Week 1 · D1🔒 LockedAbout 4 hours

When Should You Go Multi-Agent: Three First-Hand Sources and an Experiment That Reproduces Them

Start with three first-hand sources that point in opposite directions: one says splitting a task across agents scored ninety percent higher, one says do not split, and one gives the criteria. Then run the same batch of cases through a single agent, independent parallel agents, and centralized orchestration, using an offline model that knows nothing about the topology, and watch centralized orchestration win on parallelizable work, multi-agent lose on tightly dependent work, and the token bill multiply.

Today's goals 0/3

Can state the conclusion and the premise of each of the three first-hand sources from Anthropic, Cognition, and Kim et al., and use parallelizability, dependency density, and the single-agent baseline to explain why all three are right
Can run the same batch of cases through the single-agent, independent, and centralized topologies, and read off the scoreboard who wins where, who loses where, and how many extra tokens it cost
Can explain why the model interface accepts nothing but one string of context, and name the class of conclusions that turns into a manufactured false green the moment it knows which topology it is in

Sign in, then link your WeChat mini program account on the account page. Linked accounts share one free article per day. Unlock additional articles with rewarded videos in the mini program.

Scan in WeChat to open Dayward AI. Open the same course and chapter, choose the rewarded video, then refresh this page after completion.

Unlocked — refresh to read Sign in / scan with WeChat →

Interview questions

When would you decline to split a task across multiple agents? Give your criteria.什么情况下你不会把一个任务拆给多个 Agent？给出你的判据。
Common in ChinaCommon overseasIntermediate#multi-agent#architecture-decision
🔒 Unlock this chapter to see the answer keys.
A multi-agent system costs far more than a single agent. What exactly does the extra money buy?多 Agent 系统比单 Agent 贵很多，多花的钱到底买到了什么？
Common in ChinaCommon overseasIntermediate#multi-agent#cost
🔒 Unlock this chapter to see the answer keys.
Why do independently answering agents merged at the end amplify errors more than a centralized orchestrator that consolidates their work?为什么多个 Agent 各自独立作答再合并，比有一个编排方统一汇总更容易放大错误？
Common in ChinaCommon overseasIntermediate#multi-agent#error-amplification
🔒 Unlock this chapter to see the answer keys.
When you run a multi-agent comparison experiment, why must the model interface not know which topology it is running in?做多 Agent 对比实验时，为什么模型接口不该知道当前用的是哪种拓扑？
Common in ChinaCommon overseasDeep dive#multi-agent#experiment-design
🔒 Unlock this chapter to see the answer keys.

Interview questions

Comments