LangSensei

Dumb Routers, Smart Specialists

A pattern for multi-agent task execution where coordination is mechanical and intelligence lives at the edges.

This is an idea file. It describes a general architecture pattern, not a specific tool. Copy it, adapt it, build on it. The goal is to communicate the core insight; your implementation will vary.

---

The default architecture for multi-agent systems is a smart orchestrator. You build one powerful agent — the "brain" — that understands the task, plans the approach, delegates to sub-agents, handles errors, and synthesizes results. This agent is typically an LLM itself: it reasons about what needs to happen, picks the right tools, and coordinates everything.

LangSensei

Cognitive Scaffolding for Autonomous Agents

A pattern for making LLM agents sustain disciplined reasoning across long tasks.

This is an idea file. It describes a general pattern, not a specific tool. Copy it, adapt it, instantiate it with your agent of choice. The goal is to communicate the core insight; your agent can figure out the specifics.

---

Most people's experience with LLM agents and complex tasks looks like this: you give the agent a task, it starts well, makes progress for a while, then gradually drifts. It forgets the original plan. It skips steps. It redoes work it already did. It loses track of what it found. By the 50th tool call, the agent is operating on vibes — reacting to whatever's in its immediate context window rather than following a coherent strategy.