What You Need to Know
A multi-agent system splits one task across several agents. In the production-proven orchestrator-worker shape, a lead agent (the coordinator) analyses the goal, develops a strategy and decomposes it into subtasks. It spawns subagents that explore in parallel, each with its own context window, and synthesises the results they return. Two skills apply: knowing when multi-agent is justified, and orchestrating so the classic failure modes cannot occur.
Take an MOT garage comparing five parts suppliers on price and delivery. Each supplier can be researched alone.
When Multi-Agent Is Justified and What It Costs
Multi-agent architectures excel at breadth-first work that parallelises cleanly:
- research across independent directions
- analysis of many independent units
- tasks exceeding a single context window
- work needing distinct specialisms with different tools and permissions
Subagents work alone, so the split pays off when no direction needs another's result, as with the garage's suppliers. Multi-agent struggles where agents must share context tightly or depend on each other's intermediate results. Anthropic's engineering assessment: most coding tasks have fewer truly parallelisable parts than research does.
Anthropic's production data puts agent interactions at roughly 4× the tokens of chat, and multi-agent systems at roughly 15×. Each subagent spends its own tokens. The task's value has to pay for that increased performance, as a better supplier contract might for the garage. A proposal with no value case is incomplete (see the business value pillars in objective 1.6).
Key Concept
The exam tests restraint. Multi-agent for a task that fits one context window is a wrong answer, however competent the orchestration. Justify the architecture first.
The Orchestrator-Worker Architecture
Subagents share no memory. A subagent does not inherit the coordinator's conversation or see other subagents' findings. Every piece of context it needs must be passed explicitly in its instructions. The third supplier's subagent cannot know the garage wants only brake and exhaust parts unless told. Any design that assumes ambient shared knowledge between agents is broken by construction.
Delegation quality determines system quality. The brief is all a subagent knows. Anthropic's research-system team found that without detailed task descriptions, subagents duplicate work, leave gaps, or fail to find necessary information. An effective subagent brief contains:
- an explicit objective
- an expected output format
- tool and source guidance
- clear task boundaries
The third supplier's brief asks for prices, each with a source, and excludes the other four suppliers. "Research this topic" alone is a defect.
Failure Modes and Their Root Causes
Failures fall into a short list, each with a signature:
- Narrow decomposition. The coordinator splits a broad goal into subtopics covering only part of the space. Output has systematic gaps, such as no subagent checking delivery. The root cause is the coordinator's decomposition, not subagent capability, and not agent count.
- Duplicated work. Subagent scopes overlap because boundaries were never partitioned. Two subagents both price brake pads at the same supplier. Signature: near-identical sections in the synthesis.
- Effort mismatch. Ten subagents spawned for a simple fact-check, such as one supplier's delivery charge. Scale agent count and tool budget to task complexity. Simple fact-finding wants one agent and a handful of tool calls. Only genuinely complex research justifies many subagents with divided responsibilities.
- Lost attribution. Findings passed between agents as free prose lose their sources. The report names the cheapest supplier and nobody can trace the price. Structured claim and source data must cross every agent boundary, or synthesis produces confident, uncitable claims.
Duplication plus gaps in one output is the signature of unpartitioned decomposition. Overlapping scopes cause the duplicates, and uncovered space causes the gaps. Both start in the coordinator's briefs, so the fix is its briefing strategy. More agents with the same overlapping briefs produce more overlap. The garage gives each part and supplier to one subagent.
Orchestration Strategy
- Parallel where independent, sequential where dependent. Parallel spawning cuts wall-clock time for independent subtasks; dependencies force sequencing. The five supplier checks run at once, and the comparison waits for all five.
- Durable state and checkpointing for long-running systems, so a mid-run failure resumes from a checkpoint rather than restarting. At 15× token cost, restarts are expensive by definition. If the run fails on the fifth supplier, the first four results are kept.
- Observability of agent decisions: which agent did what, with which tools, at what cost. The garage can see which subagent priced which supplier, and what that cost. (Monitoring depth is Domain 3 and 4 territory; the design obligation lands here.)
- Through-the-coordinator remains the default communication topology. It concentrates control and observability in one place. Peer-to-peer agent protocols exist, and choosing between them is examined in Domain 3.
In the exam
At the Professional level the exam tests both skills. It tests the second largely through diagnosis questions: a broken multi-agent system, four candidate root causes from the failure modes above, one correct. The two architecture facts carry most of the weight. "Add more subagents" and "use a larger model" are the planted wrong answers.
Exam Traps
Diagnosing overlapping or missing report sections as a subagent capability problem
Duplication plus gaps is the signature of unpartitioned coordinator decomposition. A stronger model given the same overlapping briefs produces the same overlap — trace the failure to its origin.
Assuming subagents can see the coordinator conversation or each other's findings
Subagents share no memory. Everything a subagent needs must be passed explicitly in its brief. Designs that rely on ambient shared context are broken by construction.
Proposing multi-agent for a task that fits comfortably in one context window
The roughly 15x token multiplier, added latency, and orchestration complexity buy nothing a single agent could not do. The exam rewards restraint and a value case, not architectural ambition.
Fixing systematic coverage gaps by adding more subagents
More agents with the same unpartitioned briefs produce more duplication, not more coverage. The fix is mutually exclusive, explicitly bounded briefs from the coordinator.
Practice Scenario
A due-diligence system uses a coordinator plus six parallel research subagents. Final reports keep arriving with three sections nearly identical to each other and two risk areas missing entirely. What is the root cause?
Build Exercise
Design a Bounded Orchestrator
What you'll learn
- When a multi-agent architecture is justified and when it is not
- How to write a coordinator decomposition strategy with partitioned scopes
- What a complete subagent brief contains
- How to preserve attribution across agent boundaries
- How to write the value case that clears the token multiplier
- Pick a genuinely parallelisable research task — competitor analysis across five companies works well — and state why it parallelises
Why: The justification step is the one most designs skip. Breadth-first, independent directions is the pattern that earns multi-agent cost.
You should see: One sentence naming the independent directions and confirming no direction depends on another finding.
- Write the coordinator brief: the decomposition strategy, the number of subagents, and an effort budget per subagent
Why: Effort scaling is a designed property. Anthropic embeds explicit guidelines — simple fact-finding gets one agent and a handful of tool calls; complexity earns more.
You should see: A short paragraph: how the goal splits, how many subagents, and a tool-call budget per subagent with a sentence of justification.
- Write one subagent brief in full: objective, output schema with source-attribution fields, tool guidance, and explicit scope boundaries
Why: Delegation quality determines system quality. The complete brief is the artefact that prevents duplication, gaps, and lost attribution simultaneously.
You should see: A brief another person could execute without asking questions: one objective, a JSON output schema including a sources field, named tools, and a boundary list.
- Define the synthesis step: how the coordinator merges structured findings, and what happens on conflict between subagent claims
Why: Synthesis is where lost attribution and unresolved conflicts surface. Designing the merge rules in advance keeps the final output citable and consistent.
You should see: Merge rules in three or four lines: how findings combine, how conflicts resolve, and how sources carry into the final report.
- Write the value case: estimate the token multiplier against a single-agent baseline and state why the task value clears it
Why: The professional argument for multi-agent is economic, not aesthetic. Anthropic reports multi-agent at roughly fifteen times chat tokens — the value must pay for it.
You should see: Three lines: single-agent cost estimate, multi-agent estimate, and the value statement (time saved, coverage gained) that justifies the difference.
Sources
- How we built our multi-agent research system — Anthropic
- Subagents in the SDK — Anthropic
- Building Effective AI Agents — Anthropic