Practiced one out loud? Paste your answer into the free self-check and get a score in 30 seconds
What the interviewer is testing
Whether you resist adding complexity without evidence, and whether you know the specific conditions under which splitting work across agents pays off.
Answer framework
- Default: start with a single agent and a good tool set; it is cheaper, faster and easier to debug.
- Multi-agent helps when work splits into independent parallel parts, such as researching several sources at once.
- It also helps when sub-tasks need very different tools, permissions or instructions that would crowd one prompt.
- Costs: more tokens, more latency from hand-offs, lost context between agents and harder debugging.
- Decide by evaluation: compare quality, cost and latency of both designs on the same tasks.
Sample answer
I start with a single agent unless there is a clear reason not to. One agent with well-designed tools keeps all context in one place, costs less and is much easier to trace when something goes wrong.
Multi-agent setups earn their cost in a few situations. The clearest is parallel work: a research task that needs to read ten sources can fan out to sub-agents, each reading in its own context, while an orchestrator combines their findings. Another is separation of concerns: if one part of the job needs database write access and another only reads public pages, separate agents with separate permissions and focused instructions are safer and simpler than one huge prompt.
The costs are real. Total token use can grow several times, hand-offs lose context, agents can disagree or duplicate work, and failures are harder to locate. So I would prototype both designs on the same evaluation tasks and switch to multi-agent only when it clearly improves quality or speed enough to justify the extra cost.
Likely follow-ups
Common mistakes
- Choosing multi-agent because it sounds more advanced.
- Ignoring the multiplied token cost.
- Passing full transcripts between agents instead of focused summaries.
- No tracing across agents, so failures cannot be located.
The 0 to 1 AI Engineer Interview Playbook covers the answer frameworks, system-design walkthroughs and an 8-week plan.
Get the playbook 40% off with code EARLYBF until Nov 3, 11:59 PM PTEARLYBF: 40% off until Nov 3Smaller start: Score your answers against 50 rubrics, $9. 50 AI engineer questions, one rubric each, PDF on Gumroad.
Interviewing for PM or TPM? See the full PM package, $59 on Gumroad: a resume system, a PM, TPM and PGM interview handbook, a 1:1 cheat sheet and a PM playbook. PM and TPM content, not AI engineer prep.
Related questions
Based on question #47 in the full question bank.