Chat, Cowork, and Code are three tabs in the Claude desktop app on one subscription. Side-by-side table, decision questions, and the five misallocations we see most often.
TL;DR. One Claude subscription gives every user three tabs in the desktop app. Chat is for talking. Cowork is for working on files. Code is for building software. The most common mistake at a mid-market company is buying Claude Code for an ops team that needed Cowork — or trying to make Chat do what Cowork does. Neither costs extra to fix; both cost momentum.
Three tabs. One subscription. One mental model.
If a user does not need to act on files, do not give them a Cowork seat brief. They will use Chat happily. If a user needs to write production code, do not point them at Cowork. They will get frustrated within the hour.
| Chat | Cowork | Code | |
|---|---|---|---|
| Primary user | Anyone — execs, sales, knowledge workers | Operators, finance, ops, marketing, legal, HR | Software engineers |
| Where it runs | Browser + desktop tab | Desktop tab (Mac, Windows) | Terminal + desktop tab |
| Computer / file access? | No | Yes — one folder you grant | Yes — repos, shell, full system |
| Approves actions step-by-step? | N/A — no actions | Yes — plan-then-act loop | Yes — configurable |
| Output formats | Text in the chat window | .docx, .xlsx, .pptx, .pdf, images, code, anything in the folder | Source code, commits, PRs |
| Best for | Thinking out loud, drafting, Q&A | Recurring document workflows, multi-file tasks | Building or maintaining software |
| Subscription | Same as the others | Same as the others | Same as the others |
| Notable risks | Hallucination on facts the user won't check | Editing the wrong file if the workspace is messy | Running an unintended shell command |
The subscription column is not a typo. Pro, Max, Team, and Enterprise plans all unlock all three tabs on every seat. The decision is which tab a given person uses, not which one they are licensed for.
The fourth question is the one mid-market companies miss. Buying once and routing per role is dramatically simpler than splitting procurement.
We see the same five misallocations repeatedly. None of them are catastrophic; all of them cost weeks.
"We bought Claude Code for the ops team." Wrong tab. The ops team needed Cowork. They downloaded the desktop app, opened Code, saw a terminal-style interface, and quietly stopped using it.
"Our finance team is using Chat to draft invoices." They are leaving roughly seventy percent of the value on the table. Chat can describe what an invoice should look like; Cowork generates the workbook with formulas, exports the PDF, and saves it where the rest of the team will find it.
"We built a custom in-house ChatGPT wrapper." Cowork already does this, with governance and an audit trail you would otherwise have to build. Custom wrappers usually surface the wrong problem — the team did not need a UI; they needed Cowork's file access.
"We tried to share Cowork outputs across the team via email." This is what the workspace pattern fixes — see The workspace pattern. One shared folder, three subfolders, every output discoverable.
"We asked the developers to evaluate it." Developers default to Code because that is the tab built for them. The people who would use Cowork most — finance, ops, marketing — never get to weigh in. Evaluate Cowork with operators, not engineers.
The leverage of buying one subscription across the org is that the tabs talk to each other through shared files and shared context.
An engineer in Code writes a feature spec. An operator in Cowork turns the spec into a customer-facing release note in the company's tone. Both reference the same CLAUDE.md so the wording is consistent.
A marketing operator in Cowork compiles a research brief. An executive in Chat asks follow-up questions on the brief without opening the source file. The brief is the artifact; the chat is the conversation about it.
This is the lever for buying once across the org instead of three procurement cycles.
Pro, Max, Team, and Enterprise plans all unlock Chat, Cowork, and Code on every seat. There is no separate Cowork SKU and no separate Code add-on inside any of these plans. The decision a buyer needs to make is per-seat and per-role, not per-product.
That means procurement is one decision: how many seats, on which plan tier. The plan tier governs usage envelope and admin features; the seat is bound to a person and unlocks all three tabs. Full plan breakdown is in Pricing & plans.
We map roles to tabs in week one of every engagement. The split usually lands around 70% Cowork seats / 20% Chat-mostly seats / 10% Code seats for a 50–500 person company. That is a useful prior, not a recipe — the right number for your team comes from a shadow week, not a spreadsheet.
The mistake we see most often is over-buying Code when Cowork would have served. Re-allocating seats during rollout is normal and cheap; specifying the wrong number for a year and discovering it in month four is neither.
A two-minute self-check: list the five most-touched documents on your team this past week. If three or more are Office, PDF, or image, Cowork is your answer. If three or more are source files in a repo, Code is. If they are all PowerPoint and Word with no programmatic component, you may genuinely only need Chat — and you should not over-buy.
Book a 30-minute call. We'll ask where you are, what your team needs, and which systems Cowork should touch.