You're using Claude, but maybe not the right way

Anthropic describes Claude Desktop like this:

"The Claude desktop app gives you three ways to work with Claude: Chat, Cowork, and Code, from quick questions to complex research to building software."

It sounds simple, but those three words stand for three completely different working mindsets. And chances are some of you are using the wrong mode without realizing it.

💬 Chat

Claude Chat

Essentially the same claude.ai you already know, but because it runs natively on your machine it gains a few genuinely convenient things:

• Quick entry: Double-tap the Option key (macOS) or Ctrl + Alt + Space (Windows) and Claude pops up right over whatever you're working on, no app switching, no new tab. Perfect when you need to look something up instantly.
• Screenshots & window sharing: Capture your screen and share it straight into the chat instead of laboriously writing a long description. Claude sees and analyzes exactly what you mean.
• Dictation: Speak instead of type. I think of it as an upgraded "Hey Siri" or "Hey Google", giving you information hands-free.
• Desktop connectors: Connect the tools running on your machine so Claude can work seamlessly with them.

🤝 Cowork

Claude Cowork

This mode is different in nature. You don't "ask Claude", you hand a task to Claude.

You give it a goal; Claude plans on its own (shown in the sidebar), pulls data from connected sources, runs, and returns a result. While it works you can see exactly what it's doing: which file it's reading, what it's synthesizing, how far along it is.

What Cowork can do:

• Folder access: Point Claude at a folder on your machine so it reads it, figures out what's relevant, and saves results back in the right place. Or upload files / paste content straight into the conversation.
• Scheduled tasks: Hand off recurring work on a schedule, a daily briefing summarizing Slack and calendar each morning, a weekly roundup of what shipped, inbox triage sorting email by priority. Claude can even go gather the latest news every morning, summarize it, and lay it out neatly for you. Set it once and Claude runs it. If your machine is off when the task is due, it runs it when you reopen.

Auto-collect and summarize the news for you

• Subagents: For complex tasks, Claude splits the work and runs multiple sub-tasks in parallel, each with its own context and job, then merges the results into a single deliverable.
• Dispatch: A continuous conversation thread between desktop and mobile. From the Claude Mobile App you can hand Claude a task that uses everything on Claude Desktop, files, connectors, plugins, even desktop apps. The catch: both the desktop app and the computer have to be on. Personally I'd still call this one beta, not fully polished yet, and I'm looking forward to the complete experience.
• Projects: Group related tasks into their own workspace, with their own files, context, instructions and memory, like Projects on claude.ai (Claude Chat) but running locally and built around the tasks you run through Cowork.
• Browser use: Connect the Claude in Chrome extension so Claude can visit websites directly, interact with their content and carry out your requests, even on pages with no API.
• Computer use (research preview, macOS): When there's no suitable connector or plugin, Claude navigates your machine directly, clicking, typing and opening apps like a real user. The priority order is connectors first, then Chrome, then screen interaction. You'll get a permission prompt before Claude touches each app, and you can set a blocklist for anything you want to restrict.
• Plugins/Connectors: Extend Claude with things it doesn't have built in. Rather than trying to be one all-in-one app, Claude focuses on doing its core job well and then connects out to third parties, partners and other tools. You can hook up Gmail, Drive, Google Calendar to have Claude check your schedule, draft mail or create docs. Or add dev and design plugins like Figma, GitHub, so every sketch can be delivered as code and pushed to a repo you control on GitHub with ease.
• Protected environment: Cowork runs in an isolated space. Claude only reads, creates and edits files in the folders you allow, it can't reach outside that scope. You stay in control of your own safety.

💻 Code

Claude Code

This is essentially Claude Code running right inside the desktop app, a software development environment with a visual interface.

With Code, Claude works directly in your project/codebase: reading what's there, creating or editing code, running commands. A visual diff shows every change, the built-in terminal shows the commands running, and git tracks the whole history so if you slip you can roll back immediately.

Unlike Cowork, which only works within the folders you allow, Code has full access to the file system, terminal and the project's development tools.

You choose where it works:

• Local: Access a folder on your machine; Claude works directly with those files, reads the project, creates a CLAUDE.MD (via /init), accesses local tools, and runs a dev server so you can preview in the browser.
• Remote: Connect a GitHub repo and Claude runs in a cloud environment. The session keeps going even after you close the app, kick off a big refactor, go to sleep, and check the result in the morning. Good for large codebases or when you want to keep development off your local machine.

Three interaction modes to control how autonomous it is:

• Ask: Claude proposes each change and waits for your approval. Review the visual diff, accept or reject before anything is modified.
• Code: Claude applies file changes itself, but still asks before running any terminal command.
• Plan: Claude lays out the entire approach before touching a single line of code. A separate plan viewer lets you review and track progress throughout.

You can run multiple sessions in parallel across different projects, filtering by status (Active / Archived) and environment (Local / Cloud) from the sidebar.

(The Code tab is available on the Pro, Max, Team and Enterprise plans.)

As a senior developer, my workflow with these three modes is pretty clear:

Chat for quick research and debugging, reading docs, analyzing logs, clearing up things I don't understand. I also use it to extract data from docx, pdf and the like into markdown, broken down section by section so the planning and coding that follow are easier to breathe through.

Cowork is where I go before I start coding. I create a project, pull in both the business docs and the existing codebase, and let Claude analyze the relationship between them. Plan the implementation, trace data flow across microservices, clear up unknowns in the architecture with specific questions (e.g. which service does the endpoint apis/abcxyz originate from, which services does it touch). The best part is that Claude keeps context across the business logic and the technical layer, something I used to have to do by hand.

Claude Code is the final step. Once the plan is clear, I move here to execute. I don't let Claude code right away. I force it to plan first. After countless rounds of review, when it finally sits right with me, I accept and Claude Code executes under my supervision.

Those three steps run in order, and each mode does its own job well.

What workflow are you using Claude with? 👇