Learn 80% of Claude Cowork in Under 20 Minutes

You're in the right place if you use claw chat every day, are intimidated by cloud code, and know co-work exists, but not why you need it. Because in this video, we're breaking down the seven

core capabilities of cloud co-work, each with realworld use cases, so that by the end, you know exactly what it can do and where it fits in your workflow. Quick

heads up, I've been making AI videos for about 3 years now, and I've never felt this way about a single tool. Let's get

started. Kicking things off with the three biggest differences between Claude Chat, the chatbot we're all very familiar with, and Claude Co-work, the native desktop app. First, while both

Claude Chat and Co-work require an internet connection, chat uploads files to the cloud, which limits you to 20 files per conversation and 30 megabytes per file. But since co-work can access

per file. But since co-work can access your local files, neither of those limitations apply. And so, Co-work can

limitations apply. And so, Co-work can easily handle more files with larger file sizes. On top of that, co-work has

file sizes. On top of that, co-work has a much larger context window, meaning you can do a lot more work before triggering the compacting conversation function, which is basically Claude

summarizing your previous context to free up space, but potentially losing important details in the process.

Difference number two, Claw Chat gives you a response in a chat window that you have to then copy, format, or download yourself, while Co-work actually does the work and delivers readyto-use files

directly in your folder. Spoiler alert,

the final output isn't limited to just local files since co-work can connect to external platforms and produce polished pages in notion directly. For example,

let's not get ahead of ourselves. We'll

cover that later. Third, as a result of those first two differences, we need to prompt co-work chat. With chat, we use task first

chat. With chat, we use task first language. So, we tell it what to do step

language. So, we tell it what to do step by step. For example, review my raw

by step. For example, review my raw thumbnail photos and recommend a naming convention and folder structure I should use. You get back a recommendation in

use. You get back a recommendation in text and then you go do the work yourself. With co-work, we use outcome

yourself. With co-work, we use outcome first language where we define the end result, the constraints, and the quality bar. For example, I have 15 raw

bar. For example, I have 15 raw thumbnail photos in this folder. I need

them organized into subfolders by topic with descriptive file names. You let

cowork run and 3 minutes later, the work is done. If you're struggling with this,

is done. If you're struggling with this, I've prepared a prompt template where you can just describe what you're trying to do and it spits out a prompt that's optimized for claude co-work linked down below. Now, let's go over a few

below. Now, let's go over a few essential settings so that co-work is set up properly. First, make sure the co-work tab is selected. And then under settings, the first general tab here,

this is where you'll find personal preferences that apply to both claw chat and co-work. But if you go down to the

and co-work. But if you go down to the co-work tab, you'll be able to add instructions that apply to co-work only.

I've actually put together a starter pack of instructions beginners can just copy and paste immediately out of the goodness of my heart obviously and not at all because I'm trying to build a parasocial relationship so that one day you'll buy my paid products. What? No.

No. That's just obviously a joke.

Anyway, these instructions are basically guard rails that stop co-work from touching your files without asking. Like

before deleting, overwriting, or renaming any existing file, show me what will change and wait for confirmation.

Think of these like training wheels on a bicycle, right? You can remove them as

bicycle, right? You can remove them as you get more experience with co-work.

Next, under settings capabilities, enable both memory features. I leave

location metadata off like that's going to help with data privacy. Um, set tool access to load tools when needed and enable all of these features down below.

And finally, whether you're on Windows or Mac, head on over to your documents folder and create a new subfolder called co-work playground so that everything we do today is contained in that one folder

and your other files stay untouched.

Now, when we open up a new conversation, point co-work at your playground folder by choosing the folder under documents, co-work playground. And if this is the

co-work playground. And if this is the first time you're giving access, it will ask you for access and you can click always allow. And just know that Co-work

always allow. And just know that Co-work is very strict about file access. So if

you drag a file from your downloads folder, for example, into co-work, right? Co-work will be able to load it,

right? Co-work will be able to load it, but it can't read that file. So if you want to follow along today, make sure all your files are within the playground

folder. All right. Capability number

folder. All right. Capability number

one. In plain English, co-work can create, edit, and organize files directly on your computer. Diving right

into an example, I've added a folder of over 100 receipts into the playgrounds folder that are a mix of PDFs and JPEGs.

