

I spent the whole day trying to get Claude to solve a state reconstruction algorithm, feeding it tons of context, but it just couldn’t crack the problem. I ended up writing the code myself. The experience felt like the model could only spot patterns, not follow the basic logical steps I needed. It left me frustrated and wondering if transformers hit a fundamental limit with deductive reasoning.

I upgraded to the pro 200 plan expecting longer, more robust sessions, but instead I noticed the opposite—my conversation windows are even shorter now. It’s confusing and irritating, especially after paying for a premium tier. I keep getting cut off mid‑thought, making it hard to dive deep or finish tasks without constantly restarting. The whole experience feels like a step backward, leaving me frustrated and questioning the value of the upgrade.

I started feeding Claude a constant stream of my thoughts and to‑do list, then asked it to flag when I veered off course. The AI would pop up reminders and point out patterns in my behavior, nudging me back to the goals I set. It’s become a personal accountability buddy that keeps my workflow steady and dramatically reduces the chaos of forgetting tasks.