Agent swarms are the next scaling law: 32 agents hit 19x speedups on NumPy tasks
andrew_n_carr · x · 2026-11-18
@pirroh argues agent swarms are the next scaling law in AI — after pretraining and reasoning, we now scale the number of cooperating agents. On 13 NumPy optimization tasks, 1 agent hit up to 4x speedups, 16 agents 17x, and 32 agents 19x. Replit launched Missions to put swarms to work, and Andrew Carr adds this is also where most safety and steerability challenges will come from.
Related event: Agent Swarms Hailed as AI's Next Scaling Law(2 posts)→
More from coding & agent
- Bryan Johnson Uses a Devin Bot to Text His Mom a Daily Health Report — marvinvonhagen · 2026-11-18
- Every launches Everycut: agent-native Mac screen recorder where every control is an agent tool — kieranklaassen · 2026-11-18
- ForgeGUI launches browser-based Roblox Studio this Saturday, claiming 1M+ developers — imjustnewatai · 2026-11-18
- Rillet Labs: switching from multi-agent to single-agent harness doubled accounting accuracy — mattturck · 2026-11-18
- Early access reviewer: GPT-6.2 Astra is a powerhouse agentic video editing partner — craigsdennis · 2026-11-18
- Microsoft launches Autopilot, a persistent personal agent working across Teams and Outlook — testingcatalog · 2026-11-18