Silicon Valley is abuzz with excitement over a thread on X from Boris Cherny, the creator of Claude Code at Anthropic. In it, Cherny shares his personal terminal setup, revealing a workflow that has left industry insiders calling it a watershed moment for software development.
Cherny's approach is surprisingly simple yet allows a single human to operate with the output capacity of a small engineering department. He runs five AI agents in parallel, using system notifications to manage work streams and hand off sessions between his local machine and the web. This validates Anthropic's "do more with less" strategy, which prioritizes superior orchestration of existing models over building massive infrastructure.
One striking revelation from Cherny's disclosure is that he exclusively uses Anthropic's slowest model, Opus 4.5, despite it being bigger and slower than others. He argues that paying the compute tax upfront eliminates the correction tax later, as the agent is better at tool use and requires less human oversight.
Cherny also detailed how his team solves AI amnesia by maintaining a single file in their git repository, which serves as a self-correcting organism. When humans review pull requests and spot errors, they tag the AI to update its instructions, transforming every mistake into a rule. This practice has been praised by industry observers, who see it as a game-changer for software development.
