The short answer: If the thing you're making gets committed to a repository, reach for Claude Code; if it becomes a document, spreadsheet, deck, or a decision pulled from scattered inputs, reach for Cowork. Claude Code is Anthropic's terminal coding agent for developers — it lives in your terminal, IDE, and CI. Cowork is Anthropic's general office agent for founders and operators — it runs on desktop, web, and mobile, works across your files and apps, and keeps going after you close the laptop.
The one-line rule#
Both are Anthropic agents. The split is about where your work lives.
- Use Claude Code if… the work is code in a repo — a bug fix, a feature, a refactor, an internal tool, anything that ends in a commit or a PR.
- Use Cowork if… the work is knowledge work around the business — research, docs, spreadsheets, decks, analysis, meeting prep — anything that ends in a file or a decision, not a commit.
- Rule of thumb: Does it get committed to git, or does it become a doc/sheet/deck? That single question resolves ~90% of cases.
What Claude Code is#
Claude Code is Anthropic's terminal-native CLI coding agent. It runs in your terminal, integrates with your IDE, and can run headless in CI — and it's locked to Anthropic's own models (docs). The point of the tool is that it operates on a real repository: it reads your codebase, makes coordinated edits across many files, runs your tests, and opens pull requests.
The pieces that make it a genuine engineering agent rather than a chat window:
- MCP (Model Context Protocol) to connect external tools and data sources.
- Subagents — built-in agents like Explore, Plan, and General-purpose that run in parallel to speed up codebase research and multi-file work.
- Plan mode — it drafts an implementation plan you can review before it touches anything.
- Hooks, slash commands, and skills for wiring the agent into your own workflow (CLI reference).
One recent change worth knowing: as of August 14, 2026, Claude Code's default permission mode flipped to auto mode for Pro, Max, and Team users — a safety classifier now adjudicates each command instead of prompting you for every one. We covered what to check before that switch. If you're weighing Claude Code against other terminal agents, see our head-to-head on Claude Code vs Codex CLI vs Gemini CLI and our current pick for the best LLM for coding.
Pricing: Claude Code has no standalone price — it's bundled into paid Claude plans (Pro, Max 5x, Max 20x, Team, Enterprise) or available pay-as-you-go via the API (pricing coverage). Terms move; as of publication, see Anthropic's pricing page for current numbers.
What Cowork is#
Cowork is Anthropic's general-purpose office agent — the sibling of Claude Code that lives in your operations instead of your repo. It's built for the knowledge work that isn't code: research, document drafting, spreadsheets with working formulas, decks, analysis, and prep (tutorial).
The specifics that matter:
- Where it runs: It launched as a desktop app for macOS and Windows in January 2026, and expanded to web (at claude.ai) and mobile (iOS/Android) on July 7, 2026 (TechCrunch).
- What it works on: It reads, edits, and creates files directly on your machine, and coordinates subagents for parallel work.
- Connectors: It plugs into your apps through the MCP ecosystem — Microsoft 365 (Outlook, OneDrive, SharePoint, Teams), Zoom, and more (agentic plug-ins).
- Async, always-on work: It keeps running after you close the laptop, runs scheduled tasks, and pings your phone only when it hits a decision that's yours to make.
The signal underneath the product: when Anthropic analyzed 1.2 million anonymized sessions, 90%+ of Cowork usage had nothing to do with coding — business operations led at ~33% and content creation at ~16%, while software development was under 9% (VentureBeat). The bet is that the next AI battleground is the rest of the office, not the IDE. We wrote about Cowork going mobile for founders, compared it head-to-head in Claude Cowork vs ChatGPT Work, and stacked it against the enterprise field in ChatGPT Work vs Gemini Enterprise vs Claude Cowork.
Pricing: Like Claude Code, Cowork is bundled into paid Claude plans with no separate SKU — though feature access can roll out plan-by-plan (mobile and web landed on the Max plan first). As of publication, see Anthropic's pricing page for current terms.
Which one for which job#
Concrete scenarios, because the rule is easier to feel than to state:
- Ship a bug fix → Claude Code. It's a repo change that ends in a commit.
- Turn a messy folder of notes into a client report → Cowork. Files in, a finished doc out.
- Build an internal tool → Claude Code. Multi-file code, tests, a PR.
- Weekly ops review from five dashboards → Cowork. Pull across apps, synthesize, hand back a briefing — and schedule it to run every Monday.
- Refactor a service and update its tests → Claude Code. Coordinated edits across the codebase.
- Prep a board deck from last quarter's spreadsheets and emails → Cowork. It reads the inputs, builds the slides, leaves the follow-up drafted for your review.
If you're ever unsure: the deliverable decides. A commit is Claude Code's job; a document, sheet, deck, or decision is Cowork's.
Using both#
The realistic founder workflow uses them in sequence. Cowork plans and researches; Claude Code builds.
Say you want to ship a new feature. Start in Cowork: have it research how competitors solve the problem, pull the relevant threads and notes from your connected apps, and turn all of it into a clear one-page spec. That's the messy, cross-app, "figure out what to build" half — exactly Cowork's strength. Then hand the spec to Claude Code: point it at your repo, let it draft an implementation plan in plan mode, make the edits, run the tests, and open a PR. That's the "build it" half — exactly Claude Code's strength.
Many founders keep Cowork open on their phone for ops and research during the day and let Claude Code run in the terminal for the build. Same account, two agents, two jobs.
Both Claude Code and Cowork are Anthropic products built for different jobs — code in a repo versus knowledge work across your apps. Pricing, plan eligibility, and features move fast, so check the source links above (and Anthropic's pricing page) for current terms.
