Your First Task

The quality of Claude's work depends heavily on how you describe the task. This isn't about magic prompt formulas — it's about being specific the way you'd be specific with a human colleague.

Be clear about the output. Don't just say "help me with my files." Say "organize the files in my project folder into subfolders by file type, and create a README listing what's in each folder."

Give context. If you're asking for a report, mention who it's for and what decisions it should support. If you're organizing files, explain the logic you want.

Define success. What does "done" look like? A spreadsheet? A sorted folder? A summary document? Tell Claude what you expect to receive.

Here's the difference:

Vague: "Clean up my files."

Specific: "I have 30 files dumped in my work folder — a mix of PDFs, images, Word docs, and spreadsheets. Sort them into subfolders by type (documents, images, data), rename them with a consistent format (YYYY-MM-DD-description), and create an inventory spreadsheet listing every file with its original name, new name, and folder."

Here's something that sets Cowork apart from other AI tools: Claude shows you its plan before executing.

When you give Claude a complex task, it doesn't just start working blindly. It breaks the task into steps, shows you its reasoning, and waits for your approval on significant actions. This transparency is by design — you're always in the loop.

A typical plan looks like this:

1. First, I'll scan the folder to see what files exist

2. Then I'll categorize them by file type

3. I'll create the subfolder structure

4. I'll move and rename each file

5. Finally, I'll generate the inventory spreadsheet

Always review the plan. This is your chance to catch misunderstandings before any work happens. If Claude's plan doesn't match what you expected, correct it now — not after it's reorganized 200 files the wrong way.

Safety tip: For tasks that modify existing files, tell Claude "don't delete anything" unless deletion is specifically what you want. Claude respects this instruction.

Once you approve the plan, Claude starts working. And unlike a chat response that just appears all at once, Cowork shows you what it's doing in real time.

You'll see Claude's reasoning as it works — which files it's reading, what decisions it's making, which sub-tasks it's executing. This isn't just a loading bar. It's a transparent window into the agent's thought process.

This matters for two reasons:

Trust. You can see that Claude is doing what you asked, not something unexpected. If it starts heading in the wrong direction, you catch it immediately.

Learning. Watching Claude work teaches you how it approaches problems. You'll start to understand its strengths, its patterns, and where it might need clearer instructions next time.

You're not a passive observer. At any point during execution, you can interrupt Claude and give new instructions. This is one of Cowork's most powerful features.

Common mid-task corrections:

- "Wait — skip the image files, I'll handle those myself."

- "Actually, use the client name instead of the date for the file prefix."

- "Stop. Let me review what you've done so far before you continue."

- "Add a 'priority' column to that spreadsheet you're building."

Claude handles interruptions gracefully. It pauses, acknowledges your correction, adjusts its approach, and continues. You don't lose progress — it picks up where it left off with the new instructions incorporated.

Think of it like working with a junior colleague. You give them a task, check in on progress, and redirect when needed. The difference is that Claude doesn't get frustrated when you change your mind.