Hand Claude Cowork your first task

Estimated time: 15 minutes

Learning objectives

By the end of this lesson you'll be able to:

• Walk a Cowork task end-to-end: through delegation, mid-task steering, and final review
• Respond to Cowork's clarifying questions in a way that produces better output
• Course-correct when Claude heads in the wrong direction

Watch one go end-to-end

This video walks through a single, real piece of work from delegation to deliverable: the prompt, the clarifying questions Claude asks, the moment the user reroutes mid-task, and the finished result at the end. Notice the rhythm — it's less like prompting a chatbot and more like briefing a smart colleague who's about to disappear into a task and come back with something done.

Key takeaways

• **Claude asks before it commits. ** Most non-trivial tasks start with a clarifying question or two. The questions are how Claude closes context gaps before the work starts, not friction in the way.
• **You can steer mid-task. ** If Claude is heading in the wrong direction, jump in. You don't have to wait for it to finish and start over.
• **The finished deliverable is the artifact, not the chat. ** Claude makes changes to the files on your computer. Your job at the end is to review it the way you'd review a colleague's work.

Delegate your first task

You picked a task in previous lessons. Now is when it leaves your plate. Open Cowork, point it at the folder where the context for the task lives, turn on any connectors that hold relevant context for the task, and write your prompt.

A good Cowork prompt does three things:

• **Names the deliverable. ** "A one-page brief," "a slide for the QBR," "a ranked list with notes per candidate. " Specifics about format and length save you a regenerate.
• **Names the inputs. ** Which folder, which channels, which date range, which application. Cowork is only as good as the context you point it at.
• **Names additional specifics or nuances Claude would need to get the job done. ** _"I want to see base, best, and worst case scenarios that also account for the 3 new locations we opened in Q3 last year. "_ That sentence directs Claude on the type of output you're looking for, and provides key insights from your expertise that Claude might not see.

Build an example Cowork prompt in the interactive below.

Build the prompt

Deliverable

four-page memo slide for the QBR ranked list with notes

Inputs

Q3 Competitive Review folder last quarter's memo as format reference analyst-call PDFs only

Nuance

for the executive team at the leadership offsite — help them decide on the new pricing tier lead with the recommendation flag anything we can't verify

Write a message…

Ask Sonnet 4. 6

0 of 3 rows complete

That's a complete delegation — deliverable, inputs, and the context Cowork can't guess.

Leave a row empty and Cowork will usually ask you for it.

If you're used to prompting a chatbot, this will feel like more upfront work — it is. The trade is that more of the back-and-forth happens upfront, before Claude starts, instead of across five rounds of "actually, can you also... "

Answer the clarifying questions

Claude asks questions to clarify any ambiguities about your objectives before it gets into the work. In Chat, you collaborate as you go — context shows up turn by turn. In Cowork, you delegate and it returns with a completed artifact. Anything that might trip it up — any gaps in its understanding — comes up through the questions before it dives in.

Most are easy: Claude presents a couple of options and you click the one that fits. If none of the options fit, you can answer in your own words.

Below are some examples of the questions you might receive from Cowork after submitting the prompt you built above.

Cowork is asking you a few things

###

1 of 4

Skip

↑↓ to navigate Enter to select Esc to skip

Pick an answer to see what it tells Cowork.

Try again

Steer mid-task

Watch Cowork's plan and progress as it works. If it's going off-track — wrong source, wrong format, wrong tone — interrupt. You can also stop the run if it's substantially off, refine the prompt, and start again with what you've learned. Most people's instinct from Chat is to wait until the response is done and then regenerate. Resist that. Cowork is built for course corrections, and the cost of a redirect is low.

Steer the example task below and see how Claude responds.

Steer mid-task

Help me draft the Q3 competitive memo for the leadership offsite.

I'll read the analyst-call PDFs and last quarter's memo, cross-reference each competitor's pricing changes against our positioning notes, then draft against the Q2 board-memo template.

Read 3 files, searched 2 folders

Writing draft · section 2 of 4

Progress

1Read source PDFs and last quarter's memo

2Cross-reference pricing changes

3Outline three competitor moves

4 Drafting against Q2 board-memo template

5Review against original prompt

Write a message… (click Queue to steer mid-task)

Queue

**You don't need to wait for it to finish and then regenerate. ** You can correct the plan while it's still running — Cowork picks up from where it was.

Reset

Review the finished deliverable

When Claude is done, you can preview the file in the app. The file is also in your folder, so you can open it up and see it there.

Review it the way you'd read a draft from someone you trust but don't yet fully know — with your discernment turned on. Things to check, especially the first few times:

• **Does it meet the actual objective? ** Or did it produce something subtly different from what you asked for?
• **Are the facts accurate? ** If Claude references facts or figures, confirm they're correct by reviewing the source material. Start by asking Claude to identify the docs it pulled from, then check them yourself.
• **Does anything sound made up? ** A specific date, name, or quote that you can't trace to an input is a flag, not a feature.

If the draft is mostly right, tell Claude what to change rather than starting over. Claude remembers the conversation and edits faster than it regenerates.

If the draft is wrong in a load-bearing way, the prompt was missing the load-bearing piece of context. Point Claude to the new context and ask it to make adjustments.

Try it now

Take the task you identified in Lesson 3 — or pick one now. Hand it to Cowork using the process you just walked through — name the deliverable, the inputs, and any nuance; answer the clarifying questions specifically; steer if it drifts; review the result with your discernment on. The first run is the one that teaches you the most.

What's next

You've now done it once. In Module 2, you'll start making Cowork yours — keeping the context that worked, the workflows you want repeated, and the expertise your team relies on, so you don't re-supply it every time you start a task.

Feedback

As you progress through the course, we'd love to hear how you're using concepts from it in your work, plus any feedback you may have. Share your feedback here.

Acknowledgments and license

Copyright 2026 Anthropic. All rights reserved.