Running more than 2 sub-agents in parallel is wasted tokens, dev warns
ALEXEIMARTOV · x · 2026-09-16
A developer asks whether parallelizable work (e.g. 100 identical tasks across 3 sol lanes with Astra medium) justifies running many sub-agents. The reply offers a practical rule of thumb: running more than 2 sub-agents at once almost certainly burns tokens for zero quality gain, because agents don't trust each other enough to skip double-checking each other's output, leading to redundant verification overhead. A useful signal for agent orchestration: parallelism has a quality ceiling.
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