Claude Cowork Projects Is Here (And It's Changing How I Work)

Hello legends. In this video, I'm going to take you through the new projects feature for Claude Co-work. We're going

to be going over what projects are, including what's the difference between starting a new conversation from the projects panel versus just starting a new session directly from this main panel. We'll look at how to set up a

panel. We'll look at how to set up a brand new project because you can start a new project from scratch. You can

import a project from the Claude chat mode, which is the other way to use Claude. And you can also get an existing

Claude. And you can also get an existing co-work session that has a folder and then convert that into a project. And

then finally, I'll take you through some tips and tricks of how I use projects.

Now, projects are a brand new feature for Claude Co-work, but they're not a brand new feature for Claude. We

actually had projects in the Claude chat mode. Now, you may have already used the

mode. Now, you may have already used the project feature in Claw Chat. So, you

probably already know how this works.

But just to give you a refresher, you can start a new conversation within a project. All these sessions that you

project. All these sessions that you have will basically just line up over here. So, you can very quickly and

here. So, you can very quickly and easily access any of the historical sessions. You have built-in memory. So,

sessions. You have built-in memory. So,

Claude automatically goes through each of the sessions you have within this project. It generates new memories and

project. It generates new memories and adds them into an internal file. And

then whenever you start a new session or are an existing session, you can just quickly do the clawed memory search into any existing session within this project. And then we also get custom

project. And then we also get custom instructions and then custom files for our projects as well. Now that's

essentially the basis of claude co-work projects where we're described over here. It's all about grouping related

here. It's all about grouping related work together into dedicated workspaces.

And for Claude co-work, we get the exact same things that we just saw in a chat mode. We can set custom instructions

mode. We can set custom instructions that apply to only sessions within that project. We get to do scheduled tasks.

project. We get to do scheduled tasks.

So, this is a co-work feature, but yep, scheduled tasks apply there as well. We

get context, so we get access to files on our local computer. We can also drop in external resources. Uh, Claude can also access different skills and plugins and connectors. So, everything you're

and connectors. So, everything you're used to using in Claude Co-work already applies here as well. And then you get the memory add-on as well. Actually, the

memory is the most interesting and exciting part for me because one of the issues that I have with Claude co-work is that even though you can set a new folder for a piece of work that you're working on and then every time you start

a new session from here, you can just assign it to that same folder. And I've

just opened up a demo session that I had in Claude Co-work. But the issue is that you have to create an instructions MD file and you have to instruct your claude to actually write learnings and

write memories to an M memory MD file or a learnings MD file. And this type of memory recollection is what I call hard memories. Stuff that you actually want

memories. Stuff that you actually want to store because you personally, you know, it's important to store. that

sometimes if you're working on a piece of work in a folder and you're 10 sessions deep, there's so much more nuance in the general conversation that you have, which I would call soft

memories of random bits and pieces that you haven't added to your learnings or memory MD file that is now lost in those sessions. So the fact that we get the

sessions. So the fact that we get the clawed memory function built in directly into the project which once again you're setting a folder on your co-work and every time you have a new session it

just drops directly into that folder and it persists in that folder. But the fact that we get the built-in memory function opens up two new memory possibilities for us. So the first is that we can

for us. So the first is that we can search past conversations with Claude at the individual project level. These

conversation searches are restricted to only the sessions within the project folder. So when you're searching up for

folder. So when you're searching up for memories, it's not going to get polluted with a conversation you had for a totally unrelated piece of work in a different project that you were doing.

When you're within that project and you're prompting up about a previous conversation, Claude knows to only look at conversations within that project.

And a second interesting thing that we get is that Claude's built-in memory function actually does a sweep of all the conversations every 24 hours at the project level. It then creates a

project level. It then creates a dedicated memory resource in the background that now every session and every new session will be able to access. So this is like having a

access. So this is like having a dedicated person scrape through each of the conversations you have every 24 hours and just pick out really important things and add them into the context for future conversations. So if you think

future conversations. So if you think about it, you can now set up the hard memory using the learnings.mmd or

memory.mmd files that you tell your agent to create and manage. And you also get access to the soft memory, which is the ability to rely on Claude's mechanism for remembering and

recollecting information. So to set up

recollecting information. So to set up your first project, let's go across to Claude Co-work and then click on this plus button over here. And we see three different options. I'm actually going to

different options. I'm actually going to go with option two first, which is import a project from Claude Chat. Now,

like we saw before, if you go across to Claude Chat, you have this projects folder, and we're already inside a project, which is moving to moving to the city. So all we need to do here is

the city. So all we need to do here is click on import a project. Let's drop

down and let's select move to the city.

