⚡ Real Productivity

Claude writes a script that spins up to 1,000 agents for massive jobs, processing them very effectively in parallel.

When a task is huge and divisible, parallelizing cuts the total time. It’s the foundation for several features listed above.

Uses a lot of tokens; it’s new and still finding its place in everyday use.

Runs agents in ISOLATED copies of the repository so they never collide, then merges everything at the end.

Important when working on a team or on several features at once, so one agent doesn’t step on another’s work.

Isolation up front, merge at the end—safe parallelism.

Slash command that spins up a FLEET of agents in the CLOUD (like code review) to check every bug before reporting.

Fewer false positives: each finding is checked before it reaches you.

Runs in parallel in the cloud — it doesn’t block your terminal.

You ask for research and it automatically invokes a dynamic workflow, spreads searches across the web (and your apps), and returns ONE cited report.

Deep research without opening 30 tabs; sources are included.

Advanced Research on the same parallelism engine.

A sidebar that clicks, fills out forms, and acts across tabs; and Computer Use (point and click) to use any app when there’s no connector.

Extends Claude’s reach to everything in your browser or on your screen.

A connector when available; point and click as a universal fallback.

Projects (workspaces with chats, files, and instructions together), Memory (pulls in context from old chats), and /recap (shows what the session was doing when you come back).

It’s the everyday glue: reusable context and frictionless pickup, great when you have lots of tabs open.

More: Artifacts & Interactive Charts, Browser Recording, Claude for M365, Mobile Push, Cross-Device Convo and Connectors Directory.

Runs Gmail, Drive, Docs, Sheets, and Calendar through the TERMINAL; you can build skills around it.

Proves Claude can drive any CLI—your office workflow becomes automation.

Technically, it’s not an official Claude “feature,” but it’s so useful that it earns honorable mention #1.

Dispatch: start a task on your phone, and Claude runs it on your desktop and sends you the result. Channels: connect Telegram, Discord, or iMessage to a live session.

Good for continuing the conversation from your phone and talking to the session from wherever you are.

Channels works both ways: you send and receive.

Desktop App: a Mac/Windows home for chat, Cowork, connectors, and managing local and cloud routines in one place. Claude Design: describe what you want, and it creates polished designs/slides/one-pagers.

The desktop app is also better for copying and pasting text without the terminal's weird wrapping; Design is great for teams.

Design: comment, view design systems, and generate PPTs quickly.

Plan Mode: researches and proposes a plan before touching anything. Auto Mode: approves safe actions and blocks risky ones without constantly asking.

Planning ahead prevents rework; Auto Mode removes the friction of approving everything all the time.

Auto Mode is, in practice, a permission level.

Hooks: automatically run your own commands on session events. Effort Levels: adjust reasoning depth from low to maximum (/effort).

Automate the surrounding workflow and calibrate how much brainpower the task deserves.

Ultracode: the maximum-effort tier — reasoning at full throttle + automatic multi-agent workflows.

Agent View (dashboard of all terminals/agents running in the background and their status), /btw (a parallel question without interrupting the task) and /context (what's filling up the window).

You can see what’s running and how much it costs, and control it.

More: /usage, Prompt Caching (reuses repeated context and saves tokens), and /background.