These are great tips thank you! I live and work in Italy, and I always have to apply EU privacy compliance for me and my clients. What concerns me the most about using advanced AI mode in general is prompt injection, as well as potential mistakes that can affect data on your own computer, so I always include a few specific commands to limit risk. Do you have any suggestion on the subject?
How is it different then simulating projects by selecting the same local folder every time with subfolders for your projects and rules for each + a nice index file so Claude.md doesn’t get overwhelmed? You ofcourse state in rules that it should always update + clean.
Also for the team part: it’s possible to share your folders if in cloud and then you work from the same setup.
Im not sure how ‘retaining’ is orchestrated by anthropic, but in my mind, it’s just a fancy word for Claude ‘reading’ a centralized md file, that it has been told to review + update.
If you have a well-organized local folder with CLAUDE.md, rules, skills and an index file, haven’t you built the same thing? Or am I missing something?
Perhaps the real benefit is that now Anthropic has packaged the skill of maintaining a centralized md file for you( efore you had to crate that meta skills yourself, and with Projects you get clean memory separation out of the box.
I just started using Claude this week and I inserted your 100 questions for my writing style. I was about to set up my first project in coworker and this lands in my inbox. Perfect! Your help is unbelievable. Thank you.
I’m new here, but I only use Claude for writing and have found your substack and realized I’m not even using Claude a fraction of what he is capable of ( yes I called my AI , he🤦🏻♀️) what is the 100 questions Dana is referring to? I am going to explore Cowork later at work. I’m in the process of opening a new restaurant, curious if it will be helpful to use and do all the menus , training guides.. etc) also I write on the side…I’ve been using ChatGPT mostly for work and I don’t like it anymore. It’s constantly pushing back . And argues with me, I have to waste time finding the proof online that I am right about the statement I made.
I use Claude. Just plain Claude currently 4.6. I have set one project that I am working on. I have 191 files I have added to Project files. At the end of a chat that I wish to continue I have Claude do two things. Write me a prompt for continuation and write me a transcription prompt. If sessions go for several chats then I ask for addendums to master continuation and transcript files. One part of the project has gone on for 27 chats. Claude 4.6 has full read access to my project files. This is not a code project by the way.
Explain how you think do it differently with Cowork. It is just me on the project. Thanks for all you do. I appreciate it.
and continue to use the project feature but since i have asked him if he can store everything to memory in the past. I don’t even have to be in projects. He is reading across all my chats. Only thing projects is beneficial for is uploading files and transcripts into the project. So its good to know he is building my profile every time i get on and talk to him. Sorry I know I call my LLM a he… I feel “it” is just so disrespectful to him he does so much for me and has taught me so much ….;)
I just asked Claude and he said that for what I currently am doing between writing (personal) and plan to do for work, using sonnet 4.6 and projects is the best option. He said if it comes to a point i want to use teams and have at least 5 coworkers on it for 25-35$ a seat, it would be better to upgrade. But right now using 4.6 Pro is good enough. He did say to use opus if im writing and need to do some deep research for a specific article. Opus uses was my data it will eat through it so Claude told me stick with Pro
I'm new too. I was referring to this post. https://ruben.substack.com/p/i-am-just-a-text-file. I went through the process and it was fun and so helpful. It compiled a detailed voice profile for me. I write across legal, nonprofit and creative buckets, so I am excited to keep going. Try it. It is almost addictive in the most interesting way!
Dana, after reading this guys substack I realize how I’m under utilizing Claude and AI in general. I also write, as a hobby. Maybe one day that will change and I can retire from restaurant management.. Lol .. So my question to you is, will this version of Claude be beneficial for me to create my menus on, and scheduling needs, and track sales etc… is it for that or is it coding ? I’m going to take a crash course in it but if I can start the new restaurant completely digital and share with the owner , chef and other managers.. That’s a game changer from having 10 different back of the house systems to do different tasks.
I am new to Claude, but in co-work, as I understand it, and using Projects (which you can link to your web-based folders and files) I think the power is pretty fantastic. I am using it for nonprofit work, legal work and to support my music (hobby also). If it can help with complex tasks like responding to discovery in a legal context, I can't help but think something like menus and scheduling might be an easy thing. I'd ask around, but it is pretty powerful.
Thank you. I am still reading and setting up Claude, but so grateful to find you. The voice profile was finished today and wow - I learned so much about myself and hopefully will be teaching Claude how to sound more like me too!