I tell Co-work, "Look, I need an expense report from the receipt photos in my receipts folder. Uh, give me Excel

receipts folder. Uh, give me Excel spreadsheet with a date, vendor, category, amount, and a totals row. If

anything's blurry or unclear, mark it, verify." Thanks to the starter pack

verify." Thanks to the starter pack instructions, co-work first lists the steps it plans to take, waits for my confirmation, then reads every single image, extracts the relevant information, and outputs a formatted

Excel file directly in my folder. And as

you can see, there are also flagged rows for me to doublech checkck. Now, is that possible in claw chat? Not really,

because chat has a 20 file limit per conversation, and we have more than 20 receipts, obviously. And even if we had

receipts, obviously. And even if we had fewer receipts, chat would still only give us a spreadsheet in a chat window.

we then have to download and like save somewhere else ourselves. Example two, I have a massive 400 plus megabyte PDF that I need broken into smaller chunks so it's easier to upload onto AI tools.

I can just tell Co-work I have a massive PDF I need broken into separate files, one per chapter or major section with descriptive file names so I can find

what I need at a glance. and co-work

unlike chat will identify natural section breaks, create all those individual files and then save them all back in your folder. Example three,

Notebook LM can create amazing PowerPoint presentations, but the slides are images that can't be edited. So, we

share the file with co-work and say, "These slides aren't editable. Rebuild

this as a clean editable PowerPoint.

Same content, same slide order." and

co-work will first read all the slides to understand the underlying content before recreating the entire presentation with real text boxes you can actually modify. Although it's not

perfect, as you can see here, it's better than just having static images.

Persistent memory is the most important capability we're covering today, and it's only possible thanks to capability number one, co-working local file access. Let me explain with a silly

access. Let me explain with a silly example. If I go into claw chat, open

example. If I go into claw chat, open personal preferences and add always address me as Heisenberg. At the start of every conversation, I start a new chat and Claude will always open with,

"Hey Heisenberg say my name, Heisenberg. You're godamn

right."

That's technically persistent memory.

You tell the AI something and the AI remembers that thing in future conversations without you having to repeat yourself. But because Claude

repeat yourself. But because Claude chat's memory is stored online, there's a hard limit to how much it can hold.

Co-work saves memory to actual files on your computer. Meaning it can remember

your computer. Meaning it can remember every decision you've asked it to hold, every preference you've set for as long as you need. For example, if I ask Coowork, how many newsletter editions have we produced together so far? Break

it down by which app each edition covered. By the way, you should

covered. By the way, you should definitely sign up for my weekly newsletter if you haven't already. It'll

load up the relevant context. Tell me

we've produced seven editions together.

two for Gmail, two for Chrome, and so on and so forth. And remember, this was from a brand new session with no prior conversation. Here's a use case you can

conversation. Here's a use case you can try yourself. I have a meeting

try yourself. I have a meeting transcript within a subfolder in co-work playground and I say, "Summarize the meeting transcript limit to 200 words."

And after it gives me that first output, let's say I don't like the structure of this and I move the context up top, for example. Not that that's something I'd

example. Not that that's something I'd ever do because context always comes after. But I tell co-work, I've made

after. But I tell co-work, I've made some changes to your summary. Compare

your version with mine and save those preferences in a visible root cla MD and memory MD files so you remember them next time. Co-work then reconciles the

next time. Co-work then reconciles the two versions, figures out what I changed, and most importantly, and I can't stress this enough, most importantly, it creates something called

a CLAM MD and memory MD file. First, you

can see that co-work created these two files under your playground's root folder. How these work is more advanced

folder. How these work is more advanced than what we'll cover today. So, I'll

save that for the next video. For now,

just know whenever you tell Coowork to remember something, it writes to those files. And the more it writes, the

files. And the more it writes, the better Co-works at working the way you want it to. So, keep those two files there. By the way, I recently revamped

there. By the way, I recently revamped my free AI toolkit from scratch. New

tools, new workflows, new everything.

So, if you signed up for the original, this is a completely different experience. Link down below. Next up,

experience. Link down below. Next up,

connecting co-work to tools. By default,

co-work can only see what's in your folder. Connectors let it reach into the

folder. Connectors let it reach into the tools you already use like Gmail, Google Drive, or Notion, right? So, it can read from them and work directly inside them.

