The software development landscape is experiencing a profound transformation thanks to Boris Cherny, creator and lead of Claude Code at Anthropic. Recently, Cherny shared his personal AI-enhanced coding workflow, sparking widespread discussion and admiration among developers and industry experts alike.
Starting from a thread on X, Cherny’s insights have become a viral touchstone, with influential voices in the programming community recognizing it as a pivotal moment for Anthropic and potentially for the future of software engineering.
Parallel AI Agents: Turning Coding into a Strategic Operation
Unlike traditional sequential coding, Cherny treats software development like commanding multiple units in a real-time strategy game. He runs five Claude AI agents simultaneously within his terminal, assigning numbered tabs and monitoring system notifications to manage input demands efficiently.
Each AI agent focuses on distinct tasks—running tests, refactoring legacy code, drafting documentation—while Cherny also operates additional Claudes through his browser using a ‘teleport’ command that seamlessly hands off sessions between local and web environments. This multitasking strategy exemplifies Anthropic’s ‘do more with less’ philosophy, prioritizing orchestrated AI workflows over massive infrastructure expansion.
Choosing Intelligence Over Speed: The Opus 4.5 Model Advantage
Contrary to industry trends emphasizing speed, Cherny relies exclusively on Anthropic’s heaviest and slowest model, Opus 4.5. Despite its latency, this model requires less human intervention due to superior reasoning and tool usage capabilities, ultimately accelerating overall development by reducing correction time.
This approach highlights a critical insight for enterprise AI adoption: investing in smarter models upfront pays off by minimizing the “correction tax” that usually hinders productivity when using faster but less capable AI.
Building a Self-Learning Codebase with CLAUDE.md
Cherny’s team combats the common AI limitation of session amnesia by maintaining a dedicated file named CLAUDE.md within their Git repository. Whenever Claude makes an error, the team documents it here, effectively teaching the AI to avoid repeating mistakes in future sessions.
This system turns the software codebase into a self-correcting organism where every human fix becomes a rule update for the AI. As product leader Aakash Gupta emphasized, “Every mistake becomes a rule,” meaning the AI continuously grows smarter the longer it collaborates with developers.
Slash Commands and Specialized Subagents Streamline Development
Automation plays a central role in Cherny’s workflow. He employs custom slash commands stored in the project repository to execute complex operations quickly. A prominent example is the /commit-push-pr command, which autonomously manages version control tasks such as committing code, pushing changes, and opening pull requests, boosting efficiency.
Additionally, Cherny uses specialized AI subagents tailored for different development phases, including code simplification and application verification. These subagents ensure that the software architecture remains clean and that comprehensive end-to-end tests are conducted before deployment.
Verification Loops: Elevating AI-Generated Code Quality
A core factor behind Claude Code’s rapid growth to a reported $1 billion in annual recurring revenue is its integrated verification loop. Unlike AI models that only generate code, Claude actively tests every change using browser automation and test suites.
Cherny explained that Claude opens a browser, rigorously tests the user interface, and iterates until both the code functions correctly and the user experience is satisfactory. This self-verification multiplies code quality by two to three times, ensuring reliability and reducing human debugging effort.
Implications for the Future of Software Engineering
The enthusiastic response to Cherny’s workflow signals a paradigm shift in software development. Where AI was once merely an autocomplete assistant, it now acts as an operational workforce capable of managing multiple complex tasks simultaneously.
As Jeff Tang succinctly put it, “Read this if you’re already an engineer… and want more power.” The tools to multiply human output several times over are already accessible, demanding only a mindset change to treat AI as a collaborative labor force rather than a simple helper.
Developers who embrace this new approach will not only enhance their productivity but fundamentally change the way software is created, leaving traditional typing-only methods behind.
Fonte: ver artigo original
Top AI-Powered Gadgets Revolutionizing Pet Care in 2025
Google DeepMind’s Bioacoustic AI Model Excels in Whale Detection Using Bird Call Training
Apple and Netflix Collaborate to Stream Formula 1 Canadian Grand Prix
MWC 2026 Validates AI-Native Networks as a Reality Beyond 6G Promises
