You know that feeling when you realize you've been using 10% of a tool's capabilities for months? Like driving a sports car exclusively in first gear?
That's what happened to a lot of developers this weekend when Boris Cherny, the person who actually created Claude Code, shared his 15 favorite "hidden and under-utilized features." The thread racked up 2.3 million views in under 24 hours, and the most common reply was some version of: "Wait, it can do THAT?"
Here's the thing: Claude Code has quietly grown from a terminal coding assistant into something closer to an autonomous development platform. But most people, even daily users, are only scratching the surface. Boris' thread is basically a cheat sheet from the person who built the thing.
We broke down every tip, explained what it actually does (in plain English), and added a beginner's guide so you can start using these features today, whether you're a seasoned developer or someone who just learned what a "terminal" is last month.
First up, the TL;DR
Boris Cherny, the creator of Claude Code (Anthropic's AI coding tool that lives in your terminal), just shared 15 features he uses daily that most people have never touched. The thread hit 2.3 million views in a day. His message: Claude Code has grown into a full development platform.
Here's what happened:
- Boris revealed he writes most of his code from his phone using Claude Code's mobile app, and moves sessions between devices with a single command (
/teleport) - He runs dozens of automated Claude agents simultaneously using git worktrees (separate copies of his code so agents don't step on each other), with tasks like auto-reviewing pull requests every 5 minutes
- Two features,
/loopand/schedule, let you set Claude to run tasks on repeat for up to a week, turning it into an always-on coding assistant - Hooks (custom triggers that fire during Claude's workflow) let developers auto-format code, log commands, or even route permission requests to WhatsApp
- Cowork Dispatch extends Claude Code's power to non-coding tasks; Boris uses it daily to catch up on Slack and manage files from his phone
Why this matters: Claude Code contributed to roughly 4% of all public GitHub commits by early 2026, about 135,000 commits per day. These power features explain how: they turn a single developer into something closer to a team. If you've been using Claude Code as a fancy autocomplete, you've been leaving the best parts on the table.
Our take: The gap between "using Claude Code" and "using Claude Code well" is massive, and it's widening. Boris' thread reads like a preview of where all AI tools are headed: less typing, more delegating. The developers who figure out loops, worktrees, and hooks first will ship faster than teams twice their size. And yes, coding from your phone while waiting for coffee is now a real workflow. The future is weird.
The Basics: What Even Is Claude Code?
Before we dive into Boris' power user tips, let's get everyone on the same page.
Claude Code is Anthropic's AI coding assistant. Unlike ChatGPT or regular Claude (which live in your browser), Claude Code runs directly inside your terminal (the text-based interface developers use to interact with their computer; think of it as the command line where you type instructions instead of clicking buttons).
Here's what makes it different from other AI pre-coding agent coding tools:
- It reads your entire codebase. Claude Code sees every file in your project and understands how they all connect to each other, which means it can make changes across multiple files at once without breaking things.
- It runs on your machine. Your code stays local. Claude reads and edits files on your computer, not in some remote server (note: it's not private though; Claude still reads and processes your code over the cloud).
- It asks permission. Every time Claude wants to change a file or run a command, it asks you first. You stay in control (depending on the setting!).
- It speaks plain English. You don't need to write code to use it. You can say "add a login page to my app" and Claude figures out the rest (this is especially useful for those tricky terminal commands you never learned, or at least never memorized and always forget).
You can use Claude Code in the terminal, in VS Code or JetBrains (popular code editors), as a desktop app, on the web, or even from your phone. It costs $20/month on the Pro plan, with heavier usage tiers at $100 and $200/month ($100 is the sweet spot!).
One key concept you'll see throughout this article: CLAUDE.md. This is a simple text file you put in your project folder that tells Claude how your project works, what your coding style is, what rules to follow. Think of it as onboarding documentation for your AI teammate. Every time you start Claude Code, it reads this file automatically.
Got it? Good. Now let's get into Boris' tips.
The 15 Features, Explained
1. Code From Your Phone
Boris says he writes "a lot of his code" from the iOS app. That's not an exaggeration for effect; Claude Code has a full mobile experience. Open the Claude app on iOS or Android, tap the Code tab, and you're in a coding session.
Why this matters for everyone: You don't need to be at your desk to fix a bug, review code, or kick off a task. Waiting for your kid at soccer practice? You can tell Claude to refactor that messy function. It's not ideal for marathon coding sessions (small screens are small screens), but for quick fixes and monitoring, it's a game-changer.
2. Teleport and Remote Control: Move Sessions Between Devices
Here's where it gets interesting. Say you start a coding session on your laptop at the office. You need to leave, but Claude's in the middle of something. Two options:
/teleport(orclaude --teleport): Takes a session running in the cloud (on the web or mobile) and pulls it onto your local machine. Think of it like picking up a phone call on a different device./remote-control: The reverse. Control a session running on your local machine from your phone or any browser. Your code never leaves your computer; only the chat messages travel through an encrypted connection.
Boris keeps "Enable Remote Control for all sessions" turned on permanently, so he can always check in from his phone.
