Claude
At most ten steps, in order of how much they help. Each step is from Anthropic's own pages, linked. Re-checked against those pages every week; anything that can't be confirmed is marked unverified rather than removed.
- 01
Hand over the whole job, briefed like a new colleague
In the new experience you do not pick a mode first; just say what you want. If you still see two options, use chat for something that fits in a few exchanges and Cowork for a deliverable, such as a file, a deck or a spreadsheet. Say what the output is (format, structure, who will read it) and why it matters, and show an example if you have one. Add: "Before we begin, repeat my ask back to me so we're aligned, then ask me as many clarifying questions as you have." (one Claude, Cowork best practices, prompting best practices)
- 02
Write your standing instructions once
Put how you like to work under "Instructions for Claude" in Settings > General: your preferred methods, terms you use, and how you want answers and files laid out. They apply to every conversation, so you stop repeating yourself. In the new experience this is also where the old Cowork "Global instructions" now live. (Cowork, until your account moves over: Settings > Cowork, then "Edit" next to Global instructions.) Folder instructions add context for a local folder on desktop, and Claude can update them itself during a session (Cowork, until your account moves over). (personalisation, one Claude, get started with Cowork)
- 03
Give each ongoing piece of work its own project
A project keeps files, instructions and context together, with its own memory, so what Claude learns there stays there. In the new experience, projects work in every conversation. On claude.ai, go to claude.ai/projects, click "+ New Project", then "Set project instructions". Claude does not see the project name or description, so put anything it needs in the instructions. Add knowledge with "+" in the project knowledge panel, using clear file names. Free accounts can have up to five projects. (Cowork, until your account moves over: open Projects in the left navigation and choose +, then start from scratch, import a claude.ai project, or use an existing folder. These projects hold local folders and stay on your computer.) (create and manage projects, RAG for projects, Cowork projects guide, one Claude)
- 04
Give Claude rich context: folders, connected apps and memory
Anthropic says the gap between a mediocre and a great result is almost never the prompt, but whether Claude has enough context. Point Claude at whole folders rather than single files; local files and folders need Claude Desktop. Connect your apps under Customize > Connectors, pick one, select "Connect to Claude" and sign in; "+" then "Connectors" turns one on or off for a single chat. Check what memory holds in Settings > Memory, and ask Claude to find a past chat when you need it (chat search is on Pro, Max, Team and Enterprise). For questions that need many sources, type
/deep-researchor click "+" then "Research". (Cowork best practices, get started with connectors, chat search and memory, using Research, one Claude) - 05
Ask for the finished thing, not a description of it
Ask for a document, spreadsheet, deck, design, dashboard or small tool, and Claude hands back the file. In chat you can also choose "Output" in the message box to start from a template (a beta on paid plans). An artifact opens beside the chat, where you ask for changes or edit directly. Artifacts start private; "Share" sets who can open one. "Unpublish" cannot be undone and deletes the artifact's stored data. (one Claude, what are artifacts, share artifacts)
- 06
Let long tasks run while you do something else
Bigger tasks keep running in the cloud if you close your laptop or leave the page. Tasks that use files or apps on your computer need Claude Desktop open. A task you start at your desk shows up in Claude Mobile, where you can check progress, answer Claude's questions and redirect it. On Pro and Max, from 6 October 2026, new Cowork tasks run in the cloud and the "Only on your computer" option in Settings > General goes; tasks already started on your computer stay there. Longer tasks use more of your plan's limits than quick questions, so group related work into one task and check Settings > Usage. (one Claude, scheduled tasks, get started with Cowork, usage limit best practices)
- 07
Give Claude a working folder and keep untrusted content away from real actions
Claude can read the files you grant it and act for you. Use a dedicated working folder, keep financial documents, credentials and personal records out of reach, and keep backups. Claude asks for explicit permission before permanently deleting a file. Web content is a main route for prompt injection, so only give Claude internet access to sites you trust, and be deliberate about which tabs are open in Claude in Chrome. Before turning on computer use, block sensitive apps such as banking or health apps, and start with low-stakes tasks. (use Cowork safely)
- 08
Match the permission mode to the stakes
The mode selector in the chat box offers Manual (Claude asks before acting), Auto (Claude checks actions for safety risks and pauses only if something looks unsafe) and Skip (no asking and no safety checks). Auto uses more of your limits than the other modes. For work with real consequences, such as money, messages sent as you or important files, stay close and review, or switch back to manual approval. Use "Allow once" for a connector tool until you trust it. (get started with Cowork, get started with connectors)
- 09
Schedule one weekly job you already know well
In the new experience, describe the task and how often it should run, for example "Every Monday at 9 AM, summarise last week's messages in my team's Slack channels." (Cowork, until your account moves over: click Scheduled in the left sidebar, then New task.) Scheduled tasks are on all paid plans and keep running when your computer is asleep. Do not schedule tasks that touch sensitive files, send messages as you, buy things or are hard to undo. Since 1 October 2026, scheduled tasks Claude creates for you default to "Automatically approve" where your organisation allows it, so check the approval mode. (scheduled tasks, use Cowork safely, changelog)
- 10
Turn repeated work into skills and plugins, after reading them
Skills are saved instructions (and scripts) Claude loads when a request matches their one-line description, so that description matters most. Switch them on in Customize > Skills, or type
/in the message box to pick one. They need a paid plan and "Code execution and file creation" on in Settings > Capabilities. Read a shared or uploaded skill's files before turning it on, as Anthropic does not review them. Installing a plugin does not connect its connectors; do that from the plugin's Connectors tab. (skills overview, plugins guide, Cowork overview)
Unverified
What Anthropic's own pages disagree on, or could not be confirmed.
- The rollout reaches accounts on the same plan at different times. The "one Claude" page names Pro and Max first and says more plans will follow; it does not say when Free or Team get it. Enterprise admins get at least 30 days' notice. Once your account moves over, you cannot switch back.
- In the new experience these are not available yet: GitHub integration ("Add from GitHub"), branching a conversation from an earlier point, search across older Cowork tasks, and Dispatch for new users. Use Dispatch only if you already have it. (Dispatch guide)
- It is not clear how Cowork projects (local, with folders) and claude.ai projects fit together in the new experience. The "one Claude" page only says projects work in every conversation. Step 3 gives both paths.
- The scheduled tasks page says both that a task needing local files or apps "will only run locally" and that scheduled tasks "can't be tied to a folder on your computer". Its manual setup also lists an optional folder. Step 9 does not rely on either.
- The
/schedulecommand was in the old Cowork playbook but is not on the scheduled tasks page as read on 2026-10-04, so it is left out. - The 1 October 2026 "Automatically approve" default applies to scheduled tasks that Claude creates for you, per the changelog. Whether it also applies to tasks you set up by hand is not stated.
- Importing and exporting memory to and from other AI services is described as experimental on the memory page, but its own article was not read, so it is not a step here.
- The usage limits page says cross-conversation memory is for paid plans, while the memory page says memory is on Free too. Step 4 follows the memory page.