And now we're creating a brand new folder within the claude projects workspace. And importantly our memory is

workspace. And importantly our memory is on by default. So I'm going to go ahead and create this. And now very similar to the layout that we had in the chat. We

have our custom set of instructions over here. We have any schedule tasks that we

here. We have any schedule tasks that we create within this project will appear over here. And then we have the context

over here. And then we have the context which is a collection of the files that we have on our computer, the memory from uh different conversations that we had, the actual project work that we already previously worked on. But then

additionally, if you add URLs into here, other custom context, Excel files, any work that Claude generates will all live within this project as well. Now, the

second way to create a project is to use an existing folder from previous Claude co-work sessions you've had. So most

likely you already have a bunch of project type work that you're already working on with Claude that you don't want to start from scratch. So all you need to do is click this button, click on select folder, and then find your appropriate folder. You can set your

appropriate folder. You can set your custom instructions over here. You can

add new files into here. But most likely if you add the folder, all that kind of stuff already exists. So I would just go ahead and create this. And then the last way to create a new project is completely from scratch, which is what

we're going to do for our example. So

recently I've been using projects a lot for managing client work. Every time I get a new client, I create a new project. I get custom instructions and I

project. I get custom instructions and I explain who the client is, what we're working on with the client, what the goals and objectives are. I'll actually

upload the proposal as an attached file into here as well and I do something really cool with that. But basically, I just drop in all the different information and context into a brand new project. And then as I keep working with

project. And then as I keep working with the client, so let's say I build out something inside N. I create the workflow. I delivered the first piece of

workflow. I delivered the first piece of work and then I might even add that workflow if it makes sense as context into this uh project folder and then every time I have a new session with the agent he'll be able to look at the

previous n workflows that we have or scripts that we created or apps that we're working on. If I ever have a meeting with the client and I'm using AI to transcribe and summarize that meeting I'll then maybe add that meeting into a

subfolder with within this project level as well. So actually a project is good

as well. So actually a project is good to just start a new piece of work but also to manage an ongoing piece of work.

So a lot of my projects are over six months old and the way that they shape and they hold the context is amazing. I

can literally come in, start a brand new session, ask a random nuance question about work that I did two months ago for the client and I can quickly get up to speed and complete the new work that I need to do. So for our example, I have a

new client called Apex Digital. This is

just a Claude makeup madeup client. It's

not a real one. Now, when it comes to instructions, you have the global instructions that you've set that tell Claude how to speak with you, to be short and direct, to be chirpy, to always be factual, to always, you know,

explain things logically. Whatever your

global instructions are will be will apply here as well. But when it comes to project level instructions, some stuff that you might want to put here are you explain who the client is. You might

drop the URL to their website. You might

give an explanation of what the client's products or services are and who the customers that they're targeting. You

might give an overview of your engagement with the client. So, so

they're hiring you as a consultant to build out workflows or automations or apps. So, for example, I'm going to

apps. So, for example, I'm going to start by saying, "I'm working with a new client called Apex Digital. Uh, they're

an agency that helps businesses get onto the internet by building them websites.

I'm going to be a consultant for them and help them build out automations using nan and claude code. And we're

currently at the onboarding stage and I've attached their proposal. Let me

just drop in their proposal as well.

This is a claude generated proposal that's just got a bunch of text here.

But an interesting thing that I want to spot out is that it has a timeline with a bunch of phases, deliverables, and due dates. We're going to be doing something

dates. We're going to be doing something interesting with this in just a second.

And actually, at this level, this is probably enough information for us to give. Of course, we can always come back

give. Of course, we can always come back into the instructions file, which I recommend that you do regularly, maybe after each stage that you work on, or once a week or once a month, depending on your engagement. You want to come

back into here and just make sure that you do some housekeeping. Remove any old files that shouldn't be here or update the instructions based on your new work that you're working on. Let's go ahead and create this. So, one of the first

things that I typically do when I upload a proposal and get a new client is I say, "Hey, read the proposal and let's start building a task list for all the work that we need to work on."

Previously, because I was using chat GPT or just regular claw chat mode, it was all like in an MD file or in an Excel sheet and a task list had all the different things we were working on, which is actually pretty good. But now

we're going to be trying something different because we have the ability to actually schedule tasks. And we just saw that in that proposal, we had a timeline of all the deliverables. Hey, we just started working with a new client called

Apex Digital. Can you please read the

Apex Digital. Can you please read the proposal that I attached and find the timeline and then create scheduled tasks within co-work based on that timeline?

So, this is the first client that I'm importing into co-work. I actually

haven't tried to do this based on a proposal, but I'm pretty sure it's possible. Like, schedule tasks are built

possible. Like, schedule tasks are built into here, so should be able to do this.

Okay, so looks like Co-work found the timeline and is now creating schedule tasks. Let's accept the first scheduled

tasks. Let's accept the first scheduled task. And here we have all the tasks

task. And here we have all the tasks were created. I just approved all the

were created. I just approved all the tasks and it looks like we've got 25th of March, 4th of April, 10th, 14th, and 15th. 25th, 4th, 10th, 14th of Okay, so

15th. 25th, 4th, 10th, 14th of Okay, so yeah, it's the exact same timeline and it's now actually created separate tasks for us for our main project folder. So

let's just go back one step and here we are. We have all the scheduled tasks

are. We have all the scheduled tasks that we just created in that session. We

have the kickoff, phase one, two, testing, and launch. So, this is actually the first time that I've taken uh I've literally uploaded a document and I've had someone create a plan for me. What would have been even more

me. What would have been even more interesting to do is actually issue off that command, that first prompt, but give it access to my Google calendar or my Gmail, actually my ClickUp, so that

it can create tasks automatically based on the proposal that we sent. It can

create a new folder if I had a ClickUp skill, a new folder in my ClickUp, give me notifications within here for co-work, and then actually go off and assign work to different members of my team. So, that's also possible all the

team. So, that's also possible all the regular skills and uh features that we have with Claude Co-work. Now, like we spoke about before, another thing that I do at the start of my client projects is since we have this soft memory already

built in, this is really cool. Uh but

one thing I do is hey we're working on a brand new uh working with a brand new client called Apex Digital and um I know that along the way I'm going to be learning a bunch of different stuff for the initial projects that we're doing.

