One desktop app. A spectrum of help — from thinking alongside you, to handing off whole projects.
Tip — click any part to jump straight to it.
Through earlier AI training, you’ve already been working hands-on with AI tools — so a lot of today will feel familiar. Here’s how the pieces fit together.
You’ve already built work plans, decks, and dashboards in AI tools. Those instincts carry straight over.
Get comfortable with Chat and Cowork and the key features within them — and how they compound into a way of working.
Practice-specific offerings — built for Integration, Diligence, and Divestiture — are in development. More to come.
The difference isn't intelligence — it's how much you hand off. As you move from one to the other, you give Claude more room to act on its own, with your approval at the gates.
You drive. Claude thinks, drafts, and builds with you, turn by turn.
You delegate. Claude plans and executes multi-step knowledge work, you approve.
You can get a long way just by chatting — but it goes further when you treat this as a way of working, where context, skills, and automation compound. We’ll build to that.
The conversational surface at claude.ai and in the desktop app — where you and Claude break problems down together.
Chat is the everyday surface for writing, learning, research, and analysis. You work turn by turn: Claude expands on your logic, simplifies complexity, and turns a back-and-forth into finished work.
Generate interactive visualizations, tools, and reports in a side panel you can iterate on live.
Persistent workspaces that keep context, files, and instructions together across many sessions.
Upload documents and images for Claude to read, analyze, and reason over directly.
Hands-free, spoken conversation for working on the go or thinking out loud.
Pull in current information and cite sources without leaving the conversation.
Connect OneDrive, SharePoint, and other tools to ground answers in your own context.
↪ More on connectors in the Cowork section.
A Project is a persistent workspace in Claude Chat — it holds your files, your instructions, and the thread of past conversations, so every chat inside it starts already knowing the context. The same instinct as a primed Cowork folder, on the Chat side.
Upload the documents the work draws on once — every conversation in the Project can read them.
Standing guidance Claude follows across the whole Project — tone, format, and what to avoid.
Past chats stay with the Project, so you pick up where you left off instead of re-explaining.
Share the Project with teammates so everyone works from the same files, instructions, and history.
↪ Why a Project, not Cowork? In a Project, Claude reads your files to think with you — it doesn’t change them. When the files themselves need to change, that’s Cowork (more in the Cowork section).
Reach for a Project whenever you’ll return to the same body of work more than once.
“Help me think it through — and make it with me.”
Best when you want to stay in the loop on every step.
Hand off whole tasks. Claude makes a plan, gets your approval, and does the tedious knowledge work for you.
Cowork is the autonomous mode in the desktop app for knowledge work. You point it at a folder — your engagement workspace — then describe an outcome; Claude works with the files in that folder and your connected apps to complete the whole job — checking in on the steps that matter.
Point Cowork at a folder on your computer and it works inside it — reading every file for context, then creating and editing files right there. Chat keeps its work inside the conversation; Cowork works on the real files in your folder, on your machine.
It pulls context from every file in the workspace — no copy-paste, no re-explaining the deal each time.
It drafts new files, edits existing ones, and organizes them — saved straight into your folder, not pasted back as text.
Chat works inside the conversation — even the artifacts it builds live there, apart from your files. Cowork edits and creates the actual files in your workspace.
Tell Claude what you want in plain language — “sweep these contracts for change-of-control clauses,” “build the weekly status report.”
It shows you the steps and the access it needs before it starts. You decide what it can touch, and whether it checks in along the way.
It carries out the work autonomously, keeping you informed of significant actions along the way.
Set the posture per task — “act before asking” to approve each step, or “act without asking” once you trust it. Either way, you decide which folders and connectors Claude can touch.
Recurring daily, weekly, or monthly workflows that run on their own — like a morning email digest.
Sorts, renames, and categorizes files using sensible, consistent conventions.
Pulls details from screenshots, receipts, and invoices into clean, formatted spreadsheets.
Produces branded documents, decks, and reports from your source materials.
Watches inboxes and metrics, then summarizes what changed into a digest.
Compiles and analyzes data into repeatable, template-based outputs.
Connectors let Claude reach into your real systems — SharePoint, OneDrive, Outlook, Teams — using your existing permissions. It's how Cowork works with your actual files and mail, not just what you upload.
Search sites and libraries; read the documents you already have access to.
Search mail and calendar — pull threads and find availability.
Search chat messages and conversations for context.
Read-only and permission-respecting — Claude only sees what your organization already lets you see. It's also what lets you point Cowork at your team's shared folder.
Any job you can describe, Cowork can put on a schedule — recurring or one-time. The tell: if doing it once right now wouldn’t satisfy the request, schedule it.
