You’ve done setup, turned on memory, and written global instructions. Time to actually run some sessions. This module walks through three real tasks end-to-end — small, medium, and large — so you can see what a Cowork session actually looks like, what to type, and where to stop.
The golden rule: start by picking a folder. Cowork always operates inside a single folder and everything below it. Pick that folder deliberately — it is the sandbox for the session.
Session 1: Clean up your Downloads folder
A great first session because it is genuinely useful and nothing is at stake.
Point Cowork at Downloads. Click the folder picker, select your Downloads folder (or whichever folder has accumulated mess). If this is the first time Cowork has touched that folder, it will ask for permission. Grant it.
Pick your model. Opus 4.7 with extended thinking if you’re new. Sonnet if you’re on the $20 plan and watching tokens.
Prompt:
This is my Downloads folder. I want you to:
1. Put all the screenshots in one folder.
2. Put all the contracts in another folder.
3. With the remaining files, organize them into
three folders — High, Medium, Low — based on
your judgment of how important they look.
Don't delete anything.
What happens. Cowork lists the files, reads enough of each one to classify it, builds a to-do list in the sidebar, and starts moving files. You can watch it work. If you change your mind mid-flight — say, you want a “Medium” bucket — just type it into the chat. Cowork adds it to the to-do list and adjusts.
What to verify. When it’s done, open Finder. Check that nothing was deleted. Spot-check a few files in each bucket to see whether the classification matches your instinct. If “High” is wrong, tell Cowork what it missed; it will reshuffle.
This one session captures the whole Cowork paradigm: point at a folder, give it a task, let it work, approve or redirect. You stay in the driver’s seat.
Session 2: Build a briefing document from a pile of PDFs
A step up. This one produces a real deliverable.
Pick a folder full of source material. In this walkthrough, imagine a folder called global-outlooks containing a dozen investment bank outlook PDFs — a synthesis task a junior analyst would spend a day on.
Pick your model. Opus 4.7 with extended thinking. This task involves reading many PDFs and producing a well-formatted output, which is exactly what extended thinking is for.
Prompt:
Go through all of the investment outlooks in this
folder. They are forecasts for the current year.
Create a document for my boss, who is a portfolio
manager and doesn't have time to read all of them.
My boss wants:
- A synthesis of the clear themes for the upcoming year
- An analysis of the key risks
- A table at the bottom showing forecast returns
by asset class, attributed to each bank
Create the document in Word. Make it well structured
with a little color to give it some personality.
What happens. Cowork lays out a plan, reads each PDF in turn, and produces a Word file in the same folder. Expect this to take several minutes — PDFs are token-intensive, which is why a cleaner pipeline would convert them to text first. For the first time, just let it run.
What to verify. Open the Word document. It should have headers, a themes section, a risks section, and the table. Check a couple of attributions against the source PDFs — is JP Morgan’s forecast actually what the report says? This is where you, the human, earn your keep. AI-generated synthesis is a strong first draft, not a finished product.
Bonus step. Once you have the Word doc, try:
Adapt this document for PowerPoint. Create a new file.
Condense it to no more than seven slides, including the title.
Cowork generates a .pptx in the same folder. Open it. Notice that it didn’t just dump bullets — the themes became cards, the forecasts became a table, risks got their own slide. This is the part that feels like magic the first time. It is also the part where you will find small design things you want to change. Fix them in PowerPoint directly; the file is yours.
Session 3: A real project with accumulated context
Now the big one. This session uses projects to give Cowork persistent context about a body of work.
Set up the project. In Cowork, click Projects → New project → Use an existing folder. Point at a folder on your machine — say, a folder called axe-capital containing two or three old due-diligence questionnaires (DDQs) from past clients.
Name and instructions. Name the project by outcome: Update Axe Capital DDQs. Add short custom instructions:
This project helps me respond to due-diligence
questionnaires for a hedge fund. Use a professional
tone. Assume the reader has intermediate understanding
of investment concepts, so don't over-explain.
Dollar figures in thousands, no decimals.
Start a chat in the project. Try a low-stakes question first:
Describe the organizational structure of the firm.
Cowork uses the old DDQs as context and produces an answer grounded in those files — legal entity, ownership, senior leadership, and so on. Verify one or two facts against the source docs. This is the sanity check that your project is wired correctly.
Do the real task. Now the big prompt:
There is a blank DDQ in this folder called
"hedge-fund-DDQ-blank.docx". Use the information
from the old DDQs in this project to fill in the
new one to the best of your ability.
Do it in a Word document. For every answer, cite
which source document the information came from,
and note your confidence level (high, medium, low).
What happens. Cowork walks the blank DDQ section by section, pulling answers from the context, and writes a filled Word doc with citations and confidence tags. Expect several minutes of runtime and a 10–15 page output.
What to verify. This is a first draft, not a finished DDQ. Walk every section. High-confidence answers are usually right; medium and low are where you’ll find hallucinations, stale data, or missed nuance. AI produces a draft in minutes that used to take days — but a human still has to ship it.
Common mistakes in your first few sessions
- Not picking the folder deliberately. Cowork will happily work in whatever folder is open. Pick a scoped folder so it can’t accidentally touch something you care about.
- Starting too abstract. “What can you do?” gets vague answers. “Clean up my Downloads folder” gets work.
- Panic-cancelling the first long task. Real Cowork sessions take 5–15 minutes. That is normal. Keep an eye on the to-do list in the sidebar to see that it’s making progress.
- Skipping verification. Confidence tags and citations exist precisely because the output needs human review. Use them.
- Using chat instead of Cowork. If you catch yourself uploading files and downloading outputs one at a time, you’re in the wrong tab.
What you’ve learned
After these three sessions you’ve seen: file operations in a scoped folder, multi-source synthesis with a Word and PowerPoint deliverable, and a project with persistent context used to complete a real document. Every later module — skills, scheduled tasks, Excel workflows — is a variation on this base loop. Point, prompt, approve, verify.
Next up
Next module: artifacts. Interactive mini-apps Cowork can generate — chatbots, data visualizations, even small games — and share with colleagues who never need to learn AI themselves.