AI Agent Swarms Are Amazing
summary
Demonstrates building agent swarms with OpenClaw/Grimlock experiments and OpenCode, running up to 12 subagents in parallel via a Pi harness, using MDX/MDSVX “.brain” notes and a publishable wizard site, handling a support inbox peaking at hundreds of emails per day through a feedback-and-memory loop, deploying an Agent Dungeon on Cloudflare Durable Objects and Durable Artifacts for persistent state, reports an Opus45-driven productivity spike, an Anthropic usage restriction, and warns that the orchestration tax and human attention remain the real bottlenecks.
AI-written summary · about
chapters (16)
- 0:00Introduction: Developer Superpowers in the AI Era
- 0:52The Dream of Agentic Swarms
- 1:56How AI Changed My Development Workflow
- 2:48Harness Engineering Explained
- 4:09Building and Owning Your AI Harness
- 5:56Swarm Tools and Multi-Agent Experiments
- 7:37AI Skills, Agent Personalities, and OpenClaw
- 9:28Pi: The Minimal Agent Harness
- 12:24Extensions, Subagents, and Custom Workflows
- 14:04Pi Notes: Building Memory for Agents
- 16:24The AI-Powered Support Inbox
- 19:15Agent Dungeon: Simulating Teams with AI
- 20:55Cloudflare, Durable Objects, and Agent Infrastructure
- 22:06The Reality of Agent Swarms
- 22:49The Orchestration Tax and Human Attention
- 23:13Final Thoughts: Slow Down and Build Intentionally
related talks
49mAI Agents: Collaboration in Real-Life Applications
30mAI Agents Unchained
40mAgents in Action: Advancing Open Source AI Missions
25mDirecting a Swarm of Agents for Fun and Profit
07h 47mAI Agents Conference 2026 - Day 5
26mAI powered automation & multi-agent orchestration
44mThe Swarm Next Door: Running Parallel AI Agents on a Real Android Codeba…
40mAI Agents in Practice