“Every weekday at 7am, brief me — what’s on my calendar, the unread mail that matters, and what needs my attention across the deal.”
“Each Friday at 3pm, compile the workstream trackers into a draft of the steering-committee update.”
“Each morning, check the data room for new uploads and log anything that touches the TSA scope.”
“Tomorrow at 9am, remind me to chase the Tax lead on the German entity setup.” Not everything recurs.
Each run is a normal Cowork session — same plan, same approvals, just on a clock.
A live artifact is a dashboard, tracker, or brief Claude builds that you open and keep — it pulls the latest each time you open it, instead of going stale the moment you export it.
A deal or workstream view that pulls the latest each time you open it.
An entity or issue list that re-queries the workspace each time you open it.
A summary you re-run on demand instead of rewriting from scratch.
“Just take care of it — and show me before anything risky.”
Best when you'd rather delegate than do.
| Claude Chat | Claude Cowork | |
|---|---|---|
| Core idea | Think & make with you | Delegate knowledge work |
| You provide | A conversation | An outcome to hand off |
| Autonomy | Turn by turn | Plan → approve → execute |
| Works with | Files, web, connectors, artifacts | Local files & cloud apps |
| Lives in | Web & desktop app | Desktop app (+ mobile pairing) |
| Reach for it when | You want a thinking partner | The task is clear but tedious |
Both a Chat Project and a Cowork folder are persistent workspaces that hold your files and context — so what's actually different? It comes down to what the work is: the conversation, or the files.
Files are loaded in as reference; Claude reasons over them and hands finished work back as artifacts in the thread. State lives in the conversation — read-mostly, and relatively static.
Like a briefed analyst in the meeting room — they've read the data room and hand you drafts, but the room doesn't change.
Claude reads, writes, and edits files in place on disk. The folder is the durable state, not the chat — read-write, and it scales to far more files, since they're pulled selectively rather than all held in context.
Like an analyst at the deal folder on the shared drive — actually rebuilding the workbook; the folder itself is what's different when they're done.
The tell — at the end of the session, what should have changed: the answer, or the files?
Both share the same models — Opus 4.8, Sonnet 4.6, Haiku 4.5 — and the same permission-first safety posture.
Beneath both modes is a layer you can shape — teaching Claude your context, your procedures, and the tools your team shares. The same idea runs across Chat and Cowork alike.
Instructions and skills — the two layers you can shape, both plain files Claude reads.
A standing-instructions file Claude reads every session — your conventions, environment, and preferences, kept right next to the work. In the desktop app, project instructions play the same role.
Packaged know-how — instructions plus optional scripts — Claude loads only when a task calls for it, so a specialized procedure (a steerco pack, your house TSA format) runs the same way every time. The built-in skill-creator builds one from a procedure you describe.
Already have skills built in another tool? Add them with Add skill — see “Where do I find skills?” below.
Both live right in Cowork — and neither is hand-crafted: describe what you want, and Claude writes them for you.
CLAUDE.md, your context, your skills — these aren’t settings buried in an app. They’re plain files in your engagement folder. Point Cowork at it and every layer above is in play at once.
One way to lay it out — not a required structure. Keep what’s useful and skip the rest; you don’t need a background, data-room, working, and deliverables folder for every workspace. Let the work decide the shape.
Get in the habit — when something’s worth remembering, ask Claude to write it to a .md and keep it in the folder.
Two boundaries keep the work clean — and keep Claude sharp. Scope a Cowork workspace to a deal, and each chat to a task. It’s the same instinct you already use for any folder on your computer: one engagement’s files belong together, and unrelated work doesn’t get buried in with them.
↪ Before you start fresh, capture what matters to a .md in the folder — the folder is the memory that carries across chats; the chat is just the working space.
“Teach Claude once — reuse it everywhere.”
You already have the desktop app — Chat and Cowork live in one place. Switch between them as the moment calls for it.
Pick the deal closest to your world, pair up, and run a short hands-on tour of Cowork together — out of a workspace that already knows the deal.
Three workspaces, the same seven-step tour. Pick one with your group — you’ll all work out of that one folder. Click a tile to load that scenario’s guided steps.
Sell-side separation — Meridian is carving out its Flow Control division; the advisor runs the SMO. You build a TSA workstream status one-pager.
Buy-side integration — Northwind is folding in Lattice Analytics; the advisor runs the IMO. You build a Day-1 readiness one-pager.
Buy-side HC diligence — Stonehaven is acquiring Northstar Advisors. You build an HC findings one-pager.
One person per team sets up the shared folder; everyone else syncs into that same copy. Follow it in order.
Open your scenario’s folder, then click Download in the toolbar to save it to your computer.
Open project-tide ↗It arrives as a .zip — unzip it first so you have a normal folder.
