01 · Executive summary
Everything that matters, compressed. If you only read one page of this pack before teaching, read this one — then use §05 as your lookup table during the session.
The ten things that matter
1. Same engine, different products
Cowork is documented as using "the same agentic architecture that powers Claude Code, accessible within Claude Desktop without opening the terminal." The reasoning loop, the tool-use model, the extensibility system (skills, plugins, MCP) are shared. What differs is the surface, the execution environment, the unit of context, and the assumed user.
2. Cowork is for knowledge work; Claude Code is for engineering work
Anthropic describes Cowork as its "knowledge work agent," aimed at marketing ops, sales, legal, finance and general enterprise teams. Claude Code is "an agentic coding tool that reads your codebase, edits files, runs commands." That said, Claude Code's own docs are careful: "While it excels at coding, it can help with anything you can do from the command line." The boundary is about interface and audience, not a hard capability wall.
3. The single biggest technical difference: where the work executes
Cowork Remote by default
Work runs "in an isolated environment on Anthropic's servers, separate from your computer and your network." Shell commands and code run inside that environment. The desktop app acts as a bridge to your local folders.
Claude Code Local by default
Runs on your machine with access to your files, terminal and git state. Cloud execution (Claude Code on the web) is an option, not the default.
This inverts most people's intuition and has direct consequences for security review, data residency conversations, and what "the app must stay open" means. See §03.
4. Both need a paid plan, but the entry paths differ
Cowork requires a paid Claude subscription: Pro, Max, Team or Enterprise. Claude Code additionally works with an Anthropic Console API key, Amazon Bedrock, Claude Platform on AWS, Google Cloud's Agent Platform and Microsoft Foundry — which matters enormously for enterprises with existing cloud commitments. Cowork has no documented third-party-provider path of its own equivalent to Claude Code's.
5. Cowork consumes usage limits noticeably faster than Chat
Anthropic states this plainly: Cowork "consumes limits faster than Chat" because "complex, multi-step tasks are compute-intensive and require more tokens to execute." Also note that Cowork's Auto permission mode "consumes more of your usage limit than the other modes" because of the extra safety checks. This is the single most common source of executive surprise. See §09.
6. Persistent context works differently in each
| Cowork | Claude Code | |
|---|---|---|
| Unit of context | Project — folders + instructions + links + linked claude.ai projects + project memory | Directory — CLAUDE.md files walked up the tree, plus .claude/rules/ and auto memory |
| Authored by | You, in a GUI form | You, in version-controlled markdown |
| Shared with a team | ✗ Projects live on your computer only, not synced or shared | ✓ CLAUDE.md is committed to the repo |
The "not shareable" property of Cowork projects is a genuine limitation worth flagging to any team lead in your audience.
7. They do not share local configuration
This is a specific, documented, easily-missed fact: Cowork "loads the [connectors, skills and plugins] enabled for your claude.ai account, synced at session start, and doesn't read the Claude Code CLI's ~/.claude directory on your machine." A skill you wrote for Claude Code will not appear in Cowork unless you add it through Customize. Expect this question in every class with developers in it.
8. Dispatch is the bridge between them
Dispatch (Pro and Max, latest Claude Desktop on macOS/Windows) is a long-running agent that takes a high-level brief, splits it into child tasks, and routes each task to the right product: coding work to Code, knowledge work to Cowork. It is the clearest evidence that Anthropic sees these as one portfolio rather than two silos. You can brief it from your phone.
9. Governance maturity is not equal
Claude Code has a deep admin story: managed policy files, server-managed settings, an analytics dashboard and Enterprise Analytics API, Zero Data Retention for qualified accounts, permission deny-rules, hooks as hard enforcement. Cowork has OpenTelemetry monitoring (Team/Enterprise), an org-wide on/off switch, curated plugin marketplaces, and cloud-session controls — but Anthropic explicitly notes that Cowork local session data "is not subject to Anthropic's standard data retention policies" and cannot be centrally managed, while cloud sessions are captured in the Compliance API. See §08.
10. Prompt injection is the named risk for both — and it is Cowork's headline risk
Anthropic's Cowork safety guidance identifies prompt injection as the central threat, and states the two conditions that must hold simultaneously for a successful attack: Claude can access content outside your trust boundary and can perform consequential actions. That framing is the best security teaching device in either product's documentation — it gives executives an actionable mental model rather than a vague warning.
The five-line version for a slide
- Same brain, different hands. One agentic engine; two products shaped for two jobs.
- Cowork = your files, your apps, your documents. No terminal. Deliverables come back finished.
- Claude Code = your codebase, your terminal, your CI. Built for engineers; runs locally.
- Both extend the same way: skills, plugins, connectors (MCP).
- Pick by the artefact, not by the person: is the output a document or a commit?
Differences that are commonly misunderstood
| Common belief | What the documentation actually says |
|---|---|
| "Cowork is Claude Code with a GUI." | Partly true architecturally, wrong operationally. Cowork executes remotely, is organised around Projects rather than directories, has no hooks-as-enforcement story of its own, no git-native workflow, no CI/CD integration, no Agent SDK, and does not read ~/.claude. |
| "Claude Code is only for coding." | Its own docs say it "can help with anything you can do from the command line: writing docs, running builds, searching files, researching topics." Output styles exist specifically to "adapt Claude Code for uses beyond software engineering." |
| "Cowork keeps my data on my machine because it's a desktop app." | The opposite. Local files opened through the desktop app "are processed on Anthropic's servers rather than staying on your computer." |
| "Claude Code needs a Claude subscription." | Not necessarily — an Anthropic Console API key or Bedrock/Vertex/Foundry works for the CLI. But a claude.ai subscription is required for Desktop, web, mobile, Slack, Routines, Remote Control, Chrome, computer use, Artifacts, and voice. |
| "Cowork scheduled tasks need my laptop on." | No — scheduled tasks "run remotely… even when your computer is asleep or the Claude Desktop app is closed." But they therefore cannot be tied to a folder on your computer. Computer use and local file access do require the app open and the machine awake. |
| "Everything is available on every plan." | Computer use and Dispatch are Pro and Max only — explicitly not Team or Enterprise. Code Review is Team and Enterprise only. This non-monotonic pattern surprises people. |
The "Pro/Max only" features (computer use, Dispatch) are the fastest way to lose credibility if you get them wrong, because a senior person on an Enterprise seat will try it live and it won't be there. Say the plan constraint out loud every time you demo one.