05 · Capability matrix
The reference section. Grouped side-by-side comparison across every dimension where the two products meaningfully differ or overlap. Keep this open during Q&A.
How to read the ✗ marks
A ✗ means either (a) the documentation explicitly says the feature is unavailable, or (b) the feature is a first-class documented concept for one product and entirely absent from the other's documentation. Case (b) is absence of evidence, not proof of impossibility. Where the distinction matters I say so in the row.
A. Identity & positioning
| Dimension | Cowork | Claude Code |
|---|---|---|
| Official one-liner | "Anthropic's agentic workspace" / "knowledge work agent" | "An agentic coding tool that reads your codebase, edits files, runs commands" |
| Primary user | Knowledge workers: marketing ops, sales, legal, finance, enterprise teams | Software engineers and engineering orgs |
| Primary artefact | Documents, decks, spreadsheets, organised files, research briefs | Code changes, commits, pull requests, builds, tests |
| Terminal required | ✗ Explicitly "without opening the terminal" | ◐ Not required (Desktop/IDE/web), but the CLI is the most complete surface |
| Maturity of documentation | ~6 doc pages + ~10 help-centre articles | 200+ documentation pages + weekly release digests |
B. Execution & environment
| Dimension | Cowork | Claude Code |
|---|---|---|
| Default execution location | Anthropic servers (isolated VM) | Your machine |
| Cloud execution option | Is the default | ✓ Claude Code on the web / Routines |
| Runs while laptop is closed | ✓ (remote sessions, scheduled tasks) | ◐ Cloud sessions and Routines yes; local sessions no |
| Access to your internal network | ✗ Environment is "separate from your computer and your network" | ✓ Local sessions run inside your network |
| User-tunable sandbox | ✗ Isolation is architectural; egress is admin-configurable | ✓ /sandbox, dev containers, worktrees, cloud VM network levels |
| Working-directory write boundary | Connected project folders | ✓ Start folder + subfolders; parents need explicit permission |
| Local shell access | ✗ Shell runs in the remote environment | ✓ "Any command you could run" |
| Git awareness | ✗ Not documented | ✓ Branch, uncommitted changes, commit history; stages, commits, branches, opens PRs |
C. Files & content
| Dimension | Cowork | Claude Code |
|---|---|---|
| Reads/writes local files | ✓ In connected folders, no manual upload/download | ✓ In the working directory tree |
| Individual file size limit | 50 MB per file read | Not published as a single number (context-bound; large outputs trigger compaction) |
| Supported file types | Documents, spreadsheets, presentations, images, data files, code, notebooks | Any text file; images; PDFs; Jupyter notebooks |
| Produces Office deliverables | ✓ Headline capability — Excel with functional formulas, PowerPoint, formatted docs | ◐ Possible via skills/scripts, not a marketed capability |
| Publishes shareable web pages | ✗ Not documented | ✓ Artifacts (Pro/Max/Team; admin-enabled on Enterprise) |
| Drag & drop into a project | ✓ Files copied into the first folder; folders mounted as additional project folders | ◐ IDE/Desktop drag-drop; CLI uses paths and --add-dir |
D. Context, memory & project setup
| Dimension | Cowork | Claude Code |
|---|---|---|
| Unit of organisation | Project | Directory / repository |
| What a project/context holds | Description, folders, instructions, links, linked claude.ai projects, memory | CLAUDE.md (+ local + rules), settings, skills, subagents, hooks, MCP config, workflows |
| Standing instructions | ✓ Project Instructions field | ✓ CLAUDE.md at managed / user / project / local scope |
