Hey
Let me tell you what my normal mornings used to look like.
I would wake up, open 6 different tabs. Check emails. Open Notion. Look at my newsletter draft. Switch back to email. Forget what I was doing. Start over.
Every single day.
I was using Claude too. But here is the embarrassing part. I was using it exactly like Google. Type a question. Get an answer. Close the tab. Next session, Claude remembered nothing. I remembered nothing. We were strangers every single morning.
I thought this was just how AI worked.
I was wrong.
A few days ago I came across Jeff Su's Cowork setup and it broke something in my brain. Because he wakes up, opens Claude, and in one session it reviews his expenses, finishes his newsletter draft, and routes tasks to the right folder without him explaining anything.
Claude remembers his projects. His voice. His rules. His context.
That is not a chatbot. That is Jarvis.
And almost nobody I know is using Claude this way.
The problem most people have with Claude (and do not even know it)
Every time you open a new Claude session, it starts fresh.
No memory of what you worked on yesterday. No knowledge of your writing style. No idea what projects you are running. You have to re-explain yourself. Every. Single. Time.
So you keep prompting Claude with the same context over and over. "I run a newsletter. My audience is non-technical. Write in a casual tone." Again. And again. And again.
This is the equivalent of hiring a personal assistant and firing them every evening. Then hiring a new one every morning and spending 30 minutes re-training them before any real work can happen.
Most people accept this as normal. It is not normal. It is just how most people use Claude.
Jeff Su figured out the fix. And it is so simple it almost feels like cheating.
What Claude Cowork actually is (and why people keep sleeping on it)
Claude Cowork is not a feature most people stumble into.
You can pay for Claude Pro every month, use it daily, and still have no idea Cowork exists in any meaningful way. Because Anthropic does not advertise it loudly. There is no pop-up. No tutorial that forces you to engage with it.
So people open Claude, type their prompt in the chat window, and move on. Completely missing the thing that would 10x their results.
Cowork is where you build Projects. And inside those Projects, you can give Claude permanent context. Files it reads every session. Rules it follows without being reminded. A voice profile it uses whenever it writes for you.
The difference between using Claude in a chat window versus using Claude inside a properly set up Project is the difference between texting a stranger and briefing a business partner who has worked with you for two years.
The architecture that makes this work
Here is what Jeff Su built, and what I think is the real unlock.
Two files. That is the whole system.
CLAUDE.md is a master rulebook. You write your preferences, your workflow rules, your voice guidelines, and how you want Claude to handle different types of tasks. Claude reads this at the start of every session. It becomes the operating system.
memory.md is a running notepad. Claude updates it as you work. Corrections you made once get saved. Preferences you mentioned once do not need to be repeated. It is how the system compounds over time.
Then you layer workstations on top of this. Email HQ. Newsletter HQ. Personal Finance. Each one has its own rules that stack on top of the master rulebook.
The result is a three-level hierarchy. Root rules that apply everywhere. Workstation rules for each area of your life. Project rules for specific tasks inside those areas.
When Claude writes an email inside this setup, it reads your root voice file first, then layers email-specific rules on top. The output sounds like you because it is literally following your documented voice, not guessing.
This is not complicated to build. Jeff does the whole thing with plain text files. No code. No developer needed.
### 100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic say AI now writes 100% of their code.
Are you using AI to write yours?
These [100+ Claude Code hacks](https://magic.beehiiv.com/v1/5f7ce6e3-9a71-416b-99a7-606c5f7e2447?email=bob@reisner.info&redirect_to=https%3A%2F%2Fcodenewsletter.ai%2Fforms%2F14166360-de71-46c4-8722-878d417fab5c&utm_source=beehiiv&utm_campaign=BUBDNZO6MS&redirect_delay=3&_bhiiv=opp_82f549d0-e014-4c81-8c0c-eb01abdb6203_94e90c2e&bhcl_id=bbc3251f-42ac-4b82-b087-18b424f2c78d_f672f2b6-1e60-4422-8448-ee13d3aa1099_22c35bc7-5e17-4448-baa5-49c4bf23f93e) show you exactly how. Sign up for The Code and get:
[Claim your free playbook](https://magic.beehiiv.com/v1/5f7ce6e3-9a71-416b-99a7-606c5f7e2447?email=bob@reisner.info&redirect_to=https%3A%2F%2Fcodenewsletter.ai%2Fforms%2F14166360-de71-46c4-8722-878d417fab5c&utm_source=beehiiv&utm_campaign=BUBDNZO6MS&redirect_delay=3&_bhiiv=opp_82f549d0-e014-4c81-8c0c-eb01abdb6203_94e90c2e&bhcl_id=bbc3251f-42ac-4b82-b087-18b424f2c78d_f672f2b6-1e60-4422-8448-ee13d3aa1099_22c35bc7-5e17-4448-baa5-49c4bf23f93e)
The moments that make this feel like actual Jarvis
Once the system is live, here is what becomes possible.
You screenshot a framework or a template you found online. You paste it in and say "figure out where this belongs." Claude reads the context of your workstations and routes it to the right folder. You do not think about it.
You say "draft a follow-up email from my last meeting." Claude pulls the context, finds the relevant project, loads the right workstation rules, and writes in your voice. You review and send. The whole thing takes 2 minutes instead of 20.
You drop 12 months of credit card statements into Claude and say "build me a spending tracker." Claude creates a full spreadsheet with categories, monthly summary, and yearly overview. You correct one thing once. It never makes that mistake again.
You say "I am going to a client meeting tomorrow, prepare a briefing." Claude already knows your client history, your communication style, and what projects are active. It produces something useful because it has the context to do so.
This is the version of Claude that most people paid for and never unlocked.
Why people are missing this right now
There are three reasons almost nobody in my network is doing this yet.
First, Cowork looks intimidating from the outside. "Projects," "files," "CLAUDE.md" sounds technical. People assume you need to know how to code. You do not. Jeff Su does this entire setup with markdown files. If you can type in Notepad, you can build this.
Second, people are addicted to the instant gratification of the chat window. You get an answer in 10 seconds. It feels productive. But it is shallow productivity. You are solving the same problems again and again instead of building a system that eliminates them.
Third, nobody talks about the setup phase. They show you the output. "Look, Claude wrote this email." Nobody shows the 2 hours of initial setup that made that possible. So people assume the magic just happens, get disappointed with basic Claude, and go back to the chat window.
The setup is the work. But you only do it once. And then Claude does the work forever.
How to actually start today
Do not try to build everything at once. Jeff's advice is very clear here.
Start with one workstation. Pick the area of your life where you repeat the most work. For most people in my audience, that is email or content creation.
Create a folder. Drop in a CLAUDE.md with 3 to 5 rules. Write your voice principles in plain English. Open that folder inside Cowork and star it so every session defaults there.
Run one task inside it. See the difference.
Then build the next workstation. And the next.
The system compounds because every correction you make gets saved. Every preference you log becomes permanent context. The person who starts this week is already ahead of the person who starts next month.
My final thoughts
I will be honest with you.
I put off setting this up for months. It felt like extra work on top of already too much work. I kept telling myself I would do it when things slowed down.
Things never slow down. You know this.
The 2 hours I spent setting up my Cowork OS saved me more time in the first week than I expected. Not because Claude became smarter. Because I stopped re-explaining myself. I stopped losing context between sessions. I stopped starting from zero every morning.
The version of Claude that most people are using is like buying a sports car and only ever driving it in first gear. It moves. But it is not doing what it was built to do.
Cowork is the full gear range.
And it is sitting there inside your Claude subscription, ready to go, waiting for you to actually use it.
Start this weekend. One workstation. One [CLAUDE.md](https://CLAUDE.md) file. Five rules.
That is all it takes to go from chatbot to Jarvis.
Stay curious, talk to you tomorrow.