Supervisor / Orchestrator-Worker Pattern

One lead agent plans and delegates; specialized workers execute in parallel contexts and report back. This is the pattern behind Anthropic's Research system (Claude Opus 4 as lead, Sonnet 4 as subagents), measured at +90.2% over single-agent Opus 4 on internal research evals. Anthropic's engineering post reports that 80% of the variance on BrowseComp is explained by token usage alone — multi-agent wins largely because each subagent gets a fresh context window. This lesson builds the supervisor pattern from the primitives and covers the 2026 engineering lessons from production deployments.

Learn + Build 1.3 hr Python (stdlib, `threading`)

Prerequisites: Phase 16 · 04 (Primitive Model)

#Problem

Research is the prototypical task that single-agent systems fail. You ask "what changed in multi-agent systems between 2023 and 2026?" A single agent reads five papers sequentially, fills half its context with their text, and then has to reason about all of them together. It forgets the first paper by the time it reaches the fifth. It cannot parallelize.

The supervisor pattern fixes this: one lead agent plans the search, delegates each sub-question to a worker, and synthesizes. Each worker gets its own 200k-token window for a narrow question. The lead never sees the raw papers — only the worker summaries.

Anthropic's production Research system reports +90.2% on internal research evals vs a single Opus 4. The same post notes that 80% of the BrowseComp variance is explained by token usage alone. Fresh context per subagent is the main mechanism.

#Problem

Keep going. The rest of this lesson is one step away.