@itsmariahbrunner — Claude at Work

Welcome to the
era of vibe
working.

Claude isn't a chatbot anymore. It's an agent. And the people who figure out how to delegate to it — not just talk to it — are about to look untouchable at their jobs.

How to set this up for your role

What vibe working actually means

You've heard of vibe coding — where developers describe what they want and AI builds it. No writing code line by line. Just describe the outcome and step away. Anthropic's Head of Product says we're entering that same era for every job, not just engineering. He's calling it vibe working.

The idea is simple: stop managing tasks, start delegating outcomes. Instead of using Claude as a sophisticated search engine or writing assistant, you hand it a real piece of work — full context, clear outcome, everything it needs — and let it figure out every step. You come back to something finished.

A year ago, Claude was something you talked to. You asked it a question, it answered. Useful, but still just a chatbot. That's not what it is anymore. Now Claude can plan, execute, use your tools, write to your files, coordinate multiple workstreams — and do all of it while you've already moved on to something else.

The shift — how people used to use Claude vs. how it actually works now

How most people still use it
Managing tasks
  • "Write me an email to my client about the project delay"
  • Claude writes it. You copy it. You send it.
  • You ask one question, get one answer, apply it yourself
  • Claude is a tool you operate step by step
  • You're still doing the coordination and execution
How it actually works now
Delegating outcomes
  • "Here's the full situation. Here's what needs to happen. Here's everything you need. Go."
  • Claude reads your project files, drafts the email, checks your calendar, prepares follow-up materials
  • You hand off a problem, not a task
  • Claude figures out every step and comes back with the finished result
  • You move on to something else while it works
"

Don't say "write me an email." Say "here's the full situation, here's what needs to happen, here's everything you need — go."

What this looks
like in practice.

The difference between task management and outcome delegation isn't just philosophy — it's a completely different way of writing prompts. Here's exactly what changes.

Old way — task management
"Write a follow-up email to the client after our call."
New way — outcome delegation
"Here's the call recording summary, the client's original brief, the proposal we sent, and the three open issues from today's call. Draft a follow-up email that confirms what was decided, addresses each open issue with the resolution we discussed, and sets clear next steps with owners and dates. Tone should match how I write — direct but warm. Save it to /drafts."
The rule of thumb

Before you send a prompt, ask yourself: "Am I describing a task, or an outcome?" If you can't answer "what does finished look like?" in your prompt — Claude can't either. The more complete your context, the less back-and-forth you need. Give it the full picture once and walk away.

The three things you need set up to actually do this

01
Projects
So Claude already knows your role and context before you say a word.

The reason most people's Claude prompts need to be so long is that they're re-explaining everything from scratch every single time. Who you are, what your role is, what your brand sounds like, what you're working on. Projects eliminate that entirely.

A Project is a persistent workspace where Claude holds your context across every conversation. You set it up once — your role, your communication style, your team structure, your current priorities, relevant documents — and from then on, every conversation in that Project starts with Claude already fully briefed. You skip straight to the work.

  • Go to claude.ai → click "Projects" in the left sidebar → "New Project"
  • Give it a name that matches how you think about your work (e.g. "Client Work," "Content," "Operations")
  • In the Project instructions, write 5–10 sentences: your role, your communication style, what this Project is for, and any standing rules Claude should always follow
  • Upload your most relevant documents — past work samples, brand guidelines, templates, anything Claude would need to understand context
  • Every conversation you start inside that Project inherits all of it automatically
What good Project instructions look like

"I'm a Senior Marketing Manager at a B2B SaaS company. I manage a team of 4. My communication style is direct — no filler, no corporate speak. This Project is for client-facing work. Always use formal language in client communications. When writing emails, lead with the key point. Never include more than 3 action items in a single message. When you're unsure about tone, ask before drafting."

02
Connectors
So Claude is living inside your actual tools, not working around them.

Without connectors, Claude works in isolation. You have to paste in your emails, copy over your notes, describe your project status. It works, but it's friction. Connectors remove that friction entirely — Claude can read from and act within your actual tools directly.

Connect Gmail and Claude can triage your inbox, draft responses in your voice, flag what needs your attention. Connect Google Drive and it can access documents by name without you uploading anything. Connect Notion and it can read your project boards and incorporate real status into everything it produces. Connect Slack and it can draft messages for the right channels. The work stops being siloed between "Claude" and "everything else."

  • Go to claude.ai → Settings → Integrations
  • Connect the tools you live in — Gmail, Google Drive, Notion, Slack, and more
  • Sign in with your account and grant permissions — 60 seconds per tool
  • Claude can now read from and act within those tools from any conversation
  • Start with the two tools you use most — email and wherever your docs live
What this unlocks immediately

"Check my Gmail, read the Notion board for Project Phoenix, and give me a 5-bullet briefing on where things stand, what needs a response today, and whether we're on track for the Friday deadline." — That prompt does nothing without connectors. With them, it takes 30 seconds to run and 2 minutes to read.

03
Cowork
So Claude can run multi-step tasks while you've already moved on.

Chat is a conversation. Cowork is a delegation. When you use Chat, Claude responds and waits. When you use Cowork, Claude makes a plan and executes it — reading and writing real files on your computer, coordinating multiple workstreams in parallel, and running tasks on a schedule you set. You describe the outcome, step away, and come back to finished work.

This is where vibe working actually lives. Cowork is available in the Claude Desktop app (Mac + Windows x64, download at claude.com/download) for Pro, Max, Team, and Enterprise subscribers. You'll see three tabs at the top: Chat, Cowork, Code. Click Cowork, give it access to a folder, describe what you want done, and step away.

  • Download the Claude Desktop app at claude.com/download — update to the latest version
  • Click the Cowork tab at the top of the app
  • Click "Work in a folder" and give it access to a project folder — start focused, not your whole drive
  • Describe the outcome specifically: what file should exist when it's done, what should be in it, what format
  • Review the plan it proposes before it runs — approve it, then walk away
  • For recurring work, save it as a scheduled task and set the cadence once
The highest-leverage use of Cowork

"Every Monday at 8am: Read the project files in /active-projects and my Notion boards, check Gmail for anything from clients over the weekend, and create a weekly priorities doc in /briefings — top 5 things I need to move this week, with any client messages that need a reply today flagged at the top." Set this up once. It runs every week while you're making coffee.

The prompt — build your vibe working setup for your specific role
I want to set up Claude properly for vibe working — delegating outcomes, not managing tasks. Help me build the full setup for my role.

My role: [job title and 2–3 sentences on what you actually do day to day]
My industry: [e.g. marketing, finance, operations, healthcare, legal]
Tools I use most: [e.g. Gmail, Notion, Slack, Google Drive, Salesforce]
What eats most of my time right now: [2–3 things that take too long or happen too often]

Step 1 — Write my Project instructions
Based on my role and context, write a complete Project instruction set I can paste into a new Claude Project right now. It should cover: who I am and what I do, my communication style and tone, standing rules Claude should always follow in this Project, what "good output" looks like for my work, and anything Claude should always ask before doing.

Step 2 — Tell me which connectors to set up first
Given the tools I listed, which 2–3 connectors should I set up this week for maximum impact? For each, give me one specific prompt I could run the moment it's connected that would immediately save me time.

Step 3 — Write me my first Cowork task
Based on what's eating most of my time, write a complete Cowork task brief I can paste in right now. Specific outcome, file format, what to do with edge cases, and a request to show me the plan before executing.

Step 4 — Give me my first outcome-delegation prompt
Write an example of what a well-structured outcome-delegation prompt looks like for one of the tasks I described — something that gives Claude full context, a clear outcome, and everything it needs to complete the work without coming back to ask me questions.

Make everything specific to my role. The goal is to walk away from this with a setup I can actually use today.

The mistakes people make when they try to shift to this

01
Still prompting like it's a chatbot

Short prompts with no context get short answers with no context. If you want Claude to operate like an agent, brief it like one. Full situation. Clear outcome. Everything it needs. A prompt that takes you 3 minutes to write can save you 3 hours of work. The investment is worth it.

02
Setting up Projects but never uploading documents

Project instructions alone are good. Project instructions plus your actual documents — past work samples, templates, brand guidelines, client briefs — are transformative. Claude learns your style from examples far faster than from descriptions. Upload 3–5 pieces of work you're proud of and it starts producing things that actually sound like you.

03
Trying to do everything at once

Don't try to set up Projects, connectors, and Cowork in the same afternoon. Pick one. Set up one Project this week, use it for everything in that domain, and notice how different it feels. Then add the next layer. People who try to overhaul everything at once usually end up using none of it consistently.

04
Not telling Claude what success looks like

The most common reason for disappointing output: Claude didn't know what done meant. Always define the output — what file, what format, what length, what's in it. "Create a 1-page summary as a Word doc in /outputs" gets you something usable. "Summarize this" gets you whatever Claude decides is a summary.

05
Treating every task the same

Not everything should be delegated. Relationship decisions, high-stakes judgment calls, anything where the nuance is yours — Claude helps you prepare for those, it doesn't replace you in them. The skill is knowing which tasks are ripe for full delegation and which ones just need Claude at the edges. When in doubt: hand off the prep work and keep the decision.

Want the full playbook?

I teach this
in depth.

I built a free resource that shows you how to set all of this up for your specific role — Projects, connectors, Cowork, and how to start delegating outcomes from day one. Follow for daily Claude content.

TikTok @itsmariahbrunner Instagram @itsmariahbrunner