Only working 2 hours a day? This Google engineer used Claude to do 80% of his work.

Title: Google Engineer Automated 80% of His Work with Claude Code. Here’s the Exact System He Built.
Author: @noisyb0y1 Compilation: Peggy, BlockBeats

Author: Rhythm BlockBeats

Source:

Repost: Mars Finance

Editor’s Note: As “AI coding” gradually becomes an industry consensus, what truly changes productivity isn’t the model itself, but how you set rules for the model, organize workflows, and embed it into a sustainable system.

Starting from a simple CLAUDE.md file, progressing to multi-agent collaboration, and then to automated development cycles, this approach transforms the development process from “human-AI dialogue” into “managing an AI engineering team.” In this process, errors are constrained upfront, workflows are structured, and code generation, testing, and review gradually shift from manual execution to system takeover.

More notably, the article also reveals an overlooked detail: in long contexts and complex systems, model behavior isn’t entirely controllable. Whether it’s hidden token consumption or diluted instructions, these factors invisibly impact output quality. This makes “how to manage AI”—not just “how to use AI”—a new core capability.

At this point, developers no longer focus solely on coding but work around rule design, process scheduling, and result verification. Those who take this step early are already shifting from “doing things themselves” to “letting the system do the work for them.”

Below is the original article:

An Google engineer with 11 years of experience automated 80% of his work using Claude Code and a simple .NET application.

Now, he only works 2–3 hours a day instead of 8, with the rest of the time basically in “relaxation,” as the system runs on its own, bringing him $28k in passive monthly income.

He possesses a method you haven’t yet learned.

Part 1—Writing CLAUDE.md According to Karpathy Principles

Andrej Karpathy—one of the most influential AI researchers worldwide—systematically summarized the most common mistakes in large language model coding: over-design, ignoring existing patterns, and introducing unnecessary dependencies.

Some have compiled these observations into a unified CLAUDE.md file.

As a result, this project gained 15k stars on GitHub within a week—arguably, 15k people changed their work methods because of it.

The core idea is simple: if mistakes are predictable, they can be proactively avoided with clear instructions. Just place a markdown file in your code repository to provide Claude Code with a complete set of structured behavior rules, unifying decision-making and execution across the project.

This file mainly contains four core principles:

· Think First, Code Later → Avoid incorrect assumptions and overlooked trade-offs
· Simplicity First → Avoid over-design and bloated abstractions
· Surgical Changes → Avoid modifying code that isn’t requested to be changed
· Goal-Driven Execution → Test first, then verify based on clear success criteria

No frameworks or complex tools are needed—just one file to change Claude’s behavior at the project level.

The real difference lies in:

· Without CLAUDE.md: Claude violates norms about 40% of the time
· With Karpathy’s CLAUDE.md: Violation rate drops to about 3%
· Setup time: only 5 minutes

Command to auto-generate your own CLAUDE.md:

claude -p "Read the entire project and create a CLAUDE.md based on: Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution. Adapt to the real architecture you see." --allowedTools Bash,Write,Read

It replaces a Claude that over-designs for simple tasks, introduces unnecessary dependencies, or modifies files that shouldn’t be touched.

Part 2—Everything Claude Code: A Complete Engineering Team in One Repository

Everything Claude Code (over 153k stars on GitHub)

This isn’t just a set of prompts but more like a complete AI operating system for building products.

Supports running on Claude, Codex, Cursor, OpenCode, Gemini, and other tools—a system available everywhere.

Installation:

/plugin marketplace add affaan-m/everything-claude-code

Or manual installation—simply copy the components you need into the .claude/ directory of your project. Don’t load everything at once—loading 27 agents and 64 skills may exhaust your context window before your first prompt. Keep only what’s necessary.

The key difference:

· Previously: You were talking to AI
· Now: You are managing an automated AI engineering team

It replaces: the weeks you’d spend building your own agent system, configuring planning/review/security tools, and paying $200–$500 monthly for various AI services.

Part 3—A Hidden “Scandal”: Claude Code v2.1.100 Quietly Drains Your Tokens

Someone intercepted and analyzed complete API requests for four different versions of Claude Code via an HTTP proxy.

They found:

• v2.1.98: 169,514 bytes request → 49,726 tokens charged
• v2.1.100: 168,536 bytes request → 69,922 tokens charged
• Difference: -978 bytes but +20,196 tokens

V2.1.100 sends fewer bytes but charges an extra 20k tokens. This “inflation” happens entirely on the server side—you can’t see it or verify via the /context interface.

Why is this important beyond billing? Those extra 20k tokens are stuffed into Claude’s actual context window.

This means:

→ Your CLAUDE.md instructions are diluted by these 20k “hidden contents”
→ Output quality in long conversations drops faster
→ When Claude ignores your rules, it’s hard to identify why
→ Claude Max’s usage quota is consumed about 40% faster

Fixing it takes only 30 seconds:

npx claude-code@2.1.98

This is a temporary workaround before an official fix from Anthropic, but in practice, you’ll immediately notice a change in session quality.

It replaces: the need to guess why Claude suddenly stops following your instructions.

Example: What a Fully Automated System Looks Like

An engineer with 11 years of experience built a system with three parts:

Results after one week:

· Previously: 8 hours coding daily
· Now: Only 2–3 hours for code review and testing
· Code quality: roughly the same—he reviews each piece
· Team status: always online—mouse moves automatically every minute
· Remaining time: free all day

This isn’t magic—it’s the combined effect of CLAUDE.md + suitable agents + a 15-minute cycle mechanism.

Complete list:

What you’ll gain after reading:

· Previously: Claude violated norms 40% of the time
· Now: With Karpathy’s CLAUDE.md, violations drop to 3%

· Previously: You needed weeks to build agents
· Now: 27 agents ready to use out of the box

· Previously: Claude Max would exhaust your quota in 2–3 hours
· Now: Downgrade to v2.1.98 to recover about 40% of usage limit

· Previously: 8 hours coding daily
· Now: Only 2–3 hours for review, rest automated by the system

· Setup time: 15–20 minutes
· Daily savings: 5–6 hours
· Monthly savings: 100–120 hours

If your time is worth $30/hour—you’re “invisibly losing” $3,000–$3,600 each month.

At $100/hour, that’s $10,000–$12,000 lost monthly—just because you’re manually writing code that Claude could do itself.

Most developers will never reach this level—not because they can’t, but because they think it’s too complicated. In reality, between you and full automation, only three commands and one file stand in the way.

The engineer I mentioned at the start isn’t a genius or a Google veteran. He simply spent one evening building the system—since then, work is done by the system, and he just lives his life.

You can do the same tonight. While others debate whether AI will replace developers, those who have set up their systems are just collecting money and relaxing.

The choice is clear. You’re building your life—so choose the right path.