When Boris Cherny, the creator of Claude Code—the world’s most advanced AI coding assistant—shared his personal workflow, the software development community quickly took notice. Over the past week, developers and industry experts have analyzed Cherny’s approach, which promises to reshape how code is written and tested in the AI era.
What started as a simple thread on X (formerly Twitter) has evolved into a viral discussion about the future of programming, with prominent voices in the field calling Cherny’s techniques a watershed moment for Anthropic and the broader AI ecosystem.
Orchestrating Multiple AI Agents: Coding as a Real-Time Strategy Game
Contrary to the traditional linear coding process where developers write and test code sequentially, Cherny likens his method to commanding a fleet of autonomous units. He runs five instances of Claude in parallel within his terminal, using numbered tabs and system notifications to manage simultaneous tasks.
Each AI agent handles different responsibilities: one runs test suites, another refactors legacy code, while a third drafts documentation. Additionally, Cherny operates multiple Claude agents in his browser, seamlessly transferring sessions between web and local environments. This multitasking transforms coding from typing syntax into directing intelligent agents, resembling a real-time strategy game rather than conventional programming.
This approach aligns with Anthropic’s “do more with less” philosophy, emphasizing efficient model orchestration over massive infrastructure expansion—a contrast to competitors investing heavily in large-scale AI hardware.
Choosing Quality Over Speed: The Power of the Smartest AI Model
In an industry obsessed with reducing latency, Cherny makes a compelling case for prioritizing intelligence over speed. He exclusively uses Anthropic’s largest and slowest model, Opus 4.5, explaining that its superior understanding and tool integration reduce the need for human corrections, effectively accelerating overall development.
This insight challenges the conventional focus on raw generation speed. Instead, the real bottleneck is human intervention to fix AI errors. Investing in a smarter model upfront minimizes this “correction tax,” resulting in faster project completion.
Building a Self-Correcting Codebase with CLAUDE.md
One major challenge with AI assistants is their inability to remember specific company coding styles or architectural preferences across sessions. Cherny’s team addresses this by maintaining a dedicated file named CLAUDE.md within their git repository.
Whenever Claude makes a mistake, the team documents the issue in this file, ensuring the AI learns and avoids repeating errors. This practice turns the codebase into a continuously improving system where each human correction informs the AI’s future behavior, making the agent smarter over time.
Automating Tedious Tasks with Slash Commands and Subagents
Cherny emphasizes rigorous automation by using custom slash commands embedded in the project repository. For example, the /commit-push-pr command automates the entire process of committing code, pushing changes, and opening pull requests, drastically reducing manual effort.
Furthermore, subagents—specialized AI personas—manage distinct phases of development. These include code simplification after initial work and running comprehensive verification tests before deployment, streamlining the lifecycle with minimal human intervention.
Verification Loops: Ensuring AI-Generated Code Actually Works
A key factor behind Claude Code’s rapid growth to over $1 billion in annual recurring revenue is its built-in verification loop. Unlike typical text generators, Claude autonomously tests every code change, using tools like browser automation and test suites to validate functionality and user experience.
Cherny explains that this verification process improves code quality by two to three times, as the AI not only writes but also confirms the correctness of its output, reducing bugs and rework significantly.
Implications for the Future of Software Engineering
The enthusiastic response to Cherny’s workflow signals a fundamental transformation in programming paradigms. Historically, AI-assisted coding was limited to autocomplete features that sped up typing. Now, AI acts as a workforce multiplier, enabling individual developers to achieve the output of entire teams.
As Jeff Tang, a respected developer, summarizes, “The tools to multiply human output by a factor of five are already here. They require a mindset shift from viewing AI as an assistant to treating it as a labor force. Those who adapt first will not just be more productive—they will be playing an entirely different game.”
In essence, Cherny’s approach exemplifies how artificial intelligence is changing everyday work for developers, highlighting AI tools that are genuinely enhancing productivity and redefining roles within the software industry.
Fonte: ver artigo original
Related Articles
Former Harvard Students Launch Controversial AI Smart Glasses with Always-On Microphone
Apple Gains Full Access to Google’s Gemini AI to Enhance Siri with Lightweight On-Device Models
Amazon Prime Video Introduces AI-Generated Video Recaps for Select TV Shows
Booking.com Reports Customer Data Breach in Recent Cyberattack
