Anthropic engineer Boris Cherny revealed during a Sequoia Capital interview that he runs "a few thousand" AI coding agents overnight, a glimpse into how Silicon Valley engineers are transforming AI from interactive chatbots into persistent autonomous assistants.
Most remarkably, Cherny manages this complex agent infrastructure almost entirely from his phone.
Cherny, who created Claude Code, described opening the Claude app, navigating to the code tab, and displaying sessions containing multiple agents. When asked how many concurrent sessions he maintains, Cherny said he typically runs "five to 10 sessions," each spawning dozens of agents.
"Usually, every night, I have like a few thousand that are doing kind of deeper work," he added casually, as if orchestrating thousands of autonomous systems was routine engineering practice.
The technology that supports this process is made possible by two aspects of Claude Code known as '/loops' and 'Routines'. The aspect /loops is responsible for scheduling recurring activities locally by way of cron, whereas Routines run recurring activities on servers, allowing agents to continue working after laptops have been shut down.
"It's just like, the coolest thing," Cherny said enthusiastically. "If you haven't experimented with it, highly recommend it." The distinction matters: traditional coding assistants respond to immediate user queries, while /loops and Routines enable agents to execute autonomously on predetermined schedules.
In January 2026, Cherny shared his agentic setup on X, calling it "surprisingly vanilla." That single post has accumulated 104,000 saves and 8.1 million views, suggesting broader interest in how expert engineers actually use AI agents. "I didn't realise that it would be surprising for anyone," Cherny reflected during the interview. "That was just like the way that I coded."