There's going to be a lot of groundwork that I need to cover. And during the course of the conversation, I want you to be creating a uh learnings.mmd file

or a memory.mmd file of all the important stuff that we come across, all the preferences of the client, any new tools that they're using, any new credentials that I had to add for myself into their ecosystem. Please create a

way to actually autonomously keep that up to date session by session. So, this

is another really good prompt that I start off with. Now I want to quickly highlight that over here we have this instructions MD file and this is actually a different set of instructions to the project level instructions that

we had. Uh pretty much every layer of

we had. Uh pretty much every layer of work with claude co-work has its own level of instructions that you can apply. So these only apply for this

apply. So these only apply for this specific session. If we want to say only

specific session. If we want to say only speak to me in Polish then we can just apply that instruction for this specific chat and then as soon as we exit out of this this session into another session we'll be speaking back in English. Now,

while Claude is setting up all these files for us, I'm going to go back into my main project folder. And I'm going to come into here and actually edit my main instructions. And let's just add a new

instructions. And let's just add a new set of instructions that say, "Hey, every time you start a new session, you must check the memory MD and learnings MD file. And also keep it up to date

MD file. And also keep it up to date during the course of the conversation that we have." And it looks like I've got some work to review. Let's just

check out these details. After each Apex digital session, review the transcripts and sync learnings to memory files.

Okay, so looks like we're creating a new scheduled task that after each session is complete, which maybe it's on a time basis. I'm not sure how this is going to

basis. I'm not sure how this is going to work. It's going to go through the most

work. It's going to go through the most recent session and actually extract all the memory and learnings from it. So

let's just schedule this. Let's actually

drill back into this task and let's just check out what those files were added to. So we have the learnings MD which is

to. So we have the learnings MD which is at the uh project level for Apex Digital. So this is actually accessible.

Digital. So this is actually accessible.

It's not uh saved at the session level.

It's saved at the project level. So

every session can access this. That's

awesome. Actually, I'm not sure how well this schedule is going to work. I've

never tried this before, but I also have a um a kind of like a carryover process.

When my session gets super super long and I know it's consuming too many tokens or it's compacting too many times and I know I want to start a brand new session, but I have some ongoing work that I need to complete, I will use a prompt like this. Please go back and

review this session. I'm about to end this session and start a brand new one.

Can you please go through this conversation, highlight any important findings or memories, and then add them into our MD files? And then when you're done, can you also take a snapshot of exactly what we're up to, what we're

working on, what the existing open task is or the very next step is, and then create me a prompt that I can give to the very next session so that I can keep going from where we're leaving off. Now,

a lot of these prompts, you can actually just say like what we're doing here, just say them in the moment. Or you can go away and create a skill or a plugin for this where you just have slash commands for different parts of your conversation. So, every time you start a

conversation. So, every time you start a new client project, you might have a forward slash new client that asks you a bunch of questions, goes through a questionnaire, make sure that you upload all the context that you need, and it might also add all those different

memory files and stuff that you've learned along the way that you need.

like maybe it's a separate task list in an Excel sheet or maybe you want to use ClickUp or maybe you want to add events into your Google calendar. Another one

is to turn this into an end of session command where it goes through the entire conversation like this and just takes all the relevant memories and adds them into our hard memory files and then it also generates you a prompt for your

very next session. So my initial version of this prompt was just saying hey update all the local memory files I'm going to start a new session and then I started the new session and I always found there was a bit of gap like there's something missing there. So, I

recently started going to my end of life session and just saying, "Create me a follow-up prompt for the next piece of work that we're doing." And I've actually had so much more success and much more fluidity between conversations. And now, if we come back

conversations. And now, if we come back into our main project folder, you can see how nice and neatly everything is organized. We've got an easy way to

organized. We've got an easy way to start a new conversation, to start a new piece of work, all the existing outputs that we had from Claude, any previous sessions. We also have our custom

sessions. We also have our custom instructions. We have scheduled tasks

instructions. We have scheduled tasks that we were working on. And if we scroll down a little bit, we also have the additional context stuff that we added into here. Workflows, memory,

URLs, any local files on our computer.

So yeah, if you're new to projects and you haven't used anything like this before, I would definitely like finish this video, open up a brand new project, and just start playing with it. I

guarantee you're going to find value from this, and it's going to change how you work. All right, guys. Thank you

you work. All right, guys. Thank you

very much for watching and I'll see you in the next