Claude Code Creator Runs Thousands of Sub-agents Overnight
Business Insider reports that Boris Cherny, the lead engineer behind Claude Code, said during a May 4 interview at Sequoia Capital that he runs "five to 10 sessions," each containing multiple agents, and "usually, every night, I have like a few thousand" AI sub-agents performing "deeper work." Cherny told the audience he tracks those sessions on his phone using the Claude app and relies on two persistent-automation features, /loops and Routines; Business Insider reports he said /loops can be scheduled locally via cron while Routines run recurring tasks.
What happened
Business Insider reports that Boris Cherny, the creator and lead engineer of Claude Code, described his personal workflow during a May 4 interview at Sequoia Capital. Cherny said he typically runs "five to 10 sessions," with multiple agents per session, and "usually, every night, I have like a few thousand" AI sub-agents doing what he called "deeper work," Business Insider reports. Cherny also said he monitors those sessions on his phone via the Claude app and uses two persistent-automation features, /loops and Routines, with /loops schedulable locally via cron, Business Insider reports.
Editorial analysis - technical context
Multi-agent and persistent-automation setups like the one Cherny described foreground several technical trade-offs that practitioners already face. Systems scaling to thousands of concurrent agents typically require orchestration for task routing, durable state management for long-running work, and observability to diagnose failures. Persistent schedulers and recurring-routine primitives reduce manual triggers but shift complexity into monitoring, retry logic, and cost control.
Context and significance
For practitioners, Cherny's remarks are an anecdotal indicator of how engineers experiment with agent-based automation in everyday workflows. Industry-pattern observations: teams adopting always-on agents often confront API quota limits, nontrivial compute bills, and reproducibility challenges when many parallel sessions alter shared code or artifacts. The specific mention of phone-first monitoring underscores a UX trend where developer tooling surfaces run-state in lightweight clients rather than full dashboards.
What to watch
Observers should track how mainstream developer platforms expose durable automation features (for example /loops -style schedulers and recurring routines), what telemetry primitives they provide for auditing and debugging, and how pricing models handle long-lived or highly parallel agent activity. Also watch for documentation and guardrails around state persistence, idempotency, and cost estimation for recurring agent fleets.
Scoring Rationale
Anecdotal but notable signal about how engineers use agent-based automation. Useful for practitioners watching tooling and orchestration trends, though not a product-level change or benchmark.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
