Perplexