Claude felt dumb on November 25, 2025.

I noticed Claude’s answers getting weaker as our chat grew, especially after using the built‑in /compact tool, which left me with a vague summary and lost context. I tackled the issue by creating a handoff plugin that snapshots the current progress, compacts, and then hands the focused summary to a fresh Claude instance. The result was a smoother continuation, letting Claude pick up right where I left off without the usual lossy drift.

I tried using Claude for front‑end work and was repeatedly let down. It kept misreading inherited styles, guessing wrong fixes, and ignoring my design framework, even creating custom buttons from scratch. It favored inline styles and struggled with CSS transitions. I ended up writing all the HTML/CSS myself, while the same model nailed complex backend data designs. The contrast was frustrating and wasted my time.

I tried using Claude after hitting my session limit and noticed a new “Plan Agent” popping up. Instead of a quick text response, it now writes a whole “calm‑strolling‑cocoa.md” file, then follows up with questions in plain text. The tool’s thinking time has ballooned, and the extra steps feel pointless. I’m confused and annoyed, wondering if a recent update caused this slowdown and how I can turn the feature off.

I spent hours adding dozens of commands to my `permissions.allow` list, only to have Claude’s Plan and Explore subagents ignore them. Every time a simple piped `find` runs, I get blocked with “Allow bash command?” prompts despite already approving it. The main agent works fine, but the subagents keep forcing me to re‑confirm, which is exhausting and breaks my workflow.

I tried to run Claude Code from the WebStorm terminal, but it instantly threw “Cannot launch Claude code” and later “No available IDEs detected.” Running it from PowerShell worked, yet after the first query the shortcut (Ctrl + Alt + K) stopped injecting context and just opened a new terminal. I’m stuck, wondering if WSL or my Windows/WebStorm setup broke the extension.

I tried using Claude to generate code, but it kept spamming extra newline characters everywhere. The output was cluttered and hard to read, forcing me to clean up manually after every snippet. That constant formatting glitch made the whole experience feel rough and frustrating, turning what should've been a smooth workflow into a tedious chore.

I spent 4.5 months wrestling with Claude, and it felt like a constant battle. The model would skip large parts of files, duplicate code, “improve” working functions only to break them, and was confidently wrong about a fifth of the time. I learned to copy‑paste only snippets, set strict boundaries, and test after every change just to keep the project alive. Despite the pain, I finally got a SOTA‑level memory system, but only after endless revert prompts and frantic debugging.

I noticed after the latest update that the model started using way more tokens than before, inflating costs and slowing me down. The responses also seemed to over‑plan, adding unnecessary steps and verbose explanations that made the interaction feel bloated. It wasn’t crashing, but the extra token drain and the overly elaborate output were frustrating and made the tool feel less efficient than it used to be.

I kept getting the same strange Claude Code error no matter what prompt I tried, even with simple requests. The message complained about “thinking” or “redacted_thinking” blocks that I never edited, saying they can’t be modified. It blocked me from using the tool entirely, so I’m stuck and wondering if it’s a bug or something I’m doing wrong.

I opened Claude Code after the latest update and was immediately confused by the usage panel. It showed my quota expiring next week, even though I know it actually ends in two days, and both the daily and weekly usage bars were stuck at 99% despite only using about 30% yesterday. The incorrect stats felt misleading and made me doubt the tool’s reliability, turning what should've been a simple check into a frustrating experience.

I’ve been trying to use Claude Code, but it keeps throwing the “Error editing file” message whenever it handles tabs and spaces. Every time I attempt a edit, the tool misinterprets the indentation and fails, which stalls my workflow and forces me to redo work manually. The repeated hiccups are irritating and make the experience far less smooth than I expected.

I dove into building a job‑search scraper with Claude Code, even though I’d never touched a scraper before. Claude handled the heavy lifting, letting me pull ~60k listings, keep them fresh, and push alerts to Discord. The experience felt smooth and empowering—Claude turned a daunting project into something I could actually launch and share for free.

I updated Claude’s code and it just spun forever. When I asked it to refactor something, it kept “gerblibbeting” and showing all the thinking animations. I tried interrupting, promoting again, reinstalling, updating Cursor, switching models, tweaking my agent config, even rebooting my PC—but nothing changed. The whole experience was maddeningly unproductive.

I was trying to use the Claude CLI inside VS Code and suddenly everything started timing out. The service checker said the API was up, my network was fine, and nothing on my machine had changed, yet I couldn’t get a response. I’m in the UK and it feels like the new release broke the connection, leaving me stuck and frustrated.

I updated the VS Code Claude extension and suddenly the Stop button and Esc key no longer halt Claude Code. I usually run it in Bypass permissions mode and rely on quick interruptions, so this break is really unsettling. I’ve checked settings and the CLI, but nothing seems to fix it—feeling stuck and annoyed.

I noticed Claude Code got way worse after the latest update. Before, I could explain a task, get a reasonable exploration, and receive a solid, concise plan to iterate on. Now it spawns massive “Explore agents” that gobble tokens, then a “Plan agent” that over‑engineers everything. I end up rejecting most plans, forcing it to do far less, yet I still burn about 55% of my Pro quota on just two tasks. This token‑hunger feels wasteful and frustrating.