Windows: right-click → Extract All · Mac: double-click to unzip.Move that unzipped folder into your own OneDrive — now you control it.
Keep it in OneDrive — not Downloads, Documents, or Desktop.In OneDrive, right-click the folder → Share and invite each teammate by name.
Accept the invite, then open the shared folder in OneDrive.
Hit Sync to pull the folder onto your machine.
Don’t see Sync? Click the ⋯ More menu, then choose Sync.Now everyone opens that one synced folder in Claude Cowork — same files, same instance. Take turns driving.
Whichever scenario you picked, the folder is the same shape — pre-built, so Cowork is current the moment you open it.
Engagement identity and working style — who Claude is on this deal, and the quality floor it holds to.
The primed context: the deal, the stakeholders, and the workplan.
Raw inputs — term sheets, excerpts, exports, and meeting transcripts.
The house format your one-pager follows — so you match a standard, not invent one.
Empty — the client-ready work you produce lands here.
Scratch space for drafts in progress.
Ask Cowork to brief you on the deal, the players, and what’s open.
Draft your one-pager to the house format. Watch it plan, then approve.
Turn what you just did into a reusable skill. (swap drivers here)
Run the skill on a second workstream — same format, new inputs.
Set a daily digest of what’s open and at risk.
Turn your one-pager into a live, updating dashboard.
Capture where it stands so a teammate — or a fresh session — can pick it up cold.
A new task is a new chat — that keeps Claude’s context clean (don’t run all seven in one). The one pair to keep in a single chat is Build → Skillify — the skill is made from the one-pager you just built.
EXERCISES.html in your folder — the exact prompt for each step is there. Start with: “Read this workspace and brief me — the deal, the key players, and what’s open.”
A crisp brief — the deal, the players, the open items and risks — pulled straight from the folder, not from you.
Notice it never asked you to explain the deal. That's the workspace doing the work.
Context you capture once pays off every session after. A primed folder means you — and your teammates — never start cold again.
A one-page draft that matches the house format, every non-obvious claim tagged to a source.
It works the ask end to end and pauses for your okay along the way — you're steering the judgment, not proofreading.
Describe the outcome, not the keystrokes — and notice you never named a format. The workspace already points Claude at the house sample, so the first draft comes back in your style, consistent and sourced. Not this one layout — hand it any example and it adopts that one. Your time goes to judgment, not assembling it.
A short, reusable skill — a name, what it captures (the format + sourcing rules), when to use it — saved to the workspace.
It should capture the format, not this one deal's facts — so it works on any workstream.
This is the leap past chatting: you're keeping the method itself, so the next one-pager starts from it — and sharing it puts your whole team on one setup. Teach Claude once; reuse it everywhere.
The same one-pager, same shape, for a new workstream — produced in a single step.
Same format, new inputs — the second time should feel instant. Don't see your skill in the / list? The teammate who built it has to Save and Share it first.
Every skill you build makes the next task faster. The returns stack across the whole engagement.
A standing task set to run on a cadence — a morning digest of what's open and at risk.
You're learning the setup — it won’t fire overnight here. Note how little you had to specify. Then open the task and hit Run now once to watch it work — approve each step as “always allow,” so the scheduled run won’t stall waiting on you.
Move from doing the recurring work to delegating it. Cowork keeps working when you're not — your time goes to judgment, not the weekly pull-together.
A dashboard that opens as a living view and pulls current data each time you open it.
It's not a static export — each time you open it, it pulls current data from the workspace on its own, no re-prompting. The view reflects today, not the day you built it.
Outputs that refresh on open beat documents that go stale the moment you save them. Build once, current every time you open it.
A tight handoff — status, open items, decisions made, and what to watch — enough that someone cold can carry on.
Prove it: open a fresh session, give it the handoff, and ask it to keep going — it shouldn’t need the deal re-explained.
Context is something you capture on purpose, not something that vanishes when the chat closes. A good handoff is how work survives people and sessions.
A crisp brief — the deal, the players, the open items and risks — pulled straight from the folder, not from you.
Notice it never asked you to explain the deal. That's the workspace doing the work.
Context you capture once pays off every session after. A primed folder means you — and your teammates — never start cold again.
A one-page draft that matches the house format, every non-obvious claim tagged to a source.
It works the ask end to end and pauses for your okay along the way — you're steering the judgment, not proofreading.
Describe the outcome, not the keystrokes — and notice you never named a format. The workspace already points Claude at the house sample, so the first draft comes back in your style, consistent and sourced. Not this one layout — hand it any example and it adopts that one. Your time goes to judgment, not assembling it.
A short, reusable skill — a name, what it captures (the format + sourcing rules), when to use it — saved to the workspace.