The non-technical version: Imagine writing a document on your office computer, then seamlessly continuing it on your phone during your commute, then picking it back up on your home laptop. Same document, same place where you left off, any device.
3. /loop and /schedule: Your AI Runs on Autopilot
Boris calls these "two of the most powerful features in Claude Code," and his examples show why. Here's the difference: /loop repeats a task at a set interval locally on your machine (for up to 3 days, with a max of 50 concurrent loops). /schedule does the same thing but runs in the cloud on Anthropic's infrastructure, so it keeps working even when your computer is off.
Here are Boris' actual loops running at all times:
- Every 5 minutes: Claude checks his pull requests (proposed code changes), addresses reviewer feedback, rebases (syncs his code with the latest version), and pushes them toward production. Automatically.
- Every 30 minutes: Claude looks at Slack feedback and creates pull requests to address it.
- After every merge: Claude sweeps for review comments he missed and files fixes.
- Every hour: Claude closes stale pull requests that are no longer needed.
The non-technical version: It's like setting a recurring calendar reminder, except instead of reminding you to do something, Claude actually does the thing. Every 5 minutes, every 30 minutes, every hour. You sleep; Claude works.
4. Hooks: Custom Triggers for Everything
Hooks are custom scripts (small programs) that run automatically at specific points during Claude's workflow. Boris highlighted four examples:
- SessionStart: Load fresh context every time you start Claude. Like having your assistant read the morning briefing before you walk in.
- PreToolUse: Log every command Claude runs. An audit trail of everything the AI does.
- PermissionRequest: Route approval requests to WhatsApp. Claude wants to delete a file? You get a text asking if that's OK.
- Stop: Poke Claude to keep going when it stops. Like a "hey, you're not done yet" nudge.
The hook system supports three types: command hooks (run a script), HTTP hooks (send data to a web service), and prompt hooks (modify Claude's behavior with instructions). They fire at events throughout Claude's lifecycle, so you can customize nearly every step of the process.
The non-technical version: Hooks are "if this, then that" rules for your AI assistant. If Claude edits a file, automatically format it. If Claude finishes a task, send you a Slack message. You set the rules once; they run forever.
5. Cowork Dispatch: Claude Code for Non-Coding Tasks
Boris says when he's not coding, he's "dispatching." Cowork Dispatch is essentially Claude Code's power applied to regular computer tasks. It's a secure remote control for the Claude Desktop app that can use your apps, browser, and computer.
Boris uses it to catch up on Slack, manage emails, organize files, and handle tasks on his laptop when he's away from his desk. You describe what you want done, and Claude does it, checking in with you for approval on anything sensitive.
The non-technical version: It's like having a virtual assistant sitting at your computer. "Organize my Downloads folder." "Pull my weekly metrics into a report." "Summarize my unread Slack messages." You can even schedule recurring tasks: "Do this every Friday."
6. Chrome Extension: Give Claude Eyes
Boris' tip here contains what he considers the single most important Claude Code principle: give Claude a way to verify its output.
The Chrome extension connects Claude Code to your browser so it can actually see what the website looks like, read console errors, and test interactions. Boris says he uses it every time he works on web code, and notes it "tends to work more reliably than other similar MCPs" (MCP stands for Model Context Protocol, basically a standard way for AI tools to connect to external services). Without it, asking Claude to build a website is like asking someone to paint a room while blindfolded.
Boris put it this way: think of Claude like any other engineer. If you ask someone to build a website but they can't use a browser, the result probably won't look good. Give them a browser, and they'll write code and iterate until it looks right.
Why this matters: Claude can now write code, open the browser, see that the button is in the wrong place, fix it, check again, and repeat until it looks right. Before the extension, you had to be the eyes. Now Claude can iterate on its own.
7. Desktop App: Built-In Browser for Testing
Similar to the Chrome extension, the Claude Code Desktop app bundles a built-in browser that can automatically start your web server and test it. Claude writes the code, spins up the server, opens the browser, and checks if everything works.
The non-technical version: Instead of building something, handing it to you, and asking "does this look right?", Claude builds it, checks it, fixes it, and shows you the finished product.
8. Fork Your Session
Ever been deep into a conversation with Claude and wanted to try a different approach without losing your place? /branch creates a copy of your current session. You can explore one direction in the branch and continue the original separately.
From the terminal, claude --resume does the same thing.
The non-technical version: It's the "Save As" of AI conversations. Copy the conversation, try something risky in the copy, and keep your original safe.
9. /btw: Side Questions Without Interrupting
While Claude is working on a big task, type /btw followed by a quick question. Claude answers without losing context on what it was doing. Boris uses this "all the time."
The non-technical version: Like whispering a question to someone while they're giving a presentation; they answer you quietly and keep going without missing a beat.
10. Git Worktrees: Run Dozens of Claudes at Once
This is the infrastructure feature that makes Boris' workflow possible. Git worktrees (copies of your code repository that can be worked on independently) let multiple Claude agents work in the same project simultaneously without stepping on each other's code.