To set this up, click the customize tab on the left, go to connectors, click the plus icon, then browse connectors, and you can search for the tool that you use. At the very minimum to follow along

use. At the very minimum to follow along today, I recommend you connect Gmail, Google calendar, uh Google Drive, and notion if you use it. Once you've

connected Gmail, you can tell co-work something like, "I want you to understand my tone of voice. Go through

my emails from the past month to extract my tone of voice, and save those as writing style principles you can follow going forward." Co-work then proceeds to

going forward." Co-work then proceeds to read through my emails, locates those with substantive writing, distills the patterns it sees and presents them to me and then saves them as writing style

principles within the memory MD file so that next time I when I draft an email, it remembers my tone of voice. Next

example, although my team and I take meeting notes in Notion, we use Google Gemini to autogenerate transcripts that are stored in Google Drive. So, if I were an anal, high maintenance,

unreasonable manager, which I'm not, I tell co-work, hey, check the Gemini transcript against the meeting notes in Notion and surface commitments that didn't make it into the notes. And

co-work will read both documents, compare what was said versus what was written, and surface the action items, the commitments that didn't make it into

the notes page. This is an example of two connectors working together via co-work. It pulls a transcript from

co-work. It pulls a transcript from Google Drive, pulls the meeting notes from Notion, cross references them, and surfaces what fell through the cracks.

By the way, if you can't find the tool you want under connectors, you can add a custom connector through something called an MCP, which we're not covering today, but just know that option exists.

All right, capability number four, skills. Let's dive right into a simple

skills. Let's dive right into a simple example. I just asked co-work to write

example. I just asked co-work to write something for me, and I want to clean up the output. So, I run it through a

the output. So, I run it through a prompt template I use all the time that makes any text more clear and concise.

Co-work gives me a version two of the output. I like it. And now I can tell

output. I like it. And now I can tell Co-work, "Hey, turn what you just did into a clear and concise skill." You can see that co-work first runs its own

skill creator skill for this task. Then

it asks me a few clarifying questions.

Uh, this skill should apply to any text.

It should rewrite and provide the change log so I can see what changed. And yes,

let's skip the test cases. And co-work

proceeds to keep going. After receiving

permission, co-work creates the skill.

And I can now just add the skill to my co-work with just one click. And

clicking the manage pop-up on the top right lets me see all the skills I currently have access to. Now, the next time Claude gives me something and I say make it more clear and concise, co-work

simply runs its last output through the skill we just created. So, that was just a simple one-step skill, which honestly isn't that impressive, right? But the

underlying concept is exactly the same for a weekly report that might take 10 steps from start to finish. You teach

co-work to do, how many steps there are, and it repeats that every single time, which is a massive timesaver over the long run. Here's a more practical

long run. Here's a more practical example. When I was at Google, different

example. When I was at Google, different teams would send me their weekly updates in completely different formats, and I had to combine everything into one clean update for leadership. So I share the raw updates from the three teams with

co-work and say combine these three team updates into one formatted weekly marketing update I can send to leadership. After a few minutes, co-work

leadership. After a few minutes, co-work produces a first draft. And although it looks great, it's way too dense. So I

give feedback lead with uh three topline metrics across all teams. Then give me three highlights and three lowlights.

Keep it within 300 words. Co-work

revises but gives me the output in chat.

So just for fun, I ask for a PDF format.

And after a few more minutes, co-work gives me the report in PDF and the structure is exactly what I wanted. So I

can finally tell co-work now go back through our conversation and create a weekly report skill that captures this entire workflow. After asking me a few

entire workflow. After asking me a few clarifying questions, co-work summarizes my request, asks for permissions again, then spends a few minutes creating the skill and running tests with

hypothetical scenarios. Once that's all

hypothetical scenarios. Once that's all done, co-work tells me I need to install the skill manually, almost like I planned for this to happen. So, I head

on over to customize skills plus icon, upload a skill, find the skill MD file that co-work generated, and boom, we now have a weekly report skill. Three more

things you need to know about skills.

First, to create your own skill, head on over to customize skills and make sure the skill creator bythropic skill is

enabled. And while there are websites

enabled. And while there are websites that offer skill templates, I highly recommend creating the first few yourself to get a hang of creating skills from scratch. Second, you can

update a skill anytime by telling co-work which skill to change, what to change about that skill, and granting permission for the change. But after it recreates it, you need to click copy

again to overwrite the previous version.

I'd also recommend backing up your skills to Google Drive because if you move to a new computer, skills do not transfer over. Third, under the

transfer over. Third, under the customize option, skills tab, you'll see if you click plus, there's an option to create a skill with Claude. I do not recommend doing this because it's much

more effective for you to actually just go through the actual workflow first, go through that back and forth process, then reverse engineer the skill at the end like I showed you earlier today.

