Claude Cowork — Win Back 15 Hours a Week — 1. Claude Cowork, Your Digital Coworker

19 min read min de lecture
Chapter 01

Claude Cowork, Your Digital Coworker

Chapter 1 of 10 · 10%

Chapter objectives

  • Understand the difference between a chatbot and an executor
  • Learn to delegate a task the way you would to a human
  • Structure a request to get a usable deliverable

A coworker, not a search engine

Claude Cowork doesn't just answer questions: it completes tasks end to end. You give it a goal, it breaks it down, executes, and brings you back a deliverable. Think "ultra-fast, tireless intern" rather than "search engine".

This distinction changes everything. When you type a question into a search engine, you get links: it's then up to you to read, sort, compare, and write. The work stays entirely on your shoulders. When you delegate to Claude Cowork, you describe the end result you need — an email ready to send, a 200-word client memo, a cleaned-up table — and that result is what you receive. You move from the role of executor to the role of delegator, and that role is precisely what frees up your time.

For Marc, this shift in posture is decisive: instead of asking "what's the VAT on this product?", he'll say "draft me a client memo explaining this VAT change, clear tone, 200 words". The first phrasing gives him information he'll still have to turn into a document. The second gives him the document. For the same stakes, the second saves twenty minutes — and Marc handles dozens of requests like this every week.

Many professionals miss this lever because they've grown used to voice assistants and consumer chatbots: you ask a small question, you get a small answer. With Claude Cowork, that reflex holds you back. The right reflex is the one you already have with a competent colleague: you don't ask them "what's a payment reminder?", you tell them "follow up with the Dupont account about their missing invoices, be courteous but firm".

The golden rule: delegate like you would to a human

The more your request looks like a brief you'd give a team member, the better the result. Provide three things: the context, the precise task, and the expected format. This trio works because it reproduces what a good manager does naturally: situate the work, define the deliverable, specify the constraints.

The context is everything your counterpart can't guess: who you are, who your clients are, what the situation is, what's at stake. Marc would never tell a new intern "write a follow-up email" without saying to whom, why, and what the business relationship is. Claude Cowork is in the same position as a brilliant intern who arrived this morning: highly capable, but it knows nothing about your business until you tell it.

The precise task delimits the work. "Help me with this email" is vague; "write a reply that politely declines the requested deadline and proposes two alternative dates" is precise. The more clearly delimited the task, the fewer surprises you'll get. Finally, the expected format frames the deliverable: length, tone, structure, language. It's often the forgotten ingredient, and yet it's what turns a correct answer into a directly usable deliverable.

PROMPT
Context: I run an accounting firm, my clients are tradespeople.
Task: write an email reminding a client that their invoices for the quarter are missing.
Format: short email (max 120 words), courteous but firm tone, with a clear deadline.
flowchart LR
  C["Context"] --> B["Complete brief"]
  T["Precise task"] --> B
  F["Expected format"] --> B
  B --> L["Directly usable deliverable"]
The 3 ingredients of good delegation — just like with a human team member.
Often finish with "and give me 2 tone variants". You then pick the one that best fits your relationship with this client.

Anatomy of a brief that works

Let's look closely at what separates a weak brief from a solid one, using a real case from Marc's firm. Weak brief: "Write an email for a client who's late on payment." The result will be grammatically correct, but generic: bland tone, no mention of the amount or the history, hollow phrases. Marc will have to rework everything — he's gained nothing.

Solid brief: "Context: the client Atelier Bernard, a carpenter and a good client for 6 years, has two unpaid invoices (€1,840 total, 45 days overdue). This is unusual for him. Task: write a follow-up email that first asks whether everything is okay, recalls the two invoices with their amounts, and offers a payment plan if needed. Format: 150 words maximum, warm but professional tone, email subject line included." The result will be usable as is, because all the information that makes the email good was in the brief.

Remember this simple rule: the quality of the output is capped by the quality of the input. If you provide concrete details — names, amounts, dates, history, relationship stakes — you get concrete text. If you provide vagueness, you get vagueness. That's not a limitation of the tool; it's the same logic as with any team member.

PROMPT
Context: the client Atelier Bernard, a carpenter and a good client for 6 years, has two unpaid invoices (€1,840 total, 45 days overdue). This is unusual for him.
Task: write a follow-up email that first asks whether everything is okay, recalls the two invoices with their amounts, and offers a payment plan if needed.
Format: 150 words maximum, warm but professional tone, email subject line included.

The first draft is never the last: iterate

Classic beginner mistake: judging the tool on its first answer. If the result isn't right, don't start over from scratch — refine. Say what's wrong, just as you would with a colleague: "too formal, loosen the tone", "too long, cut it in half", "add a sentence about our meeting next week". Each piece of feedback is applied instantly, and the conversation keeps all the context you've already given.

This feedback loop is your most powerful tool. With a human collaborator, each back-and-forth costs hours or days; here, it costs ten seconds. Marc has gotten into the habit of treating the first answer as a working draft: he asks, he reads, he adjusts in a sentence or two, and he gets his final version in under two minutes. That's very different from the "write the perfect prompt on the first try" approach, which is both exhausting and unnecessary.

