Skip to content
Multi-Agent Systems chapter route0102030405

Orchestration

Multi-Agent Systems

Anthropic and Cognition published opposite headlines about building with many agents, and agreed on the thing that matters most: the real problem is managing context, and the architecture should follow the shape of the task.

5 chapters41 slides27 minUpdated

Your route through the subject

0/41 slides completed · 0/5 chapters · 0 slides read

Start with: Single Agent versus Multi-Agent: The Core Tradeoff

Begin reading

The chapter map

Follow the route or choose a chapter.
  1. 010/7 completeSingle Agent versus Multi-Agent: The Core Tradeoff~4 minReadField checkPlate
  2. 020/8 completeThe Case For Multi-Agent: Anthropic's Research System~7 minReadPlate
  3. 030/10 completeThe Case Against: Cognition's "Don't Build Multi-Agents"~6 minRead
  4. 040/8 completeWhy Multi-Agent Systems Fail: The MAST Taxonomy~5 minRead
  5. 050/8 completeReconciling the Debate: It Was Never Single versus Multi~5 minRead

Chapters

  1. Chapter 1: Single Agent versus Multi-Agent: The Core Tradeoff

    A single agent keeps unified context and stays reliable but runs sequentially. A multi-agent system parallelizes work at roughly fifteen times the token cost.

    4 min7 slides: Single Agent versus Multi-Agent: The Core Tradeoff0/7 complete

  2. Chapter 2: The Case For Multi-Agent: Anthropic's Research System

    Anthropic's research system uses an orchestrator with parallel worker agents, and it beat a single-agent baseline by about ninety percent on research tasks.

    7 min8 slides: The Case For Multi-Agent: Anthropic's Research System0/8 complete

  3. Chapter 3: The Case Against: Cognition's "Don't Build Multi-Agents"

    Cognition argues long-running agents need one unified context to contain compounding errors, since split work across subagents causes conflicting decisions.

    6 min10 slides: The Case Against: Cognition's "Don't Build Multi-Agents"0/10 complete

  4. Chapter 4: Why Multi-Agent Systems Fail: The MAST Taxonomy

    MAST catalogs fourteen failure modes in three categories found in real traces: specification issues, inter-agent misalignment, and task verification failures.

    5 min8 slides: Why Multi-Agent Systems Fail: The MAST Taxonomy0/8 complete

  5. Chapter 5: Reconciling the Debate: It Was Never Single versus Multi

    Anthropic and Cognition are both right for their tasks: context engineering decides reliability, and read-heavy versus write-heavy work sets the architecture.

    5 min8 slides: Reconciling the Debate: It Was Never Single versus Multi0/8 complete