E2 · Task-ledger orchestrator (Magentic)

Principle

A variant of orchestrator and subagents (E1): the lead keeps a task ledger (known facts, plan) and a progress ledger. When execution stalls, it replans. The delegated agents act on external systems: web, files, terminal.

The problem it solves

On an open-ended problem whose path is not known in advance, a plan fixed at the start goes stale at the first surprise.

When to use it

• Open-ended problem with no predefined path, where a human must be able to review the plan before or during execution.
• Progress must be tracked explicitly, with blockage detection and recovery from errors.

When to avoid it

• The path is deterministic or the complexity is low: a workflow (B1, B6) or a plain orchestrator (E1) costs less.
• Speed matters: the pattern is not built for it (“not optimizing for speed”, Microsoft), and Microsoft rates its cost as the least predictable.

Synonyms by vendor

• Magentic, dynamic orchestration, task-ledger-based orchestration, adaptive planning (Microsoft)
• Magentic-One (Microsoft Research (2024))
• Deep Agents (LangChain)

Neighbouring patterns

• E1 · Orchestrator and subagents
• E5 · Group chat and debate
• F1 · Human in the loop
• B6 · Custom state graph

Sources

• Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks — Microsoft Research (A. Fourney et al.), arXiv, 2024-11-07
• AI agent orchestration patterns — Microsoft, Azure Architecture Center, 2026-02-12
• Why Do Multi-Agent LLM Systems Fail? — Cemri et al., arXiv, 2025-03