AI Research Laboratory

The Subagent centric
coding agent.

Research focused on multi-agent coordination, entropy minimization, and inference optimization.

> curl -fsSL https://subagents.com/install | sh

Research Directions

01

Subagent Entropy

Minimizing information loss during agent-to-agent delegation. When context transfers between agents, understanding degrades—we measure and reduce this.

StatusActive
02

Task Decomposition

Specialized subagents outperform monolithic execution. We study where decomposition boundaries yield optimal results.

StatusActive
03

Speculative Decoding

Task-specific speculation patterns reduce latency without accuracy loss. Each task type develops learned decoding heuristics.

StatusIn Progress

Publications

2025

On the Information-Theoretic Limits of Multi-Agent Coordination

Coming Soon
2025

Speculative Decoding with Task-Conditioned Draft Models

Coming Soon
2025

Optimal Subtask Boundaries for Agent Decomposition

Coming Soon
"Indifference to people and the reality in which they live is actually the only true deadly sin in design."

— Dieter Rams

Primary
Dieter 0