Thanks - I appreciate the comment. It was weird because I used it for a month, then watched a bunch of YT videos before I prepared my live session (which was yesterday) and I still didn't feel that on top of it - had to really tease out some of this info from Claude and Perplexity
This is one of the most useful pieces I have seen on Claude Cowork because it is not just excited about the tool, it is honest about the limits too. The practical lessons and the warning about risk make this much more valuable than a typical AI how to post.
The CLAUDE.md point is the one that changed how I work with agents. A well-stocked folder beats a clever prompt every time. I built this into our Whimsey Labs workflow: every agent session starts with a spec file that defines the arena. The agent does not need to remember anything. It just needs to read.
This is a good step, but Cowork is Mickey Mouse compared to Claude Code. And then you learn that claude.md and memory and feedback are not enough. Claude is fairly crippled for any serious work. I’ve spent probably 200 hours building solution for it. Welcome to the rabbit hole.
How do you use Code for tasks Cowork could do? I used Code to build an app or something that performs a function, yes, but not a function that can already be carried out in the existing desktop setup, know what I mean? I like "invented" a function with it, where I see Cowork as for carrying out exist functions. I'm not a pro at either (master of the Chat, here), so would love to learn a little from your POV/use cases!
If it’s one‑off and lives across your desktop/apps, use Cowork.
If it’s repeating, and you want a “magic button” you can reuse forever, I use Code once to build that button, then just click it (run it) going forward. Or cron it so it runs daily weekly whatever.
To me the diff is do you want to be a machine operator or build a system that can operate with or for you.
You don’t have to be “a programmer” to do this. With Claude Code, you mostly describe what you want in plain language, let it propose a plan, and then you just run the tool it built for you.
Thank you for this! I've been seeing so much about Claude Cowork on socials lately, but it's also been very clear that many of these posts are from people who are being compensated through Anthropic's referral program. I appreciate the fact that you included risks and drawbacks that people may not be aware of (like the token usage). This was extremely helpful!
I've been struggling through with some of it this morning and this was the missing link I needed. The CLAUDE.md tip especially. Genuinely useful, thank you.
It helped for sure but still needed a lot of my input - Claude never quite understands what I am trying to say. It also came up with the most bland intro ever and a bunch of titles that were way too long.
This is the most straightforward explainer of CoWork I've seen so far and thank you for including the risks even though most are aware of the risks.
Thanks - I appreciate the comment. It was weird because I used it for a month, then watched a bunch of YT videos before I prepared my live session (which was yesterday) and I still didn't feel that on top of it - had to really tease out some of this info from Claude and Perplexity
This is one of the most useful pieces I have seen on Claude Cowork because it is not just excited about the tool, it is honest about the limits too. The practical lessons and the warning about risk make this much more valuable than a typical AI how to post.
Thank you very much!
Thank you so much, Heather! This was simple and super helpful!
It’s an employee, not a subscription. Couldn’t agree more.
The CLAUDE.md point is the one that changed how I work with agents. A well-stocked folder beats a clever prompt every time. I built this into our Whimsey Labs workflow: every agent session starts with a spec file that defines the arena. The agent does not need to remember anything. It just needs to read.
Agreed and also hard to maintain. Because needs to be updated as things change
Hits home, even for a first timer. Thanks Heather.
This is a good step, but Cowork is Mickey Mouse compared to Claude Code. And then you learn that claude.md and memory and feedback are not enough. Claude is fairly crippled for any serious work. I’ve spent probably 200 hours building solution for it. Welcome to the rabbit hole.
Claude Code is intimidating for anyone who is not a programmer though.
Yes it was for me as well until I decided to “get over it”. You just keep talking to claude or perplexity about how to do it until it’s familiar.
How do you use Code for tasks Cowork could do? I used Code to build an app or something that performs a function, yes, but not a function that can already be carried out in the existing desktop setup, know what I mean? I like "invented" a function with it, where I see Cowork as for carrying out exist functions. I'm not a pro at either (master of the Chat, here), so would love to learn a little from your POV/use cases!
If it’s one‑off and lives across your desktop/apps, use Cowork.
If it’s repeating, and you want a “magic button” you can reuse forever, I use Code once to build that button, then just click it (run it) going forward. Or cron it so it runs daily weekly whatever.
To me the diff is do you want to be a machine operator or build a system that can operate with or for you.
You don’t have to be “a programmer” to do this. With Claude Code, you mostly describe what you want in plain language, let it propose a plan, and then you just run the tool it built for you.
Thank you! This makes sense to me.
awesome. Thank You
Really useful, thank you
Thank you for this! I've been seeing so much about Claude Cowork on socials lately, but it's also been very clear that many of these posts are from people who are being compensated through Anthropic's referral program. I appreciate the fact that you included risks and drawbacks that people may not be aware of (like the token usage). This was extremely helpful!
Wonderful! Thanks!
Learning the hard way is still learning, great share.
Very true. Thanks for commenting.
Super helpful, Heather! Will try to set it up this weekend.
I've been struggling through with some of it this morning and this was the missing link I needed. The CLAUDE.md tip especially. Genuinely useful, thank you.
This is so helpful, thank you for sharing!
Very concise and informative. Did you use Claude Cowork to write it?
It helped for sure but still needed a lot of my input - Claude never quite understands what I am trying to say. It also came up with the most bland intro ever and a bunch of titles that were way too long.