Insight for: Show HN: Lessons learned from running Claude Code swarms at scale

Fleet, an application for orchestrating and managing swarms of coding agents.

Analyzed: Jun 5, 2026

This submission highlights critical operational challenges in scaling LLM agent deployments. The core pain points revolve around inefficient token consumption due to poor abstraction mechanisms (CLAUDE.md, skills, indiscriminate plugin attachment) and rigid model behaviors (unmanageable system tools, lack of background session interaction). The proposed solutions—hierarchical knowledge bases, precise plugin management, and task decomposition for model routing—underscore a maturing understanding of LLM orchestration. The market implication is a clear demand for sophisticated agent management platforms that prioritize token efficiency, context control, and cost optimization. Developers require granular control over agent interactions and resource allocation to achieve scalable, cost-effective AI-driven workflows, moving beyond simplistic prompt engineering to structured, intelligent agent design.

coding agents swarms Python orchestrator agent lifecycle centralized SQLite DB spawn agents dependencies coder/model concurrently burn through limits maximize efficiency CLAUDE.md wasted tokens irrelevant instructions skills budget progressive disclosure plugins hierarchical knowledge base system tools context AskUserQuestion background sessions MCP- or CLI-based decompose work subtasks weaker, cheaper models save tokens context-switching

Hacker News Post

Parent Entity

Show HN: Lessons learned from running Claude Code swarms at scale

Score: 8