BuildSeniorData / MLAuto-graded

Fan out to agents and merge what succeeded

Judgement about partial failure: work split across many workers should degrade gracefully rather than collapse entirely.Real AI systems rarely run a single agent. They fan out — one sub-agent per URL to summarise, per file to review, per sub-question to answer — then gather the results and reduce them into one answer. That orchestration primitive is the most reused piece of any multi-agent system, and the catch is that a sub-agent is just a flaky function call: one of them will raise. Make the fan-out attempt every item, capture the failures, and reduce over what actually succeeded.

35 min

python

Start lab

You're done when

A worker that raises does not bring down the fan-out.
Every item is attempted, regardless of earlier failures.
Failures are captured as data rather than propagated as exceptions.
The reduce step sees only the successful results.
The reduction itself is correct.
Results are returned in the order the inputs were given, not the order they happened to finish.

What it measures

Agent engineeringBuilding the fan-out and reduce primitive that multi-agent systems are assembled from.

Context engineeringDeciding what the reduce step should see when some branches failed, so the final answer is not silently built on partial input.

Designing for partial failureTreating one branch failing as expected, preserving input order despite concurrent completion, and handling the empty case.

Your environment

Terminal + editorA full shell in the browser

Claude CodePre-authenticated, no key setup

Isolated microVMYours alone, wiped afterwards

Start this lab

35 min

Start labYou'll sign in to Anthropos to start. The lab opens in a private cloud workspace with Claude Code ready to use.

AI Labs is part of Anthropos Workforce. Not on your plan yet? Book a demo