Field Guide● Luseen & Co.● Working with Claude● № 001

Reference · 28 Jul 2026

Download as PDF ↓

Chat answers.
Cowork delivers.

A working guide to Anthropic's two modes — where each one fits, the habits that make them useful, and how to keep your data yours.

01 — When to use which

Chat vs Cowork

Same assistant, two gears. You start both from the same message box — the difference is how much you hand off.

Chat
  • One message at a time — ask, draft, think out loud.
  • Answers, writing, analysis, and artifacts: docs, apps, visuals.
  • Everywhere — web, mobile, desktop. On every plan.

Best for · quick turnarounds and back-and-forth.

Cowork
  • Goal in, finished work out — it runs multi-step on its own.
  • Reads and writes your local files; can schedule recurring tasks.
  • Desktop (macOS / Windows); in beta on web & mobile. Paid plans. Anthropic

Best for · reports, file cleanup, research you’d rather not babysit.

Rule of thumb If you’d sit and watch it happen, use Chat. If you’d rather walk away and come back to a finished file, use Cowork.

02 — Habits that pay off

Practical

Most of the leverage isn’t in the prompt — it’s in the setup.

  1. Front-load context. Use Projects to keep instructions, files, and tone in one place so you’re not re-explaining yourself every session.
  2. Say the outcome, not the steps — especially in Cowork. “Turn these three CSVs into a clean Q3 summary” beats a play-by-play.
  3. Grant Cowork only the folders it needs. It works in a sandbox and touches nothing you haven’t allowed.
  4. Turn repeat work into artifacts and scheduled tasks — a reusable dashboard, a weekly report that writes itself.
  5. Connect your stack. Connectors let Claude pull from the tools you already use instead of copy-paste.
  6. Mind the thumbs. Rating a chat is useful feedback — but submitting it shares that full conversation with Anthropic.

03 — Keep your data yours

Privacy

This is the part I get asked about most. Three settings worth knowing before you put anything real into a chat.

Keep your chats out of training

Settings → Privacy → Help improve Claude → Off

On consumer plans (Free, Pro, Max), leaving this toggle on makes new chats eligible to train future models and stretches retention to as long as five years. Switching it off keeps them out and returns you to the 30-day standard. Anthropic

Honest limit: turning it off stops future use — it doesn’t pull back anything already used to train a model. And conversations flagged for safety review can still be examined regardless of the setting. Retention policy

For one-off sensitive questions, go Incognito

Ghost icon · top of screen

Incognito chats aren’t saved to your history and aren’t used to improve Claude — even if your training toggle is on. A clean option for the query you don’t want on the record. Anthropic

A shared link is a public web page

Settings → Privacy → Shared Chats

Chats and artifacts are private by default. Hitting Share creates a public snapshot anyone with the link can open — and if that link lands somewhere a crawler can reach it (a forum, a social post), search engines can index it like any other page.

That’s exactly what surfaced over the weekend of 25–27 July, when shared Claude chats and artifacts turned up in Google results. Axios TechCrunch

The safe habit: review and unshare anything public under Shared Chats. Revoking stops new views, but cached or already-indexed copies can linger — so don’t put anything you wouldn’t publish into a chat you plan to share.