It should capture the format, not this one deal's facts — so it works on any workstream.
This is the leap past chatting: you're keeping the method itself, so the next one-pager starts from it — and sharing it puts your whole team on one setup. Teach Claude once; reuse it everywhere.
The same one-pager, same shape, for a new workstream — produced in a single step.
Same format, new inputs — the second time should feel instant. Don't see your skill in the / list? The teammate who built it has to Save and Share it first.
Every skill you build makes the next task faster. The returns stack across the whole engagement.
A standing task set to run on a cadence — a morning digest of what's open and at risk.
You're learning the setup — it won’t fire overnight here. Note how little you had to specify. Then open the task and hit Run now once to watch it work — approve each step as “always allow,” so the scheduled run won’t stall waiting on you.
Move from doing the recurring work to delegating it. Cowork keeps working when you're not — your time goes to judgment, not the weekly pull-together.
A dashboard that opens as a living view and pulls current data each time you open it.
It's not a static export — each time you open it, it pulls current data from the workspace on its own, no re-prompting. The view reflects today, not the day you built it.
Outputs that refresh on open beat documents that go stale the moment you save them. Build once, current every time you open it.
A tight handoff — status, open items, decisions made, and what to watch — enough that someone cold can carry on.
Prove it: open a fresh session, give it the handoff, and ask it to keep going — it shouldn’t need the deal re-explained.
Context is something you capture on purpose, not something that vanishes when the chat closes. A good handoff is how work survives people and sessions.
A crisp brief — the deal, the players, the open items and risks — pulled straight from the folder, not from you.
Notice it never asked you to explain the deal. That's the workspace doing the work.
Context you capture once pays off every session after. A primed folder means you — and your teammates — never start cold again.
A one-page draft that matches the house format, every non-obvious claim tagged to a source.
It works the ask end to end and pauses for your okay along the way — you're steering the judgment, not proofreading.
Describe the outcome, not the keystrokes — and notice you never named a format. The workspace already points Claude at the house sample, so the first draft comes back in your style, consistent and sourced. Not this one layout — hand it any example and it adopts that one. Your time goes to judgment, not assembling it.
A short, reusable skill — a name, what it captures (the format + sourcing rules), when to use it — saved to the workspace.
It should capture the format, not this one deal's facts — so it works on any workstream.
This is the leap past chatting: you're keeping the method itself, so the next one-pager starts from it — and sharing it puts your whole team on one setup. Teach Claude once; reuse it everywhere.
The same one-pager, same shape, for a new workstream — produced in a single step.
Same format, new inputs — the second time should feel instant. Don't see your skill in the / list? The teammate who built it has to Save and Share it first.
Every skill you build makes the next task faster. The returns stack across the whole engagement.
A standing task set to run on a cadence — a morning digest of what's open and at risk.
You're learning the setup — it won’t fire overnight here. Note how little you had to specify. Then open the task and hit Run now once to watch it work — approve each step as “always allow,” so the scheduled run won’t stall waiting on you.
Move from doing the recurring work to delegating it. Cowork keeps working when you're not — your time goes to judgment, not the weekly pull-together.
A dashboard that opens as a living view and pulls current data each time you open it.
It's not a static export — each time you open it, it pulls current data from the workspace on its own, no re-prompting. The view reflects today, not the day you built it.
Outputs that refresh on open beat documents that go stale the moment you save them. Build once, current every time you open it.
A tight handoff — status, open items, decisions made, and what to watch — enough that someone cold can carry on.
Prove it: open a fresh session, give it the handoff, and ask it to keep going — it shouldn’t need the deal re-explained.
Context is something you capture on purpose, not something that vanishes when the chat closes. A good handoff is how work survives people and sessions.
You can get a lot done just by chatting — and that’s genuinely useful. But the practitioners who optimize how they work get further: capture the context once, package what you do into skills, and let the routine run itself.
Capture the deal once; every future session starts current instead of cold.
Save the procedure once; each skill makes the next task faster.
Delegate the recurring work — what’s left is the part that needs you.
Small investments in how you work pay off every week after — and the way you work carries into everything you take on next.
Everything you did today — the primed workspace, the skills, the scheduled digest — can be bundled into a plugin: one install that sets a whole team up the same way. Anthropic ships plugins today, like finance and legal.
Practice plugins are in development. More to come soon!
Three things new users ask first — here’s exactly where they live in the desktop app.
One Microsoft 365 connector covers SharePoint, OneDrive, Outlook and Teams — read-only, within the permissions you already have. (The + in the message box has a Connectors shortcut too.)
Skills you create or add show up right here in the + menu — no menu to hunt through.
This is the whole idea — point Cowork at a primed folder and every layer above is in play the moment it opens.