When splitting a job across several AIs actually helps — and when it backfires
Splitting a task across several AI assistants only helps when each can pass a short, clear note to the next. If the next AI needs details that get lost in the handoff, one big AI holding the full picture usually wins.
Why it matters
Companies are building 'AI teams' where different AI programs each handle one piece of a job, hoping this beats one AI doing everything. This paper says: it depends on the job. For tasks that split cleanly (search first, then pick a product), the team setup helps. For tasks where every small detail matters, like planning a trip within a strict budget, sharing the full context matters more than dividing the work. Stronger AI models suffer more from lost details, so bigger models actually gain less from teamwork.
Who's behind it: Wendi Yu and colleagues, Texas A&M University. Funded by ARPA-H and Texas A&M University research programs.
Summary by the Lemma AI · how we grade