Moving on, at the risk of massively oversimplifying, co-work projects and claw chat projects are basically the same thing, except co-work projects come with all the capabilities we've already covered. Local file access, persistent

covered. Local file access, persistent memory, connectors, skills, all of it.

But there is one thing worth calling out. Co-work projects can write to their

out. Co-work projects can write to their knowledge files directly, unlike claw chat projects. Here's what I mean. I

chat projects. Here's what I mean. I

have a claw chat project called clarity partner that helps me make my writing more articulate. After a session where

more articulate. After a session where we go back and forth on a piece of writing, Claude learns something new about my style and I want to save that as a permanent principle so it remembers next time. In a chat project, after I

next time. In a chat project, after I ask Claude to codify our learnings, Claude is able to generate an updated knowledge file for me. And while I am able to add this to the project as a

knowledge file in one click, I still have to manually delete the old file from the project settings. In co-work, I can just say something like codify this principle under the clarity partner

project. Like when I'm explaining a

project. Like when I'm explaining a concept, always start with an example first. And co-work writes it directly to

first. And co-work writes it directly to the instruction file without me having to do anything. If you're watching this and thinking, I want a system like this for my own work. I'm actually building a course that walks you through exactly

that step by step. I'll leave a link to the weight list down below. So,

ironically, this next capability also highlights one of co-work's biggest weaknesses. The capability itself is

weaknesses. The capability itself is simple. If you have the cloud extension

simple. If you have the cloud extension installed on your browser, co-work can theoretically hand off tasks to the browser extension. I say theoretically

browser extension. I say theoretically because I simply can't recommend the extension right now for three practical reasons. First, it's slow as since most

reasons. First, it's slow as since most interactions require a screenshot sent back to co-work before it decides what to do next. Second, it's unreliable.

There were many times where it just stopped halfway through a task without finishing. Talk about being blueballed

finishing. Talk about being blueballed by Claude. Third, it burns through your

by Claude. Third, it burns through your usage because it overthinks every step since it has to be super careful when controlling your browser. Coming back to the weakness I mentioned earlier, while

both claw chat and co-work can search the web, claw chat gives us a lot more control. We can force web search enable

control. We can force web search enable or disable. And we can even enable or

or disable. And we can even enable or disable this mid-con conversation. With

co-work, you can't force a web search and it often falls back on the browser extension. And given how unreliable the

extension. And given how unreliable the extension is right now, that's a real limitation. To be clear, scheduled tasks

limitation. To be clear, scheduled tasks aren't new. Gemini and Chachup T have

aren't new. Gemini and Chachup T have had this feature for a while now, but I found that only Claude Co-works schedule tasks work flawlessly, and it's because of the capabilities we've just covered.

Diving into an example, let's open up scheduled tasks and click into this morning inbox triage task that runs every day at 6 a.m. It produces a report

and draft replies for all my emails.

Let's actually just open up an example to show you. And because of how I set this up, the report and the drafted replies are basically flawless. So, let

me walk you through how this works.

First, longtime viewers know I teach my inbox workflow in the Workspace Academy course. And Co-work took that inbox

course. And Co-work took that inbox workflow and distilled a set of rules to follow. And thanks to capability number

follow. And thanks to capability number one, local file access, it saved those rules in a markdown document. As you can see here, inbox your triage workflow.

And these are the instructions. Second,

since co-work is connected to Gmail through the connectors feature, capability number three, it's able to read all my emails, right? map them

against my inbox zero rules and what it knows about me to draft contextually relevant replies. But this did not work

relevant replies. But this did not work so well at the beginning. For the first week or so, I had to give feedback to co-work on how I would have rephrased certain emails. And thanks to persistent

certain emails. And thanks to persistent memory capability number two, co-work remembers those corrections and now triages my inbox the way I would. And

this does require some setup. In my

actual co-work folder, I have an email HQ subfolder. And here you can see

HQ subfolder. And here you can see within the cloud MD file, I have my inbox zero triage workflow, right? And

within the memory MD file right over here, it remembers the feedback I've given it. For example, how I like my how

given it. For example, how I like my how I like to sign off in my email signature, right? So, this represents

signature, right? So, this represents weeks of feedback and context and basically learning, so to speak. I know

that looked intimidating, but don't worry, I'm going to dive much deeper in future videos. So, see you soon. And in

future videos. So, see you soon. And in

the meantime, have a great one.