Main Agent + Sub-Agent Task Routing
Roger_M_Taylor · x · 2026-07-19
This post discusses a main agent + sub-agent execution architecture: the main agent handles planning and decision-making, while delegating actual execution to sub-agents.
The core practices include:
- The main agent's effort isn't fixed; it dynamically switches between low / medium / high based on the task. It defaults to low or medium, reserving high only for multi-step judgments.
- Sub-agents don't inherit the main agent's reasoning level; their models are explicitly specified at generation time based on the task.
The author provides routing examples:
- Routine execution, drafting, formatting, and simple tweaks go to sonnet
- Harder execution, debugging, and tasks requiring deeper reasoning go to opus
- Design work goes to GPT 5.5, and coding goes to GPT 5.6
This concept can scale to a week-long campaign: the main agent defines the launch week's angle and pace, then delegates retrieval, drafting, polishing, and scheduling to multiple sub-agents. The accompanying image emphasizes cost: high-effort tasks consume tokens rapidly, so heavy reasoning should be reserved for tasks that truly warrant it.
More from coding & agent
- Frontend Textbooks turns long Codex outputs into book-style documents — nbaschez · 2026-07-21
- A Codex harness user says OpenAI’s assistant is less smart-sounding but more decisive — McDonaghMatthew · 2026-07-21
- Octen’s search benchmark is public and reproducible for agent stacks — aakashgupta · 2026-07-21
- Octen says agent search now runs at 62ms P50 with only a 6ms P90 gap — aakashgupta · 2026-07-21
- Users say Codex side chats are becoming part of their real coding workflow — nicoalbanese10 · 2026-07-21
- LithosAI says Kimi K3 is frontier-level for open-weight agentic tasks — JiaZhihao · 2026-07-21