Limit subagents to 2 to avoid wasting tokens
Melvin Vivas · X post · 2026-09-16 · Open on X
Topics: AI Agents, Tool Use & MCP, LLMOps, Deployment & Monitoring · Level: intermediate
Summary
The post quotes a claim that running more than 2 subagents at once mostly burns tokens with no quality gain, because agents keep double-checking each other's work. The creator's Codex Astra orchestrator + Luna subagents setup now includes profiles capped at 2 subagents.
Key points
- Quoted claim: more than 2 concurrent subagents usually burns tokens for no quality gain
- Reason: agents don't trust each other, so they re-check each other's work
- codex-astra-luna-orchestrator now has profiles with a maximum of 2 subagents
- Pick a max-2-subagents profile when running the setup
Resources mentioned
- donvito/codex-astra-luna-orchestrator · repo · github.com · free
The creator's Codex skill for an orchestrator-plus-subagents setup (Astra orchestrator, Luna subagents), configured through config.toml.
Also in: Codex Orchestrator Pattern: Astra/Sol Orchestrator with Luna Subagents (Melvin Vivas on X · notes), Open-source Codex skill: Astra/Sol orchestrator with Luna subagents (Melvin Vivas on X · notes), Cost-Efficient Codex Subagents: Astra/Sol Orchestrator + Luna Workers (Melvin Vivas on X · notes), Codex Orchestrator v0.2.1: GPT-6 Sol orchestrator with Luna subagents (Melvin Vivas on X · notes) and 27 more
Try this
- Choose a profile with at most 2 subagents when running the orchestrator setup
- Cap concurrent subagents to avoid wasting tokens
More in AI Agents, Tool Use & MCP
- Agent Monitor: Visualize Coding-Agent Subagents, Traces and Costs
- Coworker: an open-source AI agent desktop app for non-technical users
- Codex orchestrator + subagents setup with configurable profiles
- OpenAI's Agentic Software Factory: Build-Review-Deploy-Observe Loop
- How OpenAI's Agentic Software Factory Works (Pragmatic Engineer)
- Coworker v0.6.1: Connect AI Coworkers to Telegram and Discord Bots