Sol lead + 16 cheap Luna planners nears all-Sol team results at 70% less cost
nouhadziri · x · 2026-10-14
The team-scaling installment of the multi-agent experiments: instead of letting the lead size its own team, researchers gave a Sol lead 2, 4, 8, or 16 cheap Luna workers at each workflow stage (planning, implementation, verification).
Findings:
- Planning kept adding reward as the team grew: each planner explores a different part of the program
- Implementation and verification plateaued at four workers: extra coders just split code into thinner slices for the lead to stitch
- 16 Luna planners hit 61.8%, near the 63.2% of an all-Sol team that delegates every stage, at $155 vs $510
- Near-frontier results at 70% less cost
More from coding & agent
- 7 lessons from building a personalized AI gift site in one hour, with $0 hosting — jamesbrooksco · 2026-10-15
- Google's ADK brings one agent framework to Python, Go, TypeScript, Java and Kotlin — rseroter · 2026-10-15
- AI agent flags a bizarre '71 >= 71' error while a server pool balloons by 1.4TB — haydendevs · 2026-10-15
- Invideo launches an MCP for editing videos directly via Claude or ChatGPT — azed_ai · 2026-10-15
- Connecting Grok to ROS via OpenClaw and AgenticROS for robotics missions — chrismatthieu · 2026-10-15
- Automating bug report fixes: a Bot triages DMs into a Cursor project — msfeldstein · 2026-10-15