{"id":"b34442ea-2bcc-4932-98a0-5baf71e44aa8","slug":"vibecodingnights-agent-orchestration","kind":"mcp","name":"agent-orchestration","summary":"The coordination problems are fifty years old. The frameworks are new.","owner":"VibeCodingNights","repoUrl":"https://github.com/VibeCodingNights/agent-orchestration","license":"MIT","licenseTier":"permissive","trust":"community","stars":2,"match":null,"fewStars":true,"format":"mcp","createdAt":"2026-03-24T03:45:15.000Z","addedAt":"2026-10-21T08:34:00.450Z","logoUrl":"https://avatars.githubusercontent.com/u/248631883?v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["utilities"],"tags":["agent-orchestration","multi-agent","vibe-coding"],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-22T21:00:13.130Z","sample":false,"url":"https://conduits.dev/tools/vibecodingnights-agent-orchestration","language":"","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# Agent orchestration\n\nSpawn a team. Feel the overhead.\n\n---\n\nYou can spawn a team now. Team lead, three workers, shared task list. It looks like coordination. The agents message each other, claim tasks, report back. It feels like a team.\n\nThen one agent rewrites a file another agent is reading. A third agent gives confident advice based on a hallucination from the first. The \"team lead\" spends 30% of its tokens just routing — not thinking. You add a fourth worker and everything gets *slower*.\n\nThese aren't bugs in the framework. The frameworks give you plumbing. The hard problems are underneath.\n\n---\n\n## The two patterns\n\nEvery multi-agent system you'll ever build is some mixture of two ideas:\n\n**The supervisor** — One agent owns control flow. It decides who acts, when, with what context. Debuggable. Auditable. A single point of failure and a throughput bottleneck. pentagon.run makes this spatial and visual — a canvas where every agent has a desk.\n\n**The swarm** — No boss. Agents coordinate through shared state, environment signals, or direct peer-to-peer messaging. Coordination emerges from local rules. The ant colony. claude-peers-mcp is the minimal version: a broker, messages, and nothing else. MiroShark is the extreme: thousands of agents with personalities forming emergent consensus.\n\nMost production systems end up hybrid. A pipeline for structure, a swarm for exploration. A supervisor for synthesis, peers for the actual work.\n\n---\n\n## Primitives\n\nThe load-bearing concepts. Each one grounded in where the research actually is.\n\n- **[supervisor](primitives/supervisor.md)** — One agent to rule them all. When it works, when it breaks, what it costs.\n- **[swarm](primitives/swarm.md)** — No boss, emergent coordination. From ant colonies to a million simulated humans.\n- **[communication](primitives/communication.md)** — How agents actually talk. Text as bottleneck. KV-cache sharing as frontier.\n- **[memory](primitives/memory.md)** — Shared state is a lie. Consensus memory, context rot, the consistency models nobody's applying.\n\n## Exercises\n\nHands-on. Pick one, build something, break something.\n\n- **[peer-to-peer](exercises/peer-to-peer.md)** — claude-peers-mcp. Two sessions, no hierarchy, just messages.\n- **[supervise a team](exercises/supervise-a-team.md)** — pentagon.run or Agent Teams. Assign roles, watch the canvas, feel the overhead.\n- **[spawn a swarm](exercises/spawn-a-swarm.md)** — MiroShark. Feed it a document, watch hundreds of agents argue their way to consensus.\n\n## Provocations\n\nThe unsolved problems. The conversations worth having.\n\n- **[the coordination tax](provocations/the-coordination-tax.md)** — Measure it. When does adding an agent make things worse?\n- **[beyond text](provocations/beyond-text.md)** — What if agents could share thoughts instead of words?\n- **[your reports are stochastic](provocations/your-reports-are-stochastic.md)** — You're already an engineering manager. But your reports hallucinate.\n\n---\n\n**Tools for tonight:**\n\n| Tool | What it is | Pattern |\n|------|-----------|---------|\n| [claude-peers-mcp](https://github.com/louislva/claude-peers-mcp) | Peer-to-peer messaging between Claude Code sessions | Swarm / stigmergy |\n| [pentagon.run](https://www.pentagon.run/) | Spatial canvas for managing agent teams | Supervisor / mission control |\n| [MiroShark](https://github.com/aaronjmars/MiroShark) | Multi-agent simulation engine, hundreds of agents with personalities | Swarm intelligence at scale |\n\n**Bring** a codebase that needs parallel work, a document worth simulating reactions to, or just curiosity about why your agents keep stepping on each other.\n\n**Leave** with an intuition for when coordination helps and when it hurts — and the patterns underneath both.\n\nNo lectures. Just the structures underneath.\n"}