Claude now has 3 layers.

Each one does a different job:

1. Claude Chat (browser/app)

Use it when the work is thinking in words. It turns vague ideas into structure fast: drafts, summaries, plans, decisions. You get clarity and high-quality writing, but you still execute the work elsewhere.

✦ Rewrite a draft in your voice, cut it by 30%

✦ Create a decision memo with trade-offs

✦ Simulate customer objections, draft responses

2. Claude Code (terminal agent)

Use it when the work lives in a repo. It navigates your codebase, edits files, and runs commands autonomously. You describe the goal, it ships the code. But you review everything before it merges.

✦ Scaffold a full app with auth, database, API routes

✦ Refactor a legacy module across 40 files

✦ Run a security scan, patch vulnerabilities

3. Claude Cowork (desktop agent)

Use it when the work is workflows, not thinking. It handles the repetitive operations you do manually: organising, extracting, filling, moving. No coding needed. Runs in a sandboxed VM on your desktop.

✦ Extract tables from 50 PDFs into one spreadsheet

✦ Rename and sort hundreds of files by type and date

✦ Pull data from Gmail, update a report in Google Drive

The decision rule:

✦ Thinking and content → Chat

✦ Code and systems → Code

✦ Files and app workflows → Cowork

The advantage isn't "using AI."

It's knowing which layer to reach for.

I wrote the complete setup guide for all three.

Free to read (drops Sunday) → charliehills.substack.c…

Repost ♻️ to help someone in your network.