Most people use Claude one-to-one: open claude.ai, type, get an answer. Cowork is the multiplayer layer on top of that — shared projects, shared chats, shared context. It changes what's possible because it changes who's in the loop.
What Cowork actually is
Three building blocks:
- Projects — A persistent workspace with its own files, memory, and instructions. Anyone you invite to a project sees the same context. Claude treats the project's files as long-lived knowledge it can reference across sessions.
- Shared chats — A conversation that more than one human can read and contribute to. Someone on your team starts a thread with Claude; you join it later, see the full history, and continue where they left off.
- Cowork-aware Claude — When Claude is operating inside a shared workspace, it knows which user said what, who's currently in the conversation, and treats the project's files as the source of truth.
If solo Claude is "I have a question," Cowork is "we have a problem."
When to reach for Cowork instead of solo Claude
Reach for Cowork when:
- Multiple people need the same answer. (Claude rewrites the answer once; everyone reads it.)
- The work outlasts a single session. Project files keep the context warm.
- You want a paper trail. Shared chats are reviewable, linkable, and forkable.
- Onboarding. New team members can read prior chats to learn how decisions were made.
Stick with solo Claude when:
- It's a quick lookup or one-off.
- The context is sensitive and shouldn't be shared.
- You're iterating on something private (a draft pitch, salary negotiation prep).
Three patterns that work
1. The team knowledge base
Create a project. Drop in your company's onboarding docs, style guide, security policy, and architectural decision records. Add a one-paragraph project instruction: "When answering, cite the document and section. If the answer isn't here, say so."
Anyone on the team can now ask "what's our policy on X" and get a sourced answer. New hires self-serve. Claude becomes the search interface to your team's tribal knowledge.
2. The async pair-programming thread
Engineer A starts a Cowork chat: "I'm stuck on this auth bug, here's the trace, here's what I tried." Claude offers a hypothesis. Engineer A logs off. Engineer B picks up the thread the next morning, reads the full conversation including Claude's reasoning, and either confirms the fix or points Claude in a different direction.
The thread is the artifact. No screen-share needed. No "let me catch you up" preamble.
3. The decision capture room
Three people, one Cowork chat: "We're choosing between Postgres and DynamoDB for the event log. Here are our constraints. Help us pressure-test each option." Claude facilitates: lists tradeoffs, asks the questions you forgot to ask, surfaces the hidden assumptions.
The chat becomes the decision document. Future you can search "why did we pick Postgres" and find the actual reasoning.
Patterns that don't work
- Treating it like Slack. Cowork is for thinking-with-Claude. If two humans want to chat, use a chat tool. Cowork chats lose value when they fill up with side conversations Claude has to wade through.
- Shared everything. A project where every doc gets dumped in becomes noise. Curate. The project is only as useful as its tightest signal.
- Letting projects rot. When the docs in a project go stale, Claude confidently cites the wrong thing. Treat project docs the same way you treat the README — review them quarterly.
Where to go next
- Set up your first Cowork project — start with one team doc and one shared chat.
- Read 10 Claude Code slash commands — the Cowork mindset (shared context, durable artifacts) carries over to working in your repo.
- Layer Cowork under your Waymaker workflow: the Brain captures personal decisions, Cowork captures team ones. Both feed your operating system.