| Path-scoped instructions | ✗ | ✓ .claude/rules/ with paths: frontmatter globs |
| Agent-authored memory | ✓ Project-scoped memory store | ✓ Auto memory; 200-line / 25 KB index budget |
| Shareable with teammates | ✗ "Projects live on your computer… not synced to the cloud or shared with other people" | ✓ CLAUDE.md, skills, settings committed to source control |
| Org-wide instructions | ◐ Via required plugins / admin settings | ✓ Managed-policy CLAUDE.md + claudeMd key in managed settings |
| Links to claude.ai projects | ✓ Can draw on a claude.ai project's knowledge without merging | ✗ |
| Context inspection | ✗ Not documented | ✓ /context, /mcp, interactive context-window walkthrough |
| Archiving | ✓ Deletes project metadata; local folders untouched | ◐ claude project purge clears local state per project |
E. Extensibility
| Dimension | Cowork | Claude Code |
|---|---|---|
| Skills | ✓ Managed via Customize | ✓ Files in .claude/skills/, user, plugin, managed scopes |
| Plugins | ✓ Marketplace, git repos as marketplaces, upload from file | ✓ Same system, plus authoring & marketplace hosting docs |
| Connectors / MCP | ✓ Via Customize; synced from claude.ai account | ✓ Local/project/user scope, tool search, claude mcp login |
| Hooks | ◐ Only as part of an installed plugin | ✓ First-class: lifecycle events, shell/HTTP/prompt/subagent hooks, JSON schema, exit codes |
| Custom subagents you define | ✗ Only via plugin-provided agents | ✓ Full definition, tool restriction, preloaded skills, own memory |
| Slash commands | ◐ Via plugins | ✓ Built-in + bundled skills + custom |
Reads ~/.claude | ✗ Explicitly does not | ✓ It is ~/.claude |
| Programmatic SDK | ✗ | ✓ Claude Agent SDK (Python + TypeScript), headless, hosting guides |
| Output styles | ✗ | ✓ "Adapt Claude Code for uses beyond software engineering" |
| Industry plugin packs | ✓ Official financial-services set (modeling, equity research, IB, PE, wealth) | ◐ Code intelligence, security plugins, marketplaces |
Plugin limits (both products): 200 MB uncompressed package, 5,000 files per package, 512 MB marketplace archive, 500 plugins per marketplace, 25 marketplaces.
F. Autonomy & delegation
| Dimension | Cowork | Claude Code |
|---|---|---|
| Permission modes | Manual · Auto · Skip | Six. In the Shift+Tab cycle: Manual · Accept edits · Plan · Auto. Flag-only: dontAsk (auto-denies anything not pre-approved) · bypassPermissions / --dangerously-skip-permissions (skips all checks; isolated VMs only; blockable by admins) |
| Plan-before-act mode | ◐ You "review Claude's approach" before execution | ✓ Dedicated Plan mode; Ultraplan for cloud-drafted plans |
| Fine-grained allow/deny rules | ◐ Admin toggles for connector "Always allow" | ✓ permissions.allow/ask/deny with tool + parameter matching |
| Background / long-running work | ✓ Remote sessions continue after you close the laptop | ✓ Cloud sessions, background subagents, agent view |
| Dispatch (brief from phone) | ✓ Routes knowledge work into a Cowork project | ✓ Routes coding work into a Code workspace |
| Scheduled / recurring tasks | ✓ Hourly, daily, weekly, weekdays, manual — run remotely (beta) | ✓ Routines (cloud), Desktop scheduled tasks (local), /loop (in-session) |
| Event-triggered runs | ✗ Not documented | ✓ Routines on API calls / GitHub events; Channels; GitHub Actions |
| Parallel workstreams | ✓ Sub-agent coordination; Dispatch child tasks | ✓ Subagents, agent view, agent teams, dynamic workflows, worktrees, /batch |
| Goal-until-condition | ✗ | ✓ /goal |
| Undo / rewind | ✗ Not documented | ✓ Checkpoints, Esc Esc, /rewind |
