When Boris Cherny, the creator and lead of Claude Code at Anthropic, shared details of his unique development workflow, the software engineering community quickly recognized it as a potential game-changer. Over the past week, developers and industry experts have extensively analyzed his approach, which leverages multiple AI agents working in parallel to amplify human productivity.
Running Multiple AI Agents in Parallel: Coding as a Real-Time Strategy Game
Unlike the traditional linear approach to coding—write, test, then move on—Cherny describes his method as commanding a fleet of AI agents simultaneously. He operates five Claude AI agents in parallel within his terminal, each focusing on distinct tasks such as running test suites, refactoring legacy code, or drafting documentation. These agents are managed through system notifications, enabling seamless multitasking and real-time oversight.
Additionally, Cherny utilizes between five and ten Claude agents in his web browser, transferring sessions between his local machine and the cloud using a “teleport” command. This orchestration aligns with Anthropic’s “do more with less” philosophy, emphasizing productivity through smart coordination rather than massive infrastructure expansion.
Prioritizing Intelligence Over Speed: The Case for Using the Slowest, Smartest Model
Contrary to the industry’s focus on minimizing latency, Cherny exclusively uses Anthropic’s most advanced and computationally intensive model, Opus 4.5. He explains that despite being slower, this model requires less human correction due to its superior reasoning and tool-use capabilities, ultimately accelerating the overall development process.
This insight is critical for enterprises aiming to optimize AI-assisted software creation. The main bottleneck is not the AI’s generation speed but the human effort needed to fix mistakes. Investing in smarter AI upfront reduces these correction costs significantly.
Transforming AI Mistakes into Continuous Learning
A common challenge with AI models is their inability to retain company-specific coding conventions across sessions. Cherny’s team addresses this by maintaining a single, shared file named CLAUDE.md within their git repository. Whenever the AI makes an error, it is documented in this file, ensuring future iterations avoid repeating the same mistakes.
This process effectively turns the codebase into a self-improving system. Developers not only fix errors but also update the AI’s instructions, creating a feedback loop where “every mistake becomes a rule,” as noted by product leader Aakash Gupta. Over time, this enables the AI to become increasingly intelligent and aligned with the team’s standards.
Slash Commands and Subagents: Automating Repetitive Development Tasks
Cherny’s workflow also employs automation through custom slash commands embedded in the project repository. These commands, such as /commit-push-pr, automate complex version control tasks like committing code, pushing changes, and opening pull requests, significantly reducing manual overhead.
Moreover, specialized AI subagents are assigned to distinct phases of development. For example, a code-simplifier agent refines architecture post-implementation, while a verify-app agent executes comprehensive end-to-end testing before deployment. This division of labor ensures higher efficiency and quality throughout the software lifecycle.
Verification Loops: The Key to Reliable AI-Generated Code
A pivotal factor behind Claude Code’s rapid achievement of $1 billion in annual recurring revenue is its robust verification loop. Unlike typical AI text generators, Claude actively tests every code change it proposes. Using tools like the Claude Chrome extension, the AI performs browser automation, UI testing, and iterative refinements until the code functions correctly and delivers a satisfactory user experience.
Cherny asserts that enabling AI to verify its own output improves code quality by two to three times. This capability transforms the AI from a mere writer of code to a responsible tester, reducing the likelihood of bugs and errors making their way into production.
Implications for the Future of Software Engineering
The enthusiastic response to Cherny’s revelations signals a profound evolution in software development. AI is no longer just an assistant offering autocomplete suggestions; it is becoming a fundamental component of the development workforce. As Jeff Tang, a respected developer voice, summarized, embracing this new paradigm empowers engineers to dramatically scale their productivity.
Developers who adapt by treating AI as a collaborative workforce rather than a simple tool will gain a significant competitive advantage. According to industry observers, this shift represents not merely an incremental improvement but a wholesale transformation in how software is created, heralding a new era where coding resembles commanding autonomous agents in a strategic environment.
Gridcare Uncovers Over 100 GW of Hidden Data Center Capacity Using Advanced AI Analytics
Grammarly Review 2026: Is Grammarly Premium Worth It?
Ilya Sutskever Highlights the Need for a New Paradigm in AI Learning
Google’s Gemini App Integrates SynthID to Authenticate AI-Generated Images
