More than 2 parallel subagents just burns tokens, harness builders say 1 orchestrator + 1 executor works best
omarsar0 · x · 2026-09-17
omarsar0 and @pvncher share harness-building experience on whether subagents are worth it.
Key points:
- Where subagents shine: parallel research, code review, monitoring/tracking, and better context management. omarsar0 trusted Claude Code subagents mainly for parallelizing research, not for coding.
- Beyond 2 parallel subagents, you're burning tokens for zero quality gain: agents don't trust each other enough and keep double-checking each other's work.
- Best current setup: one orchestrator agent paired with one executor subagent works best; adding more degrades results.
Bottom line: subagents are cost-sensitive and coordination quality decides success — don't scale beyond two for now.
Related event: Running More Than 2 Parallel Sub-Agents Is Likely Just Burning Tokens(7 posts)→
More from coding & agent
- Cohere Labs agent team-construction method cuts multi-agent costs by 70% — Cohere_Labs · 2026-11-02
- Long-read: AI's real value comes after the efficiency era—infra glut, agent tooling boom — yangyi · 2026-11-02
- offprompt: MIT-licensed MCP server hands agents API keys without ever exposing them — Vast_Ad_7338 · 2026-11-02
- Reddit idea: expose every OOP public method as AI agent tools to agentify legacy apps — out_of_nowhere__ · 2026-11-02
- DeepSeek design doc: a person can only follow 3-5 agent sessions at once, attention is the real bottleneck — heyshrutimishra · 2026-11-02
- Software fundamentals are key to getting good results from AI coding agents — bendee983 · 2026-11-02