Claude Code's Hidden Features That Change Everything

Most developers using Claude Code are using maybe 20% of what it can actually do. That's not a criticism—it's just how software works. The powerful stuff is usually buried three menus deep or hidden behind flags nobody mentions in the docs.

Boris Cherny, who built Claude Code, just published a thread walking through fifteen features he uses constantly that most people don't even know exist. These aren't minor quality-of-life improvements. They're the difference between using Claude as a chatbot that writes code and running it like an actual distributed team.

I went through each one to see what actually matters.

The Mobile Trap

The first tip sounds basic: download the mobile app. But Cherny isn't really talking about the app itself. He's talking about the workflow shift that happens when you can kick off a task, walk away, and check in later from wherever you are.

This matters more when you stack it with the other features—scheduled tasks that run overnight, dispatch commands you send while commuting, parallel sessions you monitor between meetings. The mobile app isn't the feature. It's the infrastructure for treating Claude less like a tool you sit down to use and more like something that's just always working.

Session Portability Actually Works

Claude lets you move active sessions between the CLI, desktop app, web interface, and mobile. Same conversation, same context, different environment. You start something locally, hand it off to the web interface to monitor progress, pull it back to the terminal when you need to intervene.

The video demonstrates this with remote control—enable it once, and every session becomes available across platforms. The desktop app updates in real-time as you work in the CLI, though with one quirk: if you're actively viewing a session, you need to click away and back to see updates. Small thing, but worth knowing before you waste time wondering why nothing's changing.

What's interesting here is the organizational benefit. When you're running multiple sessions—and the later features make that not just possible but practical—having them visible in a GUI instead of juggling terminal tabs changes how you think about task management.

Scheduled Tasks and Hooks: Where It Gets Serious

Scheduled tasks let you define a prompt once and have Claude run it automatically on whatever interval you want. Hourly file audits, daily report generation, overnight batch processing—anything repetitive that you'd normally either automate with a script or just do manually forever.

The syntax is straightforward: /loop 5m "tell me what files exist in this repo" runs that check every five minutes. Simple example, but scale it to actual work—monitoring for dependency updates, running security checks, updating documentation—and it starts feeling less like a feature and more like infrastructure.

Hooks take this further by letting you inject your own code into Claude's workflow. Want tests to run automatically every time Claude writes code? Set up a post-tool-use hook. The video shows Cherny just asking Claude to help configure the hook itself, which is a nice meta-loop: use the AI to automate the AI.

"Instead of relying on Claude to remember to do things, you can just actually enforce them using hooks," the video explains. That shift from trusting the model to remember to making it structurally impossible to forget is significant.

Dispatch: The Remote Control Problem

Dispatch is Cherny's term for secure remote control of Claude on your computer. You're away from your desk, you send Claude a task from your phone, it executes on your machine. File management, email processing, Slack catchup—basically anything you'd normally do sitting there.

The security implications here are obvious and concerning. You're giving an AI agent control of your computer, accessible from your phone, potentially running 24/7. The video shows enabling a "keep this computer awake" setting, which is required for dispatch to work but also means your machine is just... always on, always accessible, always executable.

Cherny uses this regularly. Whether you should depends entirely on your threat model and how much you trust the security architecture Anthropic has built. The feature exists, it's powerful, and it requires you to think seriously about what you're comfortable with.

The Chrome Extension and Verification Loop

For web development, Claude needs to see what it built. The Chrome extension gives it that ability—it can view pages, take screenshots, read DOM content, iterate on its own work until things look right.

"If you asked an engineer to build a website but never let them open a browser, the end result probably wouldn't be great," the video argues. Fair point. The extension closes that feedback loop.

The desktop app takes this further by automatically running and testing dev servers. Claude starts the server, opens the app, checks the result, makes changes, repeats. The entire verify-and-iterate cycle runs without you.

This is where AI coding tools start feeling genuinely different from traditional development. It's not just generating code—it's operating in the same workflow loop human developers use, with the same ability to check its work and course-correct.

Worktrees: How to Run Dozens of Sessions

Git worktrees let you create separate working environments from the same repository. Multiple isolated workspaces, same project. This is what allows Cherny to run "dozens" of Claude sessions simultaneously without them interfering with each other.

The video shows a practical example: maintaining a stable benchmark suite in the main branch while experimenting with new benchmarks in isolated worktrees. You can spin up a worktree with a simple flag: claude --worktree greenfield-benchmark.

Combine this with dispatch and parallel sessions, and you're no longer working linearly. You're coordinating multiple Claude instances across the same project at the same time. "This is how you go from using Claude as a tool to actually running it like a team," the video explains.

The /batch command automates this pattern—give it a large changeset, and it fans the work out across as many Claude agents as needed. The demo shows five background agents launching simultaneously to handle a simple task. Scale that to actual refactoring work or large migrations, and the time savings become substantial.

The Smaller Things That Matter

Some features are just quality-of-life improvements that add up:

• /branch forks your session to test something without losing current progress
• /btw lets you ask quick questions without interrupting long-running tasks
• --bare speeds up startup by skipping default initialization
• --add-dir lets you work across multiple repositories in one session
• --agent creates custom agent configurations for specific workflows
• /voice enables voice input for when typing is inconvenient

None of these individually transform how you work. Collectively, they remove friction points that would otherwise slow you down or force workarounds.

What This Actually Means

The gap between "I use Claude Code" and "I use Claude Code the way Boris Cherny uses Claude Code" is enormous. Most developers are using it as a smarter autocomplete. Cherny is using it as distributed infrastructure.

That gap isn't just about knowing more commands. It's about reconceptualizing what the tool is for. Scheduled tasks, hooks, worktrees, batch execution, remote dispatch—these features assume you're orchestrating ongoing work, not just generating code snippets.

Whether that's the right model depends on what you're building and how you work. But it's worth understanding what's possible before deciding what's useful. The features exist. Most people don't know about them. Now you do.

Marcus Chen-Ramirez is a senior technology correspondent for Buzzrag, covering AI, software development, and the intersection of technology and society.