G. Integrations & the outside world
| Dimension | Cowork | Claude Code |
|---|---|---|
| Browser automation | ✓ Claude in Chrome | ✓ Chrome — test web apps, console logs, forms, data extraction |
| Computer use (screen control) | ◐ Pro/Max only, research preview, macOS + Windows | ◐ Pro/Max only, research preview. Desktop: macOS + Windows. CLI: macOS only. Not in non-interactive -p mode; not on third-party providers |
| Web search | ✓ (admin-disableable under Capabilities) | ✓ (provider-dependent — see §04) |
| SaaS connectors (Gmail, Drive, Slack, Linear…) | ✓ Core to the product | ✓ Via MCP + claude.ai connectors |
| CI/CD | ✗ | ✓ GitHub Actions, GitLab CI/CD, GitHub Enterprise Server |
| Automated PR review | ✗ | ✓ Code Review (Team/Enterprise), /code-review ultra |
| Team chat entry point | ◐ Slack connector on Team plans | ✓ @Claude in Slack returns pull requests |
H. Administration, monitoring & compliance
| Dimension | Cowork | Claude Code |
|---|---|---|
| Org-wide on/off | ✓ On by default; owners can disable (Org settings → Cowork) | ◐ Governed through managed settings, auth locks and seat assignment |
| OpenTelemetry export | ✓ Team/Enterprise; logs/events protocol | ✓ All providers; metrics + events |
| Analytics dashboard | ✗ Not documented | ✓ Team/Enterprise dashboard; Enterprise Analytics API |
| Managed configuration files | ◐ Admin settings + 3P deployment guide | ✓ managed-settings.json, managed CLAUDE.md, server-managed settings |
| Curated plugin marketplace | ✓ Four tiers: Installed by Default / Available / Required / Not Available | ✓ Marketplaces, plugin dependencies, required version ranges |
| Compliance API coverage | ◐ Cloud sessions only; local session data is not centrally managed | ◐ Compliance API on Enterprise |
| Zero Data Retention | ✗ Not documented for Cowork | ✓ Qualified Enterprise / API accounts |
| SSO / SCIM | Plan-level, not product-level: SSO on Team & Enterprise; SCIM Enterprise only | |
| Spend controls | ✗ Not documented | ✓ Per-developer spend limits via Claude apps gateway; /usage; cost guidance |
I. Availability by plan
| Feature | Free | Pro | Max | Team | Enterprise |
|---|---|---|---|---|---|
| Cowork | ✗ | ◐* | ◐* | ◐* | ◐* |
| Cowork on web / mobile | ✗ | ◐ beta | ◐ beta (rollout began here) | ◐ beta | ◐ beta |
| Cowork scheduled tasks | ✗ | ◐ beta | ◐ beta | ◐ beta | ◐ beta |
| Cowork monitoring (OTel) | ✗ | ✗ | ✗ | ✓ | ✓ |
| Cowork cloud sessions default | — | on | on | on (owner can disable) | off (owner must enable + grant via custom role) |
| Claude Code CLI | ✗ | ✓ | ✓ | ✓ | ✓ |
| Claude Code on the web | ✗ | ✓ | ✓ | ✓ | ◐ premium or Chat+Code seat |
| Routines | ✗ | ✓ | ✓ | ✓ | ✓ |
| Remote Control | ✗ | ✓ | ✓ | ◐ admin-enabled | ◐ admin-enabled |
| Channels | ✗ | ✓ | ✓ | ◐ admin-enabled | ◐ admin-enabled |
| Computer use | ✗ | ✓ | ✓ | ✗ | ✗ |
| Dispatch | ✗ | ✓ | ✓ | ✗ | ✗ |
| Code Review | ✗ | ✗ | ✗ | ✓ | ✓ |
| Artifacts | ✗ | ✓ | ✓ | ✓ | ◐ admin-enabled |
| Analytics dashboard | ✗ | ✗ | ✗ | ✓ | ✓ |
| Enterprise Analytics API | ✗ | ✗ | ✗ | ✗ | ✓ |
| Server-managed settings | ✗ | ✗ | ✗ | ✓ | ✓ |
| SSO | ✗ | ✗ | ✗ | ✓ | ✓ |
| SCIM | ✗ | ✗ | ✗ | ✗ | ✓ |
| Compliance API | ✗ | ✗ | ✗ | ✗ | ✓ |
| Zero Data Retention | ✗ | ✗ | ✗ | ✗ | ◐ separate enablement, qualified accounts |
* Anthropic's get-started article still describes Cowork as "rolling out over the next several weeks starting with the Max plan, with more plans to follow." Availability on a given seat is not guaranteed — check before demoing.
The two rows to point at
Computer use and Dispatch are Pro/Max only — not Team or Enterprise. If your class is all on Enterprise seats, do not build a demo around either.