RightHand AI

Ask vs. Assign: Match Claude's Help to the Job

Claude now works in a single workspace. Cowork is built into every conversation, so there is no separate Chat versus Cowork switch. The old advice was to choose the right mode before you started. The new advice is simpler: let the task set the depth, and tell Claude what you want.

Two kinds of work

An ask is a question or a short task where you stay in the loop on every turn. You send a request, read the answer, and steer. Explaining a clause in plain English, tightening a paragraph, and brainstorming subject lines are all asks.

An assignment is a job with a defined finish line. You describe the outcome, hand it off, and review the result. Comparing twelve vendor quotes and producing a one-page recommendation is an assignment. So is turning a folder of meeting notes into a status report.

A simple test

Try to describe the finished result in two sentences. If you can, the work is probably an assignment, and Claude can carry more of it. If you cannot, start with asks, and let the conversation shape the job until you can.

How to hand off an assignment

Most weak results from larger jobs come from thin instructions. A good assignment names four things:

  • The outcome: what the finished piece looks like, including format and length.
  • The material: which files, notes, or facts Claude should work from.
  • The limits: what Claude should leave alone, and what it should flag instead of deciding.
  • The checkpoints: where you want to review before Claude continues.

Checkpoints matter most on longer jobs. A review after the first section costs a minute. A wrong turn discovered at the end costs the whole run.

Deciding how much room to give

Claude's Permission mode sits right under the message box and has two settings. Auto is the default on Pro, Max, and Team plans, so switch to Manual yourself: click Auto below the message box and choose Manual. Manual means Claude asks before using new tools or opening new sites, and it is the right starting point. Auto means Claude runs on its own and pauses to ask if anything looks unsafe, which suits work you have watched run cleanly before.

Think of it as a ladder with three rungs:

  1. Ask every time (Manual). The default for anything new. You see what Claude wants to do and approve each action.
  2. Let a trusted routine run (Auto). Reserve this for a routine you have watched run cleanly more than once, such as renaming files in one folder by a pattern you have verified.
  3. Always needs you. Deleting something, sending something on your behalf, posting something publicly, or moving a file with someone else's name on it should stay at "ask me every time," no matter how much you trust the routine.

Read each permission prompt before you approve it. It is the one sentence that tells you what is about to happen. Starting on Manual is slower on day one and faster by the end of the month.

What to do this week

Pick one task you currently do as a long back-and-forth. Write it as an assignment with an outcome, material, limits, and one checkpoint. Run it, review the result, and note what you would add to the instructions next time. That note is the start of a reusable prompt.

Go deeper

Module 6 of Claude Unlocked, Claude as Your Workspace: Ask, Assign, and Hand Off, covers section 6.2 (Ask vs. Assign: Match the Depth to the Mission) and section 6.4 (The Permission Ladder) with step-by-step prompts. The guide has 78 prompts across 10 modules, is updated for the October 2026 interface, and costs $39.99, one time.

Read about Claude Unlocked

Next in this series: where Claude's work actually runs, and what to read before you connect anything.

Published by RightHand AI. Drafted with AI assistance and reviewed and finalized by RightHand AI before publication.

Claude Unlocked is independent training material about Claude, published by RightHand AI. It is not an official Anthropic product.

RightHand AI, a DBA of Ussen Technologies, LLC, 5233 Bellaire Blvd., Suite 638, Bellaire, TX 77401.

All guides