4 min read

AI & Agents

Claude Has Three Modes. Most People Only Ever Use One.

Typing a question and reading the answer is the weakest way to use Claude. Here is when to switch to Code, Cowork or Projects, with a quick decision tree and the context each one rewards.

Claude Code vs Cowork vs ProjectsClaude CoworkClaude Projects

Claude Has Three Modes. Most People Only Ever Use One. · Products Decoded
By Debashis Nayak · Products Decoded21 Mar 2026

You open Claude, type a question, read the answer, close the tab. That works, and I think it is also the weakest way to use it.

Claude is not one tool with one door. As of March 2026 it behaves like three, each built for a different job: Claude Code, Claude Cowork and Claude Projects. Picking the wrong one is how you end up re-explaining context, losing work mid-session, and wondering why the AI shift has not reached your workflow yet. (Product details below are as I understood them in March 2026, so check current plans before you rely on them.)

What are the three Claude modes?

A one-line version of each:

ModeThink of it asBest at
Claude CodeThe builderSoftware that spans many files
Claude CoworkThe collaboratorDeep work on your own files
Claude ProjectsThe institutional memoryRecurring tasks with fixed context

Same AI. A very different experience depending on which one you open.

When should you use Claude Code?

Claude Code is closer to a junior developer on your team than to a chatbot. It reads your whole project, holds context across it, and writes files straight into your folder. No pasting generated text somewhere and then asking it to apply that to your code.

The difference that matters is that it sees the entire project: folder structure, existing files, dependencies. Most AI tools only work with whatever you put in one prompt.

Switch to Code when:

  • You are building software, a tool or an internal app.
  • A change has to land across several files at once.
  • You want a long, hands-free build session without losing context.
  • You need it to connect to outside tools such as GitHub, Jira or Slack.

A concrete case: you want a small internal tool with a handful of files. Pasting snippets into a chat leaves you as the glue. In Code, you describe the requirement and the files are written in place.

To get more out of it, add a context file (a context.md or similar) so it already knows who you are, what you are building and what the constraints are. That is how you stop re-onboarding it every session.

When should you use Claude Cowork?

Cowork lives on your desktop. It reads your files and creates new ones. Because it is tied to a folder you point it at, its output starts to sound like you, simply because it is reading your material.

Setup is a bit heavier, roughly ten minutes, and what you get for it is sessions that do not break. You drop .md or .txt files into the folder, point Claude at it, and go. No “I don’t have access to that file” halfway through.

Switch to Cowork when:

  • You are doing deep work such as analysis, writing, spreadsheets or strategy docs.
  • You want actual files created, not just text generated.
  • The session will run long and has to stay coherent.
  • The output should sound like you rather than like a generic AI.

A concrete case: a strategy doc that draws on a dozen of your own notes. In a plain chat you would paste them in each time. Here they are already sitting in the folder. If you want the folder layout I use, I wrote it up in why Claude gives you generic answers.

When should you use Claude Projects?

I think Projects is the most underrated of the three. You upload your files and instructions once, and every new chat inside that project starts with that context. No re-uploading, no “as I mentioned in the previous conversation.”

It suits recurring work: weekly reports, the same type of analysis, a knowledge base, anything in a consistent format. Your context also follows you across devices, so browser, phone and desktop all see the same setup.

Switch to Projects when:

  • You do the same task every week and want consistent output.
  • You are tired of re-teaching Claude your preferences.
  • You want context to persist across devices.
  • You are building up long-term memory for a body of work.

A concrete case: a weekly report that always has the same sections. Write the instructions once, properly, and each week’s chat begins already briefed.

What do all three modes have in common?

The quality of what comes out tracks the quality of what you put in. Each mode reads something different, so the skill changes with it:

  • Code reads your whole project, so the skill is writing clear requirements.
  • Cowork reads your files, so the skill is having good files.
  • Projects reads your instructions, so the skill is writing them once, properly.

In all three, I would say the AI is rarely the bottleneck. The limit is how well you can structure and communicate what you want. My guess is that engineers and PMs who already keep documentation habits will get far more out of these tools than someone typing questions into a chat window, because the tools reward people who already think clearly. I have not measured that, so treat it as an observation rather than a statistic.

What does it cost, and what else should you know?

As of March 2026, all three need the Pro plan at $20 a month. Cowork is desktop-only. Extended Thinking works everywhere, and if you are not switching it on for hard problems, you are leaving the best reasoning unused.

A quick decision tree

  • Building something? Use Code.
  • A deep work session that needs your files? Use Cowork.
  • Recurring tasks, same format, saved for good? Use Projects.

Before your next session, ask which of those three sentences describes the job. If none of them does, a plain chat is probably fine. If one does, switch doors first and then type the question.