Research
"97% of agent projects fail to scale beyond isolated demos." — IDC
🎯 The Problem
As Multi-Agent Systems grow from 1 agent to 50, they hit a wall: coordination costs grow super-linearly.
Current frameworks focus on connecting agents (MCP) and orchestrating them (LangGraph)—but nobody is solving how they coordinate at scale.
The result? Agents interrupt each other, miss handoffs, contradict themselves, or proceed without approval.
💡 The Solution: Harmonic Coordination Theory
HCT provides the missing coordination semantics layer. Borrowing from musical ensemble coordination—which has solved the problem of coordinating 100+ performers without central state for centuries—HCT introduces:
| Component | Description |
|---|---|
| 7 Coordination Signals | Cue, Fermata, Attacca, Vamp, Caesura, Tacet, Downbeat |
| 6-Layer Architecture | From Reference Frame to Listening feedback |
| 15 Diagnostic Patterns | Named failure modes with solutions |
📄 The Research
📚 Deep Dives
| Resource | Description |
|---|---|
| HCT Signal Specification | Full protocol spec for MCP & A2A extensions |
| The MAS Matrix | 70+ SOTA techniques mapped to HCT layers |
| Patterns Library | 15 diagnostic patterns for agent failures |
📦 Open Source
Two protocol extensions for MCP and A2A. Seven coordination signals. Four languages each.
🔗 Related Tools
The research informed practical tools built with the same rigor:
- envcheck — Environment validation with research-backed quality
- NerfStatus — Scientifically-grounded LLM quality monitoring
- SpeakOps — Real-time AI coordination applied to speech coaching
✍️ Related Writing
- SeekingSota — Weekly essays on AI × Human identity
- "The 97% Problem" — Why agent projects fail to scale
HCT Spec & RFC
The canonical specification for HCT signals and performance parameters. Source of truth for all implementations.
HCT-MCP Signals
The coordination layer for MCP. Available in Python, TypeScript, Rust, and Go.
HCT-A2A Extension
HCT coordination signals for Google's A2A Protocol. Polyglot samples in Python, TypeScript, Rust, and Go.
HCT Patterns Library
15+ diagnostic patterns for multi-agent coordination failures. Enterprise-ready solutions.
HCT Core
Reference implementation of the HCT signal bus. Layers 3-5 coordination primitives.
HCT Benchmarks
Benchmark suite for HCT coordination patterns. Reproducible performance metrics.