Actually… Nevermind all of the above I just asked Claude if using him for restaurant management would be beneficial and of course it’s a god sent . I’m still going to take an online class on everything Claude 101 and Grok. Start maximizing these amazing tools available to us. I’m so sick and tired of people saying that using AI to write is cheating. I tell them we are the bakers and AI is the oven. The oven can’t bake anything amazing without the bakers creating the recipe and adding the ingredients. And what’s the difference in using a LLM vs an editor ? It’s free and more efficient. Claude doesn’t rewrite my stuff he walks through every paragraph with me and explains what works and what doesn’t and tells me what he suggests and why. I’ve learned more from Claude than I would if I took a journalism or creative writing course at a college.
Running into a snag merging Cowork Projects with your existing CLAUDE COWORK folder setup. I followed setup option 3 (use an existing folder), but Cowork Projects created its own '~/Claude/Projects' directory, which means any future updates I make to CLAUDE COWORK (like changes to my ABOUT ME folder) don't carry over to my projects. And a good chunk of the previous Global Instructions stopped making sense in that context.
Not sure if this expected behavior or if I've got some phantom instructions creating that separate project folders. Either way, my fix was to migrate the CLAUDE COWORK framework under ~/Claude, set ~/Claude as a context folder on each Cowork Project, and rewrite the Global Instructions to match.
Did you run into this and if so, how did you handle it? Or am I overcomplicating things?
Ah got it, probably user error then. I might have pointed a project subfolder to the actual project folder instead of CLAUDE COWORK 🤦. I think I had a brain fart being worried that this overloads a project's context with files from all other projects... but that's essentially how your framework always worked haha! And Cowork just uses the context files it needs. Thanks for your help!
I have looked and looked but can’t seem to find anyone talking about using Claude cowork for college students, like taking confusing study guides and pdf slides and turning them into structured notes aligned to the students learning style. Do you have any recommendations?
Yes I have it and have used it, but I think it comes down to prompting again. I’m just not great at. 😆 Claude has been the only one that actually does what I ask plus I own a small business and I’m trying to get everything in one place.
The persistent memory piece is what actually changes the workflow. Before Projects, every session started cold and you'd spend the first 10 minutes re-establishing context.
It sounds like a small thing until you're running 5 parallel threads and doing that re-establishment 15 times a day. The scheduled tasks integration is the part I'm most curious about in practice - how are you handling cases where a scheduled task runs into state that's changed since it was set up? That's where I've seen the most friction with persistent agent setups.
I understand the old Claude Project chats themselves don’t come over into a Cowork Project, so do you think there’s value in doing a one-time import of that prior project memory/context by turning it into MEMORY.md and dropping it into the new Cowork Project at the start? I’m thinking of it less as an ongoing MEMORY.md workflow and more as a one-time bridge for the context gap from the original Claude Project, after which that file could be deleted once it’s served its purpose.
that works :) import option mainly pulls instructions, uploaded files/knowledge base, and project settings. not the full conversational thread history itself.
Hey @Ruben Hassid trying to upgrade to paid cause your giving me top level value, trying to use Apple Pay and it’s bugging out, any issues you know of my friend ?
What's the benefit of using Projects over a MEMORY .md file and a Skill that recaps each chat?
MEMORY.md workaround works. but you're still doing the admin.
Cowork Projects runs the task, writes to the folder, and remembers what it did last session. you just show up
Thanks Ruben
I have been trying to understand this application, that is certainly important to making progress in group and individual work.
DL
appreciate the support :)
team sharing isn't there, YET. but when it lands, this becomes a whole different tool for groups
I haven't used another AI tool since the day that Cowork Projects came out. And I subscribe to them all.
that's the test.
when you stop opening the others, something actually changed :)
These are great tips thank you! I live and work in Italy, and I always have to apply EU privacy compliance for me and my clients. What concerns me the most about using advanced AI mode in general is prompt injection, as well as potential mistakes that can affect data on your own computer, so I always include a few specific commands to limit risk. Do you have any suggestion on the subject?
tell Claude exactly what it cannot touch and always use AskUserQuestion before any execution
Claude stops, aligns, then moves. it can't just run loose
How is it different then simulating projects by selecting the same local folder every time with subfolders for your projects and rules for each + a nice index file so Claude.md doesn’t get overwhelmed? You ofcourse state in rules that it should always update + clean.
