Claude Cowork walkthrough

Prompt Claude Cowork for the outcome, not the answer

Most people treat Cowork like chat: "summarize this", "help me with that". It works far better when you name the finished artifact, stage clean inputs and spell out the format. This walkthrough shows that pattern on four everyday jobs.

▶ Interactive · 8 steps5 minFor RevOps, marketing and enablement
Click through it. Use Next, the arrow keys or the dots. Orange boxes show where to look.Expand ⤢ for full screen

Illustrative recreation of the Claude Cowork desktop app with fictional sample data. Not the actual product; screens and labels can change as Cowork ships updates.

One pattern, four use cases

Outcome, clean context, explicit format. Then the same pattern on explainers, decks, reports and admin.

  1. Define the outcome

    Name the artifact you want back (report, deck, prototype, dashboard), stage a clean folder and state the format.

  2. Ask for an HTML explainer

    Concept, audience, colors and format in one prompt; the output is a working file.

  3. Iterate on the prototype

    Refine in plain language: simplify, recolor, declutter.

  4. Structure a presentation

    Length, audience and narrative arc in, slide-by-slide outline with speaker notes out.

  5. Refine the slides

    One instruction at a time: stat blocks, diagrams, tighter notes.

  6. Write a structured report

    Point Cowork at transcripts and connected tools and ask for summary, findings, evidence, risks and next steps.

  7. Extract receipts into a table

    Attach the files, name the fields, and have Claude write the table into a connected tool.

  8. Call a saved skill

    Type / to call a reusable skill and add the new context after it.

How JayOh uses this

Prompts become skills, skills become workflows

When a prompt works twice, we turn it into a skill so the whole team gets the same result. Our guide on how to use skills in Claude covers that step.

  • Artifact-first briefs. Every request names the deliverable and its sections.
  • Clean context folders. Only relevant, well-named files go in.
  • Brand in the prompt. Colors, tone and structure are stated, then saved as a skill.
  • Connectors for reach. Notion, Slack and call recorders feed reports without copy-paste.
Guardrails we set first

Keep a human in the loop

  • Desktop app, paid plan. The website is Chat, not Cowork.
  • Review before sharing. Generated decks and reports get a human read.
  • Least-privilege connectors. Read first, write when needed.
  • No secrets in prompts. Keep credentials out of files Cowork reads.
8screens in this walkthrough
3parts to every prompt: outcome, context, format
4everyday use cases shown
$980illustrative receipts extracted

Common questions

Short answers before you set this up.

How should I prompt Claude Cowork?

Name the finished artifact, give it a clean folder of relevant files, and state the format and sections. "Create a two-page memo with findings, risks and recommendations" works far better than "summarize this".

What is the difference between Claude Cowork and Claude Code?

Cowork brings Claude Code's outcome-driven, multi-step execution into a desktop interface built for knowledge work. Claude Code gives finer control and is aimed at building software and scripts.

What are skills, connectors and plugins in Cowork?

Skills are reusable instructions you call with a slash. Connectors let Claude read and write apps like Gmail, Slack and Notion. Plugins bundle skills and connectors into a workflow for a recurring job.

Where does Claude Cowork run?

Cowork runs in the Claude desktop app on Mac or Windows on a paid plan. It works on folders you grant it on your computer, can use connectors to apps like Gmail, Google Drive, Slack and HubSpot, and can control a browser or desktop apps when you allow it. The website is Chat, not Cowork.

Want your team prompting for outcomes?

JayOh builds the skills, connectors and enablement that get GTM teams consistent results from Claude.

Talk to JayOh