How do I use Claude Cowork?

AI used to be seen as an assistant that generates responses, but now, with the launch of Claude Cowork, Anthropic is proposing a different model for interacting with AI – one that focuses on achieving a specific outcome and setting tasks, rather than simply using a question-and-answer format.

Why isn’t this just a chat with files?

In the standard version, Claude responds to messages but cannot access your files. Every action it takes is a prompt that is created manually.

In Cowork, it’s different. You grant Claude access to specific files in a folder, and then

1. Claude devises a plan itself,
2. shows it to you for approval,
3. reads, edits and creates files,
4. and, if necessary, opens a website or reads specific applications.

The key difference from the usual way of interacting with AI is that you don’t have to spell out ‘first do this, then do that’; instead, you simply describe how things should be done.

Examples of prompts that actually work

1 Organising files
2"The ~/Downloads folder is a complete mess. Sort the files by type
3and date, delete duplicates, and rename the screenshots to reflect their content."
4
5Report from raw data
6"Here are 5 CSV files containing sales figures for the quarter. Compile them into an Excel spreadsheet
7with formulas, a summary table and a graph showing trends by month."
8
9 Synthesis from sources
10"Read all the PDFs in the /contracts folder and create a summary table:
11counterparty, term, key risks."
12
13 Accessible via a browser (via Claude in Chrome)
14"Find 10 competitors for my product online, compile their prices
15and features into a comparison table."
16
17Processing attachments
18"Review the attachments in the /invoices folder, extract the amounts and dates,
19and compile a list of invoices due for payment in Excel."
20
21Recurring task (scheduling)
22"Every Monday at 9.00 am, compile a summary of new enquiries
23from the /leads folder into a single report and save it to /reports."

Who is this best suited to?

Cowork is not designed for developers, but for those whose work involves handling large volumes of documents

Analysts – Compiling data into a single report

Marketers – Competitor research, evaluating ad campaigns

Research – Conducting research for the compan

Content Creators – Writing articles, researching competitors

Key aspects of working with Cowork

To turn Cowork into a reliable colleague, it is important to provide it with context.

1. Brain File: my name and what I do

To begin with, you need to create a single file containing your name and your job title. In it, briefly describe your role, who your colleagues are, and what the working rules are.

1# How I work
2
3## Me
4Head of ops at a 12-person SaaS company. I value speed and plain language.
5
6## Team
7- Dana — co-founder, runs sales. CC her on anything revenue-related.
8- Raj — engineering lead. Loop him in on anything technical.
9
10## Voice
11Direct, concise. No filler words or emojis.
12
13## Rules
14- Always draft, never send emails without my approval.
15- Money/legal issues: only flag them, don’t make decisions.
16- If unsure, ask a brief clarifying question.

Claude reads this file before each task. He immediately knows who you are and how you work, so he writes reports in your style.

2. Skills: turning everyday actions into buttons with files

Skills – the Skills.md folder, which describes a useful procedure. You can also add bad examples there so that the AI does not use them.

Anthropic describes skills as follows: ‘A skill is a directory containing SKILL.md and associated files that an agent can load dynamically. It is a way of packaging your expertise for Claude.’ A skill may include instructions, resources and even executable scripts that Claude runs and receives only their output. This provides flexibility and efficiency: for example, for complex calculations, Claude simply runs a ready-made script rather than generating code on the fly.

1Build a skill that turns my raw meeting notes into:
21. Decisions (key outcomes)
32. Action items in "who — what — by when" format
43. Any open questions.
5
6Here's an example of perfect output:
7[Paste an ideal example here.]
8
9And here's an example of what NOT to do:
10[Paste a bad example here.]
11
12Save this as a skill and apply it to all my meeting notes.

3. Connectors: integration with real-world tools

Cowork runs on your computer so that it can interact with your notes, calendar, Notion and CRM – connectors are required for this. This is achieved via the Model Context Protocol (MCP) – Anthropic’s open standard for connecting AI to data and services. Using MCP, you connect external ‘servers’ (such as your Gmail or Google Calendar) and allow Cloud to read from and write to them directly.

Anthropic describes MCP as follows: “It is an open standard for connecting AI systems to data sources and tools. For example, with MCP, Claude can read emails, add events to a calendar and work with spreadsheets directly, without having to copy and paste.” Once you have connected the connector, you can make requests such as:

1Check my Notion "Tasks" database: pull all items with status "In Progress" and deadlines this week, then list them sorted by priority.

MCP supports Gmail, Google Calendar, Notion, CRM and more. You can connect any service or subscription

4. Plugins: ready-made packages for your role

You don’t necessarily have to start from scratch – there are plenty of ready-made skills, connectors and tasks for specific skills. Claude itself releases skills for marketing, design, HR and libraries.

As Anthropic states: “The plugin combines several MCPs, skills and other tools into a single download. Instead of installing them separately, you get everything at once.” But be careful: community plugins may contain third-party code or MCP servers that have not been thoroughly vetted. Anthropic recommends reading the plugin’s contents carefully before installation and trusting only verified developers.

How do I use Cowork?

To make Cowork a good employee, follow these simple rules

Start by checking the work yourself. Only assign a task to Cowork once you have checked it yourself and are satisfied with the result. First, aim to achieve a good result yourself, then turn this into a habit and set a schedule.

Just because it looks done doesn’t mean it’s done properly. Claude is very persuasive and completes the request. He might take instructions from the wrong place, send an email to the wrong address, or take data from the wrong spreadsheet. Always check the result, especially after the first few runs. Anything that can be undone (sorting folders, cleaning up, summaries) – let it do that itself, but decisions concerning money, staff or important communications – keep those to yourself.

Do not delegate decision-making. Cowork is ideal for gathering and organising information; do not delegate important strategic decisions to it. Read its reports regularly and, as Anthropic advises, keep a ‘human touch’ where it matters.

Conclusion

Cowork is the evolution from a chatbot in a notepad to an assistant that actually gets the job done. The formula for effective use

1. Run some tests to see if the task is suitable for Cowork
2. Complete the task manually once and ensure the result is perfect.
3. Save this process as a skill
4. Schedule this task at the correct intervals

Over time, you’ll free up dozens of hours a week by handing over complex routine tasks to your virtual colleague. You remain in charge: you decide what needs to be done, and Cowork carries it out. Use Brain File, skills and connectors to provide the system with context, and let Claude Cowork take care of the tedious routine aspects of your work.

If you enjoyed this article, please give it a like, leave a comment and bookmark it.

I’ve got more information about AI on telegram channel https://t.me/francescopredicts