A 311 system uses a classifier agent and a work-order agent that must share intake context. Which orchestration pattern fits?

A 311 system with a classifier agent and a work-order agent that must share intake context should use multi-agent orchestration with a shared memory scratchpad for that intake context. Shared memory lets specialists cooperate without re-asking the resident for the same location, category, and description after classification. Orchestration coordinates who acts when while the scratchpad carries structured intake forward into work-order creation. Sequential handoff passing the classification label into the work-order prompt gives the fulfillment agent only a label, not the full intake context, so key details are lost and the resident may still be re-questioned. Supervisor delegation issuing discrete task prompts to classification and fulfillment agents splits intake into isolated steps without a shared scratchpad, so context can be dropped between prompts and specialization is undermined by fragmented orchestration. Peer message exchange sharing work-order identifiers for downstream fulfillment passes only an identifier, not the intake details, so the work-order agent lacks structured context and the resident's original information is lost unless re-entered. On the exam, multi-agent orchestration plus shared scratchpad fits split intake/work-order roles; distractors pass only a classification label, delegate discrete prompts without shared memory, or exchange identifiers without intake context. Caveat: scratchpad contents should be schema-validated—unstructured free text between agents recreates ambiguity the classifier already resolved. Operational check: run an intake through the classifier into the work-order agent, verify the scratchpad carries fields without re-prompting the resident, and confirm neither agent requires identifier-only peer messages for the happy path.