A **multi-agent** system splits a task across several agents, each with its own prompt, tools, and context. A single agent does everything in one loop. The interview question is when the split actually pays off.
Three conditions favour the split. First, the subtasks are genuinely parallel, so several agents can work at the same time with no waiting. Second, specialist agents clearly beat one generalist, for example a coding agent plus a separate research agent. Third, the work is too large for one context window, so you fan it out and merge the results.
Two of the listed options are traps. A strictly sequential chain gives no parallelism, because each step waits for the one before it. You pay the coordination cost and get none of the speedup. The claim that orchestration is simpler than the task is backwards, since coordination always adds complexity.
The takeaway is that single-agent is the default. You move to multi-agent only when one of those three real benefits outweighs the extra coordination cost and the risk of errors spreading between agents.
Multi-agent architectures pay off in three situations. The first is genuine parallelism, where subtasks have no dependencies and several agents run at once. The second is specialisation, where distinct agents with their own tools and contexts outperform one generalist. The third is scale, where the work exceeds a single agent's context window and must be fanned out.
The correct answers are the parallel subtasks option, the specialist advantage option, and the scale option. Each gives a benefit that can outweigh coordination cost.
The two distractors fail for clear reasons. A strictly sequential chain offers no parallelism, since every step waits on the previous one. You pay full coordination overhead and gain nothing, so a single agent or a plain chain is better.
The claim that orchestration is simpler than the task is also wrong. Orchestration always adds complexity. The whole tradeoff is that the multi-agent benefit must exceed that added cost. The default should be a single agent, escalating only when one of the three real conditions clearly holds.
Quick reference
TL;DR
Multi-agent wins only when subtasks parallelise, specialist models beat a generalist, or volume overflows one context. Otherwise it just adds coordination cost.
Imagine you have a big project due. If the work splits into pieces that people can do at the same time without waiting on each other, hiring a team helps a lot. If one teammate is great at math and another at writing, splitting by skill also helps. But if the work is one long chain where step two needs step one finished first, a team does not finish faster. You just add the cost of everyone talking to each other and handing things off. AI agents are the same. A single agent is the simple default. You only split into a multi-agent team when the task genuinely parallelises, when specialists clearly beat a generalist, or when one agent simply cannot hold all the work in its head at once.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
The task is a strictly sequential chain where each step depends on the previous step's output. There is no parallelism to exploit, so you pay coordination overhead for zero latency gain.
A single capable model already handles the whole task within one context window. Adding agents only introduces routing, handoffs, and merge logic with no offsetting benefit.
Reliability is critical and the subtasks are tightly coupled with shared mutable state. Per-agent errors compound and one bad handoff can corrupt the merged result.
The team is still prototyping and debugging speed matters more than peak throughput. Multi-agent traces are far harder to debug than one linear single-agent loop.
Quick scan · 2–4 min
Everything important, quickly.
State that single-agent is the default, then name the three conditions that justify multi-agent: parallel independent subtasks, specialist advantage with context isolation, and scale beyond one context window. Explain why a sequential chain and 'simpler orchestration' are distractors, and close with coordination overhead and error propagation as the costs you weigh against the benefit.
Side-by-side
Dimension
Single-agent
Multi-agent
Latency on parallel work
Serial, slow
Fan-out, fast
Coordination overhead
None
High: routing, handoffs, merge
Error propagation
Contained to one loop
Compounds across agents
Context isolation
Single shared context
Clean per-agent contexts
Default choice
Yes, the baseline
Only when a real benefit wins
Real-world examples
Real products, models, and research that use this idea.
1Anthropic's multi-agent research system uses an orchestrator that spawns parallel subagents to explore independent search branches, reporting large latency wins over a single agent on broad research tasks.
2OpenAI's Swarm and the newer Agents SDK model handoffs where a triage agent routes to specialist agents, a specialisation pattern rather than a parallelism one.
3LangGraph supervisor architectures coordinate worker agents over a shared typed state, exposing the coordination overhead and termination logic that single-agent loops avoid.
4CrewAI assigns role-specialised agents (researcher, writer, reviewer) to a shared goal, useful only when the role split genuinely beats one generalist agent like Claude Opus 4.7.
Interviewer follow-ups
What an interviewer would ask next. Try answering before peeking at the approach.
QHow would you decide the boundary between subagents in a research task, and what makes a clean split?
Sign in or go Pro to see the approach for this follow-up.
QWhy does a multi-agent system often cost more total tokens than a single agent on the same task?
Sign in or go Pro to see the approach for this follow-up.
QHow do you stop one misbehaving subagent from corrupting the orchestrator's final answer?
Sign in or go Pro to see the approach for this follow-up.
Don't say this
Red flags and common mistakes that signal junior thinking. Click to expand.
Most common mistake
Reaching for a multi-agent design by default. A team of agents adds coordination cost and error propagation, so a single agent should be the baseline you must justify moving away from.