AI
Guide
Engineering
Business Strategy
Share
The Teams That Move Fastest Are Already Using Claude Code.
By Pek Pongpaet | Founder & CEO, Impekable | 8 min read
Tips sourced from "Mastering Claude Code in 30 Minutes" by Boris Cherny, creator of Claude Code at Anthropic
How long does it take your engineering team to fully onboard a new hire?
Two weeks? Three? I've heard the number as high as a month for complex codebases. It's one of those costs so embedded in how we work that most companies have stopped questioning it.
At Anthropic, that number is now two to three days.
Not because they hire superhuman engineers. Because they gave those engineers a tool that actually understands the codebase and gets to work on day one with zero setup. That tool is Claude Code, and the tips below come directly from Boris Cherny, the Anthropic engineer who built it.
If you're a business or technology leader trying to figure out where AI fits in your engineering workflow, this is one of the more concrete answers I've come across.
What is Claude Code and how does it work?
Claude Code is an AI coding agent that runs in your terminal. You install it with one command, and it reads your codebase locally. No code is uploaded, remotely indexed, or used for model training. It runs alongside VS Code, Vim, Xcode, JetBrains, or whatever editor your team already uses.
The distinction worth understanding: most AI coding tools predict the next line of code. Claude Code takes a task and runs it to completion. It reads files, writes code, runs tests, and iterates without someone approving each step.
For non-technical leaders, the practical difference is this: autocomplete tools save developers a few keystrokes. Claude Code handles entire features.
npm install -g @anthropic-ai/claude-code
How does Claude Code cut onboarding time from weeks to days?
At Anthropic, new engineers use Claude Code on day one to explore the codebase before writing anything. They ask it questions like "How is RoutingController.py used?" and get back a real walkthrough with context and file references, not just a search result.
That's the mechanism behind the 3-day onboarding number. The tool does the archaeology that new hires would otherwise spend weeks doing on their own.
For companies where onboarding a single engineer costs $15,000 to $30,000 in lost productivity and manager time, that compression adds up fast.
7 tips from Boris Cherny, the creator of Claude Code
These come directly from Cherny's session on how Anthropic uses the tool internally. I've written them for business and technical leaders, not just developers.
1. Start with questions, not code
Before writing anything, ask Claude to explain the codebase. "Walk me through how authentication works" gets a real answer with context and examples. This is the first thing Anthropic teaches new engineers. It's why they onboard in days instead of weeks.
For leaders: this means a new hire can be productive on day one without pulling senior engineers away from their own work.
2. Use it to surface what your team actually shipped
Ask "What did I ship last week?" and Claude reads the commit history, summarizes it, and connects it back to related issues. No more manual archaeology before a retro or status update.
For leaders: this is also useful for engineering managers who want a quick read on what's moving without sitting in standups.
3. Give it a way to check its own work
Most people prompt once and accept the output. The better approach: give Claude a unit test, an integration test, or a screenshot of the target state. Let it run two to three passes. The first output is decent. The third is usually close to done, with no additional input required between rounds.
4. Say "make a plan first" before any complex task
Four words before every nontrivial request: "Before you write code, make a plan." It forces the model to reason about structure before executing. The output quality is noticeably better.
5. Use CLAUDE.md to stop re-explaining your codebase
Create a file called CLAUDE.md in the project root. Fill it with your style guide, common commands, and key architecture decisions. Claude reads it automatically at the start of every session. You write it once and never explain the same thing twice.
For leaders: this is how you institutionalize engineering standards without another wiki nobody reads.
6. Commit your configuration to get team-wide results
When you check configuration files into git, everyone on the team inherits the same shared memory, permissions, and MCP server setup automatically. One commit and the whole team is running the same optimized setup.
7. Use the SDK to automate the repetitive stuff
The Claude Code SDK is a programmatic interface. Put data in, get structured output back. Anthropic uses it internally for incident response. External teams use it for CI pipelines, automation scripts, and any background workflow where you want Claude doing the reasoning without a human in the loop.
For leaders: this is where the ROI compounds. Once your engineers are comfortable with the tool, the SDK is where you start replacing manual processes entirely.
Source: Anthropic.com
How to set up Claude Code for the first time
Run these three commands right after install:
/terminal-setupenables Shift+Enter for new lines/themesets your display preference/allowed-toolspre-approves permissions so you're not clicking through confirmations on every action
Takes two minutes and removes most of the friction from early sessions.
Claude Code keyboard shortcuts
Shortcut | What it does |
Shift + Tab | Auto-accept all pending edits |
| Save a memory for future sessions |
| Enter bash mode directly |
| Add a file or folder to context |
Esc | Cancel the current action |
Esc + Esc | Navigate back through prompt history |
Ctrl + R | Toggle verbose output |
Can you run multiple Claude Code instances at the same time?
Yes. For teams with parallel workstreams, you can run multiple instances using git worktrees for isolation, SSH + tmux, or GitHub Actions. At that scale it starts to feel less like a coding tool and more like running several developers in parallel.
Claude Code also handles images. Drag a UI mockup directly into the terminal or pass a file path and Claude builds toward it, with no separate tool-switching required.
What's the difference between Claude Code and GitHub Copilot?
Copilot and most other AI coding tools work at the line or function level. They predict what comes next in the code you're writing. Claude Code works at the task level. You give it a complete objective, and it figures out which files to touch, writes the code, checks its own work against tests, and iterates.
Copilot is faster for small, isolated completions. Claude Code is faster for anything that spans multiple files or requires understanding how different parts of the codebase connect.
The tradeoff: Claude Code performs better when you give it context upfront. CLAUDE.md handles most of that automatically.
Frequently asked questions
Does Claude Code upload my code to Anthropic?
What editors does Claude Code work with?
Is Claude Code for individual developers or teams?
What is the Claude Code SDK used for?
How does Claude Code handle tasks that span multiple files?
The bottom line
Most companies are using AI to help developers type a little faster. The gains are real but they're incremental.
Claude Code is a different kind of tool. The engineers at Anthropic aren't using it to type less. They're using it to spend less time navigating the codebase and more time solving actual problems.
If your team is still treating AI as a smarter autocomplete, this is worth a closer look. Your code never leaves your machine, install takes 30 seconds, and the first useful session takes under five minutes.
© 2026 Impekable · impekable.com · Written by Pek Pongpaet, CEO