Bonus tip: when a result really pleases you, ask "what in my request allowed you to produce this?". You'll learn to reproduce your best briefs — and you'll improve much faster than by piling up random attempts.

What you can delegate starting today

To get started, target tasks that combine three characteristics: they are frequent (the gain repeats), time-consuming (the gain is noticeable), and low-risk (an imperfection is easy to catch). Drafting the first version of an email fits perfectly; a client's tax return does not — at least not without expert review.

  • Writing: emails, internal memos, meeting minutes, letters, the firm's LinkedIn posts
  • Summarizing and monitoring: legal texts, specialist articles, long reports, vendor documentation
  • Data analysis: fee spreadsheets, bank exports, invoicing trackers
  • Reformatting and cleanup: document formatting, harmonizing dates, de-duplicating lists
  • Preparation: agendas, client meeting outlines, questions to ask a prospect
One-off question"What's the new VAT exemption threshold?" — you get a piece of information, the writing work remains to be done. Useful, but limited gain.
Delegated task"Write a one-page memo for my clients affected by the new VAT threshold, with the 3 actions to take before June." — you get the final deliverable. That's where the saved hours are.

What not to delegate (and why)

A good manager also knows what they don't delegate. Three families to keep under human control. First, binding decisions: Claude can prepare a comparison or a line of argument, but you're the one who signs, validates tax advice, or commits to a client. Second, genuinely sensitive data: before pasting confidential information, check your company's policy and anonymize what you can — replace names with "Client A", mask account numbers. Finally, topics where you can't judge the quality: if you can't spot an error in the result, have it reviewed by someone who can.

This caution isn't distrust, it's professionalism. Marc applies the same rule as with his junior staff: he gladly hands them the preparation, never the signature. AI produces fast and well, but it can confidently assert something inaccurate — this is called a "hallucination". On high-stakes topics, your review remains the last line of defense.

Practical confidentiality rule: if you wouldn't email this data to an external vendor, anonymize it before handing it to the AI. "Client A, €12,400 in fees" is almost always enough to work with.

Beginner pitfalls

First pitfall: giving up after one bad result. A vague brief gives a vague result; the fix is to enrich the brief, not to conclude that "it doesn't work". Second pitfall: trying to automate everything at once. Start with a single task, master it, then expand. Third pitfall: never proofreading. Production speed doesn't excuse you from quality control, especially in the first weeks while you calibrate your trust.

Fourth, subtler pitfall: keeping your good phrasings to yourself. When a brief works well, write it down somewhere — a simple shared document is enough. In chapter 5, you'll turn these phrasings into a real workflow library for the whole team. The saved hours don't come from one genius prompt; they come from dozens of good briefs reused week after week.

🛠️ Your turn

Context

Marc receives a confusing email from a tradesperson client: three questions mixed together, a veiled complaint about a delay, and an unreadable attachment. He needs to reply clearly without offending the client, and he wants to use the opportunity to test his first real delegation. Put yourself in his shoes with an equivalent task from your own week.

Instructions

  1. Pick a real repetitive task from your week (email, memo, summary).
  2. Write a "Context / Task / Format" brief like in the chapter examples, with concrete details: names, amounts, history.
  3. Submit the brief and read the first answer without treating it as final.
  4. Refine in one or two sentences: adjust the tone, the length, or a missing detail.
  5. Ask for 2 tone variants and compare them: which one best fits your relationship with this recipient?
  6. Save the best brief phrasing in a document: it will be the first brick of your library.
  7. Estimate how long this task used to take you, and how long it took this time.
Hint — If the result is too generic, it's almost always missing context or a precise format. Reread your brief: could a human colleague have done the job well with only that information?

In summary

  • Claude Cowork executes complete tasks; it doesn't just answer.
  • A good brief = context + precise task + expected format.
  • The quality of the output is capped by the quality of the input: give concrete details.
  • The first draft is a rough draft: refine in conversation instead of starting over.
  • Start with your frequent, time-consuming, low-risk tasks.
  • Keep binding decisions and sensitive data under human control.
  • Asking for several variants speeds up the decision.
  • Save your best briefs: they will become your reusable workflows.

Quiz — check your understanding

1. What posture should you adopt with Claude Cowork?

It completes tasks end to end: you give it a goal and a format, and you receive a deliverable.

2. What does a good brief contain?

The context and the output format matter as much as the task itself: they turn a correct answer into a usable deliverable.

3. The first result doesn't suit you. What do you do?

The conversation keeps all the context: a ten-second piece of feedback ("too formal, shorten it") is often enough to get the final version.

4. Which task is the LEAST suitable for getting started?

You delegate the preparation, never the signature: binding decisions stay under human control.

5. How should you handle confidential client data?

Anonymization lets you work efficiently while respecting confidentiality: "Client A, €12,400" is almost always enough.

Auteur(s)

R

REHOUMA Haythem

Haythem Rehouma est un ingénieur et architecte IA et cloud, formateur et enseignant technique, avec un profil orienté IA médicale, AWS, MLOps, LLM/RAG et vision par ordinateur.