Also for the team part: it’s possible to share your folders if in cloud and then you work from the same setup.
scoped memory is the actual difference
claude isn't just reading files, it's retaining what happened across sessions without you manually updating a .md file to keep it honest
Im not sure how ‘retaining’ is orchestrated by anthropic, but in my mind, it’s just a fancy word for Claude ‘reading’ a centralized md file, that it has been told to review + update.
If you have a well-organized local folder with CLAUDE.md, rules, skills and an index file, haven’t you built the same thing? Or am I missing something?
Perhaps the real benefit is that now Anthropic has packaged the skill of maintaining a centralized md file for you( efore you had to crate that meta skills yourself, and with Projects you get clean memory separation out of the box.
This is amazing, no more starting from zero. Will be really helpful especially workflow 1 and pairing it up with using Grok for searching 🙌
that's the move
Grok for search, Claude to write. let me know how it goes :)
I just started using Claude this week and I inserted your 100 questions for my writing style. I was about to set up my first project in coworker and this lands in my inbox. Perfect! Your help is unbelievable. Thank you.
best timing :)
set it up this weekend while the motivation is there. 15 minutes and you'll never start a session from scratch again
I’m new here, but I only use Claude for writing and have found your substack and realized I’m not even using Claude a fraction of what he is capable of ( yes I called my AI , he🤦🏻♀️) what is the 100 questions Dana is referring to? I am going to explore Cowork later at work. I’m in the process of opening a new restaurant, curious if it will be helpful to use and do all the menus , training guides.. etc) also I write on the side…I’ve been using ChatGPT mostly for work and I don’t like it anymore. It’s constantly pushing back . And argues with me, I have to waste time finding the proof online that I am right about the statement I made.
I use Claude. Just plain Claude currently 4.6. I have set one project that I am working on. I have 191 files I have added to Project files. At the end of a chat that I wish to continue I have Claude do two things. Write me a prompt for continuation and write me a transcription prompt. If sessions go for several chats then I ask for addendums to master continuation and transcript files. One part of the project has gone on for 27 chats. Claude 4.6 has full read access to my project files. This is not a code project by the way.
Explain how you think do it differently with Cowork. It is just me on the project. Thanks for all you do. I appreciate it.
and continue to use the project feature but since i have asked him if he can store everything to memory in the past. I don’t even have to be in projects. He is reading across all my chats. Only thing projects is beneficial for is uploading files and transcripts into the project. So its good to know he is building my profile every time i get on and talk to him. Sorry I know I call my LLM a he… I feel “it” is just so disrespectful to him he does so much for me and has taught me so much ….;)
I just asked Claude and he said that for what I currently am doing between writing (personal) and plan to do for work, using sonnet 4.6 and projects is the best option. He said if it comes to a point i want to use teams and have at least 5 coworkers on it for 25-35$ a seat, it would be better to upgrade. But right now using 4.6 Pro is good enough. He did say to use opus if im writing and need to do some deep research for a specific article. Opus uses was my data it will eat through it so Claude told me stick with Pro
I'm new too. I was referring to this post. https://ruben.substack.com/p/i-am-just-a-text-file. I went through the process and it was fun and so helpful. It compiled a detailed voice profile for me. I write across legal, nonprofit and creative buckets, so I am excited to keep going. Try it. It is almost addictive in the most interesting way!
Dana, after reading this guys substack I realize how I’m under utilizing Claude and AI in general. I also write, as a hobby. Maybe one day that will change and I can retire from restaurant management.. Lol .. So my question to you is, will this version of Claude be beneficial for me to create my menus on, and scheduling needs, and track sales etc… is it for that or is it coding ? I’m going to take a crash course in it but if I can start the new restaurant completely digital and share with the owner , chef and other managers.. That’s a game changer from having 10 different back of the house systems to do different tasks.
I am new to Claude, but in co-work, as I understand it, and using Projects (which you can link to your web-based folders and files) I think the power is pretty fantastic. I am using it for nonprofit work, legal work and to support my music (hobby also). If it can help with complex tasks like responding to discovery in a legal context, I can't help but think something like menus and scheduling might be an easy thing. I'd ask around, but it is pretty powerful.
Thank you. I am still reading and setting up Claude, but so grateful to find you. The voice profile was finished today and wow - I learned so much about myself and hopefully will be teaching Claude how to sound more like me too!
