@PrajwalTomar_: WAIT. This is actually insane. A senior dev just dropped the 3-tier agent stack Boris Cherny (the creator of Claude Cod…
Summary
A senior dev describes a three-tier agent stack using Claude Code that runs thousands of agents overnight, with examples like migrating user email fields across a 4000-file monorepo using 800 agents in parallel.
View Cached Full Text
Cached at: 06/15/26, 07:06 PM
WAIT. This is actually insane.
A senior dev just dropped the 3-tier agent stack Boris Cherny (the creator of Claude Code) actually runs.
He uninstalled his IDE. Now runs 5-10 interactive Claude sessions during the day and several thousand agents overnight, mostly triggered from his phone.
→ Tier 1: /loop runs slash commands on a schedule inside open sessions (PR review, deploy watch, Slack feedback mining every 5-30 min) → Tier 2: Routines run on Anthropic’s cloud overnight against a fresh clone (morning report, deep audit, dep hygiene, weekly recap) → Tier 3: /batch + dynamic workflows orchestrate hundreds to thousands of agents on a single task
A real example: 800 agents migrated user. email to user. primaryEmail across a 4,000-file monorepo. Each in its own worktree. Each with its own adversarial reviewer.
The Bun team rewrote their entire Zig codebase to Rust this way.
His closing line is the part nobody is saying out loud:
“The compound effect is in the system, not the model.”
Most builders are still running one Claude session at a time. The smart ones are running thousands while they sleep.
Full post:
Similar Articles
@PrajwalTomar_: I still don't think people understand what just happened with Claude Code subagents. You can now hand off a whole proje…
A developer explains how to effectively use Claude Code subagents by capping and scoping them, rather than running many agents, to avoid context burn and overlapping work.
@PrajwalTomar_: Okay, this is the first Claude Code setup I've seen where 3 agents outship 20. Everyone's unleashing 20 subagents and g…
A developer shares a Claude Code setup where three well-structured agents outperform twenty, emphasizing orchestrator control, scoped tasks, and using cheap models for volume while reserving Opus for lead/reviewer roles.
@0xCodez: https://x.com/0xCodez/status/2058513716509913581
A comprehensive walkthrough on building multi-agent teams with Claude Managed Agents, covering role design, model mixing, and parallel execution to scale from one to 20 agents.
@mylifcc: After analyzing the code of Claude Code, let's chat about the design of its asynchronous agents and workflows. Claude C…
The author dissects Claude Code's source code to explain how its multi-agent system layers the Agent tool, a task registry with a command queue, disk-based mailbox teammates, and a deterministic JS workflow engine — all funneled through a shared query executor with isolated contexts and message-only communication.
@AnatoliKopadze: Head of Claude Code: "85% of our engineers are running dozens or hundreds of agents. The way you do it is graph enginee…
The Head of Claude Code at Anthropic discusses how 85% of engineers are using AI agents and graph engineering to perform the work of entire teams, highlighting internal practices and future directions.