The tech world is abuzz after Boris Cherny, the mastermind behind Anthropic's Claude Code, pulled back the curtain on his revolutionary workflow. What started as a simple X thread detailing his terminal setup has ignited a firestorm of discussion, with developers hailing it as a paradigm shift in software development. The central theme? A surprisingly straightforward system that allows a single engineer to achieve the output of an entire team.

Running AI Agents in Parallel: A Real-Time Strategy for Code

Cherny's approach deviates sharply from traditional linear coding. Instead of the typical write-test-repeat cycle, he orchestrates a symphony of AI agents working concurrently. "I run 5 Claudes in parallel in my terminal," Cherny explained. He leverages iTerm2 system notifications to manage these simultaneous work streams, with agents handling tasks ranging from test suite execution to legacy module refactoring and documentation drafting. According to Cherny, he also juggles “5-10 Claudes on claude.ai” using a 'teleport' command to seamlessly transition sessions.

This 'fleet commander' mentality showcases Anthropic's strategy of maximizing impact with existing models, as highlighted by President Daniela Amodei. Instead of pursuing massive infrastructure build-outs, Anthropic focuses on clever orchestration to unlock exponential productivity gains. This approach allows a lean team to accomplish more, efficiently.

Slow and Steady Wins the Race: The Power of a Smarter Model

Counterintuitively, Cherny champions Anthropic's most powerful, yet slowest model: Opus 4.5. "I use Opus 4.5 with thinking for everything," he stated. While seemingly paradoxical in an industry fixated on speed, Cherny argues that the superior reasoning and tool-use capabilities of Opus 4.5 ultimately lead to faster overall development.

This insight carries significant implications for enterprise technology leaders. The real bottleneck isn't token generation speed, but rather the human effort required to correct AI errors. By investing in a smarter model upfront, developers can minimize the downstream 'correction tax.' The trade-off pays off by increasing output and reducing the need for constant human intervention.

Collective Learning: Turning Mistakes into Permanent Lessons

A crucial element of Cherny's workflow is addressing the issue of AI amnesia. To ensure that Claude remembers a company's specific coding style and architectural decisions, the team maintains a single file named CLAUDE.md within their git repository. "Anytime we see Claude do something incorrectly we add it to the CLAUDE.md, so Claude knows not to do it next time," he shared.

This simple practice transforms the codebase into a continuously learning entity. Whenever a developer identifies an error during a pull request review, they not only fix the code but also instruct the AI to avoid repeating the mistake. As product leader Aakash Gupta noted, "Every mistake becomes a rule," allowing the agent to become progressively smarter over time. Cherny also utilizes slash commands and subagents to automate the mundane, streamlining development cycles.

""Every mistake becomes a rule," allowing the agent to become progressively smarter over time."

— Aakash Gupta, product leader

Cherny's reveal isn't just about a new workflow; it's a glimpse into the future of software engineering. By treating AI not as a mere assistant but as a full-fledged workforce, developers can unlock unprecedented levels of productivity. Those who embrace this paradigm shift won't just be more efficient; they'll be playing a completely different game, leaving the rest of the industry behind.