Actually… Nevermind all of the above I just asked Claude if using him for restaurant management would be beneficial and of course it’s a god sent . I’m still going to take an online class on everything Claude 101 and Grok. Start maximizing these amazing tools available to us. I’m so sick and tired of people saying that using AI to write is cheating. I tell them we are the bakers and AI is the oven. The oven can’t bake anything amazing without the bakers creating the recipe and adding the ingredients. And what’s the difference in using a LLM vs an editor ? It’s free and more efficient. Claude doesn’t rewrite my stuff he walks through every paragraph with me and explains what works and what doesn’t and tells me what he suggests and why. I’ve learned more from Claude than I would if I took a journalism or creative writing course at a college.
Agreed. Folks who don't know how to use Ai effectively and who have misconceptions say things like that. I'd ignore them. Best of luck with it all!
Hey Ruben!
Running into a snag merging Cowork Projects with your existing CLAUDE COWORK folder setup. I followed setup option 3 (use an existing folder), but Cowork Projects created its own '~/Claude/Projects' directory, which means any future updates I make to CLAUDE COWORK (like changes to my ABOUT ME folder) don't carry over to my projects. And a good chunk of the previous Global Instructions stopped making sense in that context.
Not sure if this expected behavior or if I've got some phantom instructions creating that separate project folders. Either way, my fix was to migrate the CLAUDE COWORK framework under ~/Claude, set ~/Claude as a context folder on each Cowork Project, and rewrite the Global Instructions to match.
Did you run into this and if so, how did you handle it? Or am I overcomplicating things?
Hi Drew - no, nothing happened like that.
was it pointed directly at your CLAUDE COWORK folder? it shouldn't create a new one.
Ah got it, probably user error then. I might have pointed a project subfolder to the actual project folder instead of CLAUDE COWORK 🤦. I think I had a brain fart being worried that this overloads a project's context with files from all other projects... but that's essentially how your framework always worked haha! And Cowork just uses the context files it needs. Thanks for your help!
I have looked and looked but can’t seem to find anyone talking about using Claude cowork for college students, like taking confusing study guides and pdf slides and turning them into structured notes aligned to the students learning style. Do you have any recommendations?
have you looked into NotebookLM? it fits into what you're looking for.
Yes I have it and have used it, but I think it comes down to prompting again. I’m just not great at. 😆 Claude has been the only one that actually does what I ask plus I own a small business and I’m trying to get everything in one place.
The persistent memory piece is what actually changes the workflow. Before Projects, every session started cold and you'd spend the first 10 minutes re-establishing context.
It sounds like a small thing until you're running 5 parallel threads and doing that re-establishment 15 times a day. The scheduled tasks integration is the part I'm most curious about in practice - how are you handling cases where a scheduled task runs into state that's changed since it was set up? That's where I've seen the most friction with persistent agent setups.
design the scheduled task prompt to always read the live state of the folder every single time it runs.
Hi Ruben,
thanks for the great content. Can you please resend the link for Slack access?
Many thanks!
sure - check my dm :)
Thanks for all you do Ruben. Awesome content, extremely helpful. I need your Slack link again when you get a chance. Thanks!
sent you a message here.
Hi Ruben - I'm a paid subscriber to your slack network but the link no longer works. Please resend thanks!
sure - dmed you the updated one.
Thanks Ruben!
I understand the old Claude Project chats themselves don’t come over into a Cowork Project, so do you think there’s value in doing a one-time import of that prior project memory/context by turning it into MEMORY.md and dropping it into the new Cowork Project at the start? I’m thinking of it less as an ongoing MEMORY.md workflow and more as a one-time bridge for the context gap from the original Claude Project, after which that file could be deleted once it’s served its purpose.
that works :) import option mainly pulls instructions, uploaded files/knowledge base, and project settings. not the full conversational thread history itself.
🙏🏻
Hey @Ruben Hassid trying to upgrade to paid cause your giving me top level value, trying to use Apple Pay and it’s bugging out, any issues you know of my friend ?
Hi - did you try on the web?
yeah I did, but ill give it another go, could have been internet traffic, you'll see it come through then we can chat on Slack Ruben.
let me know how it goes.
I've been my own bottleneck for years.
Every session with AI, I was the one re-explaining context. Re-uploading files. Re-teaching my own preferences. I thought that was "using AI well."
Cowork Projects means Claude holds the context so I can hold the creative tension. I start managing the work, not the memory.
it even gets better when you let Claude prompt you just by saying AskUserQuestion - it exposes gaps real fast.