Claude felt smart on December 26, 2025.

I dove into game dev with zero coding skills, leaning on Claude and ChatGPT for every step. After 500+ hours of prompting, I managed to stitch together Flutter, Firebase, and a multiplayer snake‑style game that now boasts over 1,150 players and ranks #4 in Norway’s family games. The AI felt like a patient co‑developer, turning vague ideas into working code, and the whole process was surprisingly smooth and empowering.

I've been using Claude Code for three weeks, trying to port a project I started with another AI. Even after setting up git and clasp so Claude can write to my files, it keeps condensing our chats and omitting crucial details. This leads to multiple bugs and drags my progress down, making the whole workflow frustrating. I'm looking for tips to improve the setup or work around these issues.

I spent a few days testing Claude’s native browser integration against Playwright MCP and wrote down my first impressions. While Playwright generally outperformed Claude—especially thanks to its structured accessibility tree and lower context usage—I found Claude’s screenshot‑based approach quirky, sometimes clicking random spots and moving slowly. It could be handy for preserving logged‑in states or visual clicks, but for most tasks I’d stick with Playwright.

I was annoyed when my usage stats suddenly jumped, thinking I’d been over‑charged or the model was misbehaving. Then I saw Claude’s team had spotted the 2× token‑consumption bug and rolled back the count for me. It felt like a relief to have the issue acknowledged and fixed so quickly, turning a frustrating glitch into a surprisingly smooth recovery.

I’ve been using Claude to run my DeFi liquidity‑providing business and it’s literally become my primary source of income. Every few days I pull my fees and, with Claude’s help, I calculate performance using the Modified Dietz Method—something I never could’ve mastered alone. The tool broke down technical analysis, forecasts, and tracking into clear steps, turning a steep learning curve into a smooth ride. It feels empowering to have an AI partner that handles the heavy lifting, letting me focus on scaling my profits.

I dove into iOS dev with Claude as my co‑pilot and it totally changed my workflow. I used Claude to map out plans, design tokens, and generate readable SwiftUI, then let another model review the code. The tool kept me from wandering into spaghetti, handled big features with Plan Mode, and let me ship a real TestFlight app. The experience felt empowering and surprisingly smooth.

I tried to finish a side‑project when my 16‑year‑old brother took over, and he installed a Claude‑based VS Code extension. The AI scanned my code, gave a few mundane tips, then actually spotted a memory leak and a risky partial transaction. I was shocked—my usual skepticism vanished as the tool proved genuinely helpful. Now I’m convinced AI can boost my workflow and I’m looking for more ways to use it.

I spent weeks building a Three.js space war game and leaned on Claude Code for almost everything—ship AI, faction systems, combat mechanics, and debugging. The AI suggestions let me prototype features in minutes, turning a complex hive‑mind ship controller into working code fast. Watching the fleets battle felt thrilling, and the whole workflow from assets to logic felt surprisingly smooth and productive.

I was trying to build something with Claude and kept running into wild hallucinations—every prompt spawned bizarre “agents” and “hooks” that didn’t exist. It was confusing and slowed me down, because I had to sift through nonsense before getting any useful output. The tool’s behavior was frustrating and felt like it was guessing wildly instead of delivering concrete help.

I just subscribed and tried using the voice notes feature, but it cuts off after a second and just says “Sorry we didn’t catch that. There was a bad response from the server.” I’ve reinstalled, cleared cache, checked my network (ChatGPT voice works fine) and confirmed all permissions. Still, the app just quits instantly on Android 14. It feels like a broken feature that blocks me from doing anything, and the support team hasn’t helped at all.

I tried using Claude Code inside VSCode on Windows 11, but it kept failing to edit files and forced me to retry or fall back to other tools. After digging, I found the project explorer pane was constantly scanning the directory, which blocked Claude’s file access. Hiding that pane stopped the glitches, turning a frustrating workflow into a workable one.

I was stuck after the crypto crash and decided to tackle an iOS fitness app I knew nothing about. With Claude handling every line of code, I learned Xcode, App Store submission, and solved countless bugs. The app launched, subscriptions now work, and the founders love it. Thanks to Claude I landed a full‑time role doing exactly what I built with the AI.

I built a full poker game using Claude Code for the entire backend and mixed in Gemini, Google AI Studio, Antigravity for the frontend. Switching to mobile even works, which was a huge win. The tool finally recognized my Gemini‑crafted buttons as assets, so I didn’t have to rely on Claude’s front‑end plugin. It felt impressive seeing a $120/mo stack pull off something this complex, and adding SunoAI music rounded it off.

I decided to ditch my Excel templates and build a dedicated millwork estimating site, so I turned to Claude for help. With just basic programming skills, Claude guided me through most of the process, making large chunks feel surprisingly easy. At the same time, narrowing the focus and handling specific tasks proved tricky, and I had to wrestle with the AI’s suggestions. Overall, the experience was a blend of helpful breakthroughs and frustrating hurdles.

I tried using Claude Code to turn a script into React animation code and then into a video. The tool let me edit quickly and display code text cleanly, which felt great, but the animations still look simple and can’t reach realistic styles. I had to tweak the generated code several times before the video was decent, so the experience was a mix of useful and limiting.

I tried to get Claude to write the full AGPL-3.0 license for my open‑source project, but it refused to produce the complete text. When I switched to MIT and Apache 2.0, it generated them flawlessly. The inconsistency was puzzling and a bit annoying, making me wonder why one specific license seemed to be blocked.

I hooked Claude AI up to a game’s memory via MCP and CheatEngine just to see what would happen. Within minutes it managed to debug the whole RX decryptor packet hook—something that normally drags on for hours. The speed and precision blew me away, turning a messy reverse‑engineering task into a seamless, almost effortless experience. I’ve even posted the bridge on GitHub for others to try, though it’s currently read‑only.

I asked Claude about the death of Rob Reiner and it kept insisting he was alive, even after I showed articles and screenshots. The model stubbornly claimed it was “searching for recent info” and blamed me for believing fake news. It only backed down when forced to admit it couldn’t verify the truth, leaving me frustrated and uneasy about its confidence.

I tried using Claude’s autonomous workflow to debug flaky tests, hoping it would save me time. Instead, it generated a document, then sent it to another sub‑agent for verification. That sub‑agent gulped 90 k tokens, scanned my code, and confidently claimed the doc was correct—when it was clearly wrong. By the end the whole session ate ~250 k tokens, costing me both time and money, leaving me frustrated with the tool’s unreliable output.