Claude has been showing up everywhere lately, and I wanted to get past the headlines and actually try it in a normal, real-world way. Not a polished review. Not a scripted demo. Just opening it up, installing the pieces, seeing what works, and figuring out where the hype is coming from.
I have used Claude before, mostly for coding, and I still think it is one of the strongest AI tools for that job. But for my daily work, especially writing and anything that benefits from long-term memory, ChatGPT is still usually where I end up. So the real question for me was simple: is Claude worth adding back into the workflow, and if so, where does it actually help?
Quick Answer
The big thing with Claude right now is not just the web app or the desktop app. The part people are really talking about is Claude Code in the terminal, especially planning mode. That is where Claude can slow down, ask better questions, map out the work, and then execute with more structure.
For everyday chat, writing, and browser-side work, I am not ready to replace ChatGPT or Atlas. But for larger coding projects, Claude is still very much worth checking out. My practical takeaway is this: use Claude when the coding task is big enough to benefit from planning. For smaller scripts or day-to-day writing, I would still use the tool that already has your context and memory.
Where I Started
I started in the Claude web app after upgrading the account so I could look around with fresh eyes. The settings were mostly what you would expect: profile details, appearance, notifications, artifacts, web search, memory, connectors, GitHub, and MCP server options.
I turned on the things that made sense for testing, including memory and web search. I also noticed artifacts, which are Claude’s way of packaging generated apps, code, tools, or other outputs so you can preview or download them. The terminology is a little different from tool to tool, but the idea is familiar if you have used other AI coding assistants.
- Claude web is clean and straightforward.
- Artifacts are where generated code or app-style outputs can live.
- Memory and web search can be enabled in settings.
- Connectors such as GitHub and MCP servers are available.
Web, Desktop, And CLI
Claude can be used in a few different ways: through the web app, mobile apps, desktop app, API, and Claude Code in the terminal. That last one is the important piece for coding.
I installed the Claude desktop app on the Mac and it looked clean. It adds a menu bar icon and has chat and code areas. During setup, it asked for permissions like file access and runtime dependencies so it can work with local files. That makes sense, but it is also something you should pay attention to because you are giving it the ability to read and write in places you approve.
The desktop app is fine for normal chat-style work, but it did not feel like the main reason people are excited about Claude right now. The bigger difference showed up once I moved into the CLI.
- The desktop app is clean and familiar.
- Local file access requires permission, and you should only grant it where you are comfortable.
- The desktop app does not have the same planning workflow that Claude Code has in the terminal.
Why Planning Mode Matters
The feature that stood out is Claude Code’s planning mode. In the terminal, you can switch modes with Shift-Tab and use planning before letting Claude edit or build anything. That is different from simply asking a chatbot to make a plan in a normal conversation.
The reason planning matters is that Claude can ask clarifying questions before it starts building. It can tease out what you are actually trying to make, what framework you want, how complicated the build should be, and what tradeoffs make sense. That is a big deal for coding because vague instructions usually turn into vague software.
In the test, I asked Claude to help come up with a small app idea. It suggested a few things, and we landed on a smart to-do list. In planning mode, it asked about implementation choices like plain HTML and JavaScript, React with Vite, or Next.js. For a quick live test, I kept it simple and went with the more direct option.
That is the part I think people miss if they only try Claude through the web app. Claude can plan in a normal chat, but Claude Code’s terminal workflow is more purpose-built for coding: plan first, ask questions, then execute.
- Planning mode is strongest in Claude Code CLI.
- It helps Claude ask questions before building.
- It is especially useful for larger projects where the first prompt should not be the final spec.
- For small one-off scripts, this may be more process than you need.
Installing Claude Code
On macOS, I installed Claude Code through the terminal. The first run opened a setup flow right inside the terminal, which was slicker than I expected. It let me choose a theme, authorize my Claude account, and trust the project folder I was working in.
One thing to understand: once you are inside Claude Code, you are not just using a normal shell prompt. I tried a basic command like listing files and had to remember that I was inside Claude’s interactive session. In practice, I would probably keep two terminal windows open: one for normal commands and one for Claude Code.
That setup is worth getting used to if you want the full coding workflow. The terminal is where Claude’s planning mode makes the most sense.
- Run Claude Code from a project folder.
- Only trust folders you actually want Claude to inspect or edit.
- Use a separate terminal tab or window for normal shell commands.
- Use planning mode before allowing edits on anything important.
The Demo App Surprise
Claude built a simple smart to-do list as a single HTML file. Visually, it looked nice for a quick test. The app could take natural language like a task with a due date and turn it into a more structured task card.
The catch was that the app needed an Anthropic API key. At first, that surprised me because I was already using Claude through a paid account. But the reason made sense once I asked about it: the standalone browser app was calling Claude’s API directly to parse the natural language task into structured data.
That is an important distinction. If you are building inside Claude artifacts, the experience may be included with the Claude subscription. But if Claude generates a standalone app that calls Anthropic’s API, that app needs API access and billing through console.anthropic.com.
So the app was not just a local to-do list. It used AI to interpret and rewrite the task input. That is why it needed the key.
- A generated standalone app may still need an API key.
- A Claude subscription is not the same thing as Anthropic API billing.
- If the app calls Claude from your browser or backend, you need API access.
- Always ask what the API key is being used for before dropping it into a project.
Chrome Extension And Atlas
I also tested the Claude Chrome extension because I was curious whether it would work inside Atlas. Atlas is Chromium-based, and I use it because it ties into ChatGPT and keeps my memory and context in one place.
The extension installed in Atlas, but it did not actually work in the way I needed. Clicking it did not really do anything useful. When I installed it in regular Chrome, it worked properly after logging in and authorizing it.
I then tried a workflow similar to what I use for generating social posts from a YouTube page. Claude could help rewrite the prompt and generate platform-specific copy, but it ran into the same kind of limitation I have seen elsewhere: the agentic posting side is increasingly restricted. It would prepare the content, but it would not just take over and post everywhere.
That is one of my frustrations with AI browsers and extensions right now. Some workflows work for a while, then get limited or changed. Comet has been stronger for local agentic browser actions, while Atlas is still better for me because of ChatGPT memory and the way it fits my daily flow.
- The Claude Chrome extension worked in Chrome, but not properly in Atlas during this test.
- Atlas is still useful for ChatGPT memory and browser-aware workflows.
- Comet has been better for agentic browser actions in my experience.
- Browser automation features keep changing, so I would not build an entire workflow around one fragile behavior.
Where Claude Fits For Me
After testing all of this, my verdict is pretty simple: Claude still feels like one of the best AI options for coding. The lead is not just the model quality. It is the planning workflow in Claude Code.
For writing, memory, daily research, and my usual back-and-forth work, I am still leaning on ChatGPT. The memory matters because it helps the writing sound more like me and keeps more of my ongoing context available. Claude may catch up there, but right now that is not where I would put it first.
For coding, especially larger projects, I would use Claude again. I used to use Claude through Cursor, and that still makes sense in some cases. But after seeing the planning mode in the CLI, I think it is worth getting comfortable with the terminal version if you are building something more involved.
I would not keep every paid AI tool running all the time just in case. My approach is still to turn tools on when I need them and turn them off when I do not. If I have a larger coding project every couple of months, Claude is worth activating for that. For small scripts and quick fixes, I can usually handle those inside ChatGPT.
- Claude is strongest for coding work.
- Claude Code CLI is where the planning workflow really stands out.
- ChatGPT still fits better for my daily writing and memory-heavy work.
- I would use Claude when the project is large enough to justify the extra setup.
Key Takeaways
- Claude’s biggest current strength is coding, especially through Claude Code in the terminal.
- Planning mode is the feature that makes Claude feel different because it asks questions and structures the work before editing.
- The web and desktop apps are clean, but they do not show the full coding workflow people are excited about.
- Generated standalone apps may need an Anthropic API key if they call Claude’s API directly.
- The Claude Chrome extension worked in Chrome during the test, but not properly inside Atlas.
- For my day-to-day workflow, ChatGPT still wins for writing and memory, while Claude is the tool I would reach for on larger coding projects.
Watch the Video
The video above if you want to see the full live walkthrough: the Claude web app, Mac desktop app setup, Claude Code installation, planning mode test, Chrome extension experiment, and the generated app discussion all happen in real time.