Boris says he has "dozens of Claudes running at all times." Each one gets its own worktree, its own isolated copy of the code. Use claude -w to start a session in a new worktree, or check the "worktree" box in the Desktop app.
The non-technical version: Imagine 20 contractors renovating your house, but each one has their own identical copy of the house to work in. Once they're done, you merge the best results. Nobody trips over each other's tools.
11. /batch: Massive Changes, Parallelized
/batch takes a big task, interviews you about what you want, then fans it out to as many worktree agents as needed, dozens, hundreds, even thousands, to get it done. Each agent works in isolation, tests its changes, and puts up a pull request.
Boris described this in a separate thread alongside /simplify (which uses parallel agents to improve code quality and ensure CLAUDE.md compliance). Both features automate what used to take days of manual work.
The non-technical version: Instead of one person remodeling 50 rooms one at a time, you hire 50 people and each one handles a room simultaneously. The job that took weeks now takes hours.
12. --bare: 10x Faster SDK Startup
When you use Claude Code programmatically (through code, not the interactive chat), it normally searches for configuration files, settings, and connected tools on startup. The --bare flag skips all of that, making startup up to 10x faster.
Boris notes this was a design oversight and will become the default in a future version. For now, it's a flag you opt into.
Who this is for: Developers building automated pipelines that call Claude Code hundreds of times. If you're using the interactive chat, you won't need this.
13. --add-dir: Work Across Multiple Repos
Most real-world projects span multiple code repositories. --add-dir (or /add-dir inside a session) lets Claude see and work in additional folders beyond the one you started in. Claude gets more than visibility; it gets full permissions to make changes there too.
Teams can set "additionalDirectories" in their settings file to always load specific folders.
The non-technical version: Instead of Claude only being able to see one filing cabinet, you're giving it access to the whole office.
14. --agent: Custom System Prompts and Tools
Custom agents are pre-configured Claude Code setups with their own instructions and tools. Define an agent in your .claude/agents folder, then launch it with claude --agent=.
Boris calls this "a powerful primitive that often gets overlooked." You could create a "code reviewer" agent with strict review instructions, a "documentation writer" agent that follows your style guide, or a "bug fixer" agent that always runs tests after changes. See the subagents documentation for the full spec.
The non-technical version: Instead of one general-purpose assistant, you create specialized roles. One agent is your editor, another is your fact-checker, another is your project manager. Each one has its own personality and rules.
15. /voice: Code by Talking
Boris' parting revelation: he does most of his coding by speaking, not typing. /voice in the CLI enables voice input (hold the spacebar to talk), the Desktop app has a voice button, and the iOS app supports dictation natively.
The non-technical version: You literally talk to your computer and it writes code. "Add a search bar to the homepage" becomes a spoken sentence, not a typed one. We live in the future.
How to Get Started Today
If you're new to Claude Code, here's the fastest path from zero to productive:
Step 1: Install it. Visit claude.com/product/claude-code and follow the instructions for your platform. Mac and Linux users can run the installer from the terminal. Windows users need WSL2 (a Linux layer for Windows) first. You'll need a paid Claude account ($20/month minimum).
Step 2: Open a project. Navigate to any project folder in your terminal and type claude. It scans your files and you're ready to go. Start by asking Claude to explain your codebase: "What does this project do?"
Step 3: Create a CLAUDE.md. Type /init inside a Claude session. Claude examines your project and generates a starter file. Edit it to add your preferences, coding standards, and anything you want Claude to always know about your project.
Step 4: Learn the essential commands:
/planto have Claude analyze before acting/clearto reset context between tasks/compactto compress a long conversation without losing key info/modelto switch between Claude's models (Sonnet for speed, Opus for complex tasks)
Step 5: Pick one Boris tip. Just one. Try /voice and speak a prompt instead of typing it. Or try /btw to ask a side question during a task. Or set up a simple /loop to run your tests after every change. Start small, build from there.
Where This Is All Going
Boris' thread matters beyond the tips themselves. It reveals a fundamental shift in what "coding tool" means.
A year ago, Claude Code was a chatbot in your terminal. Today it's a platform that supports scheduled automation, parallel agents, device-hopping sessions, custom workflows, and voice interaction. The trajectory points toward something even bigger: developers as orchestrators, managing fleets of AI agents the way a conductor manages an orchestra.
The most interesting signal is /loop. The idea that your AI coding assistant runs continuously in the background, monitoring pull requests, responding to feedback, cleaning up stale work, without you touching the keyboard, is a preview of where every knowledge worker's AI tools are heading. Not a chat window you visit. An always-on collaborator that keeps working while you sleep.
For now, the gap between casual users and power users is enormous. Boris runs dozens of parallel Claude agents orchestrated across devices. Most people are still copy-pasting code into a chat window. The features to close that gap are already built. They're just waiting to be discovered.
And discovering them took reading one thread from the person who built the thing. Which is either a great argument for following the right people on X, or a damning indictment of how we read documentation. Probably both.
SEO Title: SEO Description: . Tags: Claude Code, Anthropic, Boris Cherny, AI coding, developer tools, Claude, automation Categories: Explainer articles
