Yesterday's Claude OS in a Day session ran four hours.
Every person came in with a working business and a Claude account. Every person left with a functioning operating system.
Here's what that looked like.
The First Hour: Writing the CLAUDE.md
Most people have never written a CLAUDE.md. They've been using Claude as a chatbot — starting fresh every time, re-explaining context, pasting the same background into every session.
The CLAUDE.md changes that.
It's a plain text document at the root of your project that tells Claude everything it needs to know about your business before you say a word. Your name and role. What your company does. How you talk to clients. What tools you use. What's off-limits. What decisions Claude can make autonomously versus what it needs to flag.
We spent the first hour writing these live. Every person's CLAUDE.md was different because every business is different. A consultant's looked nothing like an e-commerce operator's. That's the point — you're not filling out a template, you're writing a job description for the AI that runs your company.
The shift was immediate. People could feel the difference the second they tested a prompt with their CLAUDE.md loaded versus without it.
The Second Hour: Connecting MCP Servers
A system prompt tells Claude what you do. MCP servers give Claude the ability to actually do it.
We connected three categories of tools:
Communication. Email and calendar access. Claude can now read your inbox context, draft responses in your voice, and understand your schedule — without you pasting anything in.
Data. One or two data sources per person — a CRM, a Notion workspace, a spreadsheet tracking client work. Claude can query these directly instead of requiring you to extract and paste information manually.
Publishing. Whatever each person's output is — blog posts, social content, client reports — we connected the endpoint. So Claude can draft and, in some cases, deliver directly.
This is where Claude stops being a writing assistant and starts being an operator.
The Third and Fourth Hours: Three Workflows
The workflows are where it clicks.
A workflow is a sequence Claude runs without you initiating it — triggered by a schedule, an event, or a condition. We built three per person:
Morning brief. Claude pulls from calendar, email, and project data and generates a daily context summary before you start work. No more orienting yourself every morning.
Intake handler. New lead, new client request, new support ticket — whatever the business receives, Claude classifies it, extracts the key info, and either handles it or routes it with a recommended action.
Status reporter. At end of day or week, Claude reviews what changed across your systems and generates a summary. What moved. What's stuck. What needs attention before it becomes a problem.
Three workflows. That's what most people are missing — not more prompts, but systems that run without them.
The Thing Nobody Expects
People come into Claude OS in a Day expecting to feel overwhelmed.
They leave feeling like they got their time back.
That's not marketing copy. That's what happens when the infrastructure is actually yours — built around your tools, your voice, your business logic. It doesn't feel like using software. It feels like having a team member who knows exactly how you work.
The bottleneck most people have isn't capability. It's setup. Claude OS in a Day is four hours of setup that you'd otherwise spend six months doing badly.
What's Next
If you were in yesterday's session, you have the foundation. The next layer is client delivery.
Claude for Agencies is tomorrow — Wednesday, April 29 at 10 AM ET. Built for people who are serving other businesses, not just running their own. We build the workflow layer that lets you take on more clients without adding more hours: the intake system, the delivery pipeline, the QA process Claude runs on your behalf.
It's the last live session of the April cohort.
Reserve your seat for Claude for Agencies →
— Murph
