CORE GUIDE

PATTERNADVANCED7 min read

Delegation state

Delegation state defines what information, ownership and progress move from a parent agent or workflow to delegated workers and what evidence must return before the parent can safely continue.

Mental model

Delegation is a state-transfer contract. A worker needs enough task state to act independently, but shared state should be explicit so concurrent agents do not mutate hidden assumptions or overwrite one another's progress.

Why it matters

Multi-agent systems often fail at handoffs rather than reasoning: workers receive incomplete context, two agents edit the same state, or the coordinator cannot tell which result is authoritative. Explicit delegation state makes ownership, isolation and join conditions testable.

01

Package task state and merge results deliberately

For each delegated unit, define immutable inputs, worker-owned mutable state, allowed shared resources and the evidence required at completion. Give concurrent workers separate namespaces or versions, then merge through explicit join logic that can reject conflicting or incomplete results. The parent should resume from returned evidence, not from assumptions about what the worker probably did.

02

Example: two research workers overwrite the same notes

A coordinator delegates market and technical research to two agents that share one mutable notes object. The second worker overwrites sources added by the first, yet both report success. Isolating worker outputs and joining them through a schema with source provenance preserves both result sets and exposes conflicts for verification.

Common failure modes

  • Delegating a goal without defining which state the worker owns.
  • Letting parallel workers mutate the same implicit memory or scratchpad.
  • Treating worker completion as enough without a returned evidence contract.

Engineering heuristics

  • Separate immutable task inputs from worker-owned mutable state.
  • Use versioned or isolated state for parallel delegation.
  • Define join and conflict rules before launching workers.

Takeaways

  1. 01Delegation moves state and ownership, not just instructions.
  2. 02Shared mutable context creates hidden coordination risk.
  3. 03A worker should return evidence that the parent can validate before continuing.

Reading evidence

UnseenPractice not completed

This records actions you actually took; it does not claim mastery, proficiency, or certification.

Used in

This Concept is reused across these canonical learning paths.

Related concepts from the Knowledge Graph

These relationships come from the canonical graph, not a separate Guide taxonomy.

Sequential, conditional and parallel topology trade-offsPREREQUISITEGuide
Independent verification, coordination cost and correlated failurePREREQUISITE