Your AI Co-Worker Is Already Running
See what Claude Co-Work can do in the next five minutes — before we explain how.
What you'll learn
What You Will See by Module 6
It is 6:14 AM. Your phone buzzes. It is not an alarm — it is a notification from Claude Co-Work. While you were asleep, it checked your email, scanned your calendar, pulled the agenda from your most important meeting, and synthesized everything into a single briefing: three priorities for the day, two emails that need your reply, one scheduling conflict that needs resolving before 9 AM.
You did not ask it to do this. You set it up once — last Tuesday, in about twenty minutes. Now it just runs.
That is the end state. That is what Module 5 produces. And by the time you finish this course, that morning briefing will be one of the smaller things your Co-Work setup does automatically.
Before we explain how any of it works, you need to see it — because the instinct most people bring to this course is wrong. They expect a better chatbot. What they are actually getting is a delegated co-worker.
This Is Not a Better Search Engine
Here is the fundamental difference between Claude.ai chat and Claude Co-Work:
Chat is request-response. You type a message. You get an answer. The conversation ends. Next time, it starts over. The model has no memory of last Tuesday, no access to your Gmail, no ability to do anything unless you are there asking.
Co-Work is delegation with memory. You describe a task once. Co-Work executes it repeatedly, with access to your calendar, your inbox, your files, your tools. It remembers what it learned. It improves over time. And if you tell it to, it runs while you are sleeping.
Jenny at Anthropic describes Co-Work this way: instead of a chat interface where you make requests and wait for responses, think of it as a shared to-do list. You add tasks. Co-Work works through them — with context, with your tools, with memory. This framing is Jenny's personal explanation, not official Anthropic product messaging — but it is the most accurate mental model in circulation.
The practical implication is significant: you write the task once, and Co-Work executes it as many times as you schedule it. There is no re-entering context. No re-explaining your business. No re-typing prompts. The task runs exactly as defined, every time.
Co-Work Is Not the Simplified Version
This is the misconception that stops most people from investing in it properly: they assume Co-Work is the "easy mode" version of Claude — simplified for non-technical users, with capabilities stripped back.
Felix Rieseberg at Anthropic corrected this directly: Co-Work runs the same agentic architecture as Claude Code. It is not a dumbed-down version. Think of it the way you think about VS Code versus a terminal — VS Code is not a worse terminal. It is a different interface to the same underlying power, designed for a different workflow.
Model Overhang is a term Felix Rieseberg (Anthropic) used to describe the gap between what Co-Work can technically do and what most users believe it can do. It is likely internal Anthropic terminology, not in public documentation. The underlying observation — that most Co-Work users are dramatically underutilizing its capabilities — is accurate and well-supported by community evidence.
What this means for you: every agentic capability you have read about in Claude Code is accessible through Co-Work. Computer Use. Sub-agents running in parallel. Complex multi-step workflows executing without human supervision. The interface is friendlier. The ceiling is identical.
The Three Pillars
Everything Co-Work does autonomously comes from the combination of three components. Understanding what each one does — and critically, what it is not — is the foundation this entire course builds on.
Skills
Skills are reusable instruction sets stored as plain markdown files. You trigger them with a "/" command or natural language. When you trigger a skill, Co-Work arrives pre-briefed: it already knows the role it is playing, the output format you expect, and the constraints it should follow. Skills eliminate the need to re-explain context on every task. You build a skill once, improve it over time, and run it indefinitely. Module 6 is dedicated to building and managing skills.
Connectors
Connectors are MCP-based integrations that give Co-Work read and write access to your tools — Gmail, Google Calendar, Notion, Slack, HubSpot, and approximately 50 to 70 others at launch. No code required. Authentication is OAuth. Connectors are what allow Co-Work to pull your actual email rather than ask you to paste it, check your real calendar rather than work from a description, and save output directly to the folder where it belongs. Module 4 covers connectors in depth — including the permission trust model and when to use Chrome as a connector alternative.
Scheduled Tasks
Scheduled tasks are when everything becomes a co-worker rather than a tool. You define a task — "build my morning briefing" — and attach a schedule: daily at 6 AM, every weekday at 9 PM, hourly. Co-Work runs the task automatically according to that schedule, using whatever connectors and skills the task requires. No prompt needed. No human trigger. It just runs. Modules 5 and 8 cover scheduling in full — including the hardware dependency you need to understand before relying on scheduled tasks.
The formula is: Skills + Connectors + Scheduled Tasks = an actual co-worker. Remove any one of the three and you have a sophisticated chatbot. Combine all three and you have something fundamentally different.
What Runs Where: The System Architecture
Co-Work runs in a virtual machine on your local computer. It is the same machine, but the VM provides a sandboxed environment — what happens in Co-Work stays contained within that VM boundary. Most things you do in Co-Work run inside this sandbox.
Two capabilities are worth flagging now because they sit outside that boundary:
- Computer Use runs outside the VM, directly on your host machine. It can control your actual desktop — clicks, keystrokes, browser navigation. This power requires understanding the safety model. Module 13 covers this in full.
- Dispatch is a persistent cross-device thread that bridges your desktop Co-Work instance to mobile. Available on Pro and Max plans. Covered in Module 12.
Computer Use gives Co-Work direct control of your host machine — not the sandbox. This distinction matters for safety. Before enabling Computer Use, you need to understand its permission model and guardrails. We cover everything in Module 13. For now, leave it disabled.
One more hardware note that matters: scheduled tasks require your computer to be awake and the Co-Work desktop app to be open. If your laptop sleeps overnight, the 6 AM briefing does not run. We will cover the solutions — including the dedicated Mac Mini approach used by several practitioners — in Module 5.
Who This Course Is For
This course serves three different learners, and they have different finishing lines.
Alex is the overwhelmed professional — running a small business or managing a complex individual workload, drowning in recurring tasks, skeptical that AI is actually going to help. Alex's goal is specific: stop doing the same five tasks manually every week. Modules 1 through 6 are built for Alex. By Module 6, Alex has three working automated workflows and a setup that actually sticks.
Morgan is the systematic builder — already convinced AI can help, wanting to build a proper system rather than one-off automations. Morgan's goal is architecture: a Co-Work setup that scales as the business grows. Modules 1 through 12 are built for Morgan. By Module 12, Morgan has a complete operating system: folder hierarchy, connectors, skills, scheduled automations, and Dispatch configured for mobile.
Sam is the AI-native operator — ready to go deep on Computer Use, sub-agents, compliance, enterprise deployment, and the Capstone. All 18 modules. By the end, Sam has built and documented an entire workflow category — an AI Employee Day audit — and has the architecture to run it at scale.
Even if you are clearly Sam, do not skip ahead to Module 13. The foundation — global instructions, folder architecture, connectors, skills — is load-bearing. Every advanced capability in Tier C depends on getting Tier A right. The course is designed so that speed-running Modules 1–6 is fast (a few hours), and the payoff in later modules is significant.
Before the next module, open Claude Co-Work and complete this orientation exercise. If you have not yet installed it, follow along with the screenshots and do this exercise when you are set up.
- Locate the Skills panel — find at least one installed skill (the "Writing" or "Research" skills are common defaults). Note what triggers it.
- Open Connectors — identify which apps are already connected and which are available but not yet set up.
- Open Scheduled Tasks — check the task history. It may be empty right now. That changes by Module 5.
- Write one sentence: "The task I most want to automate is ___." Keep this sentence. It is the throughline exercise for the entire course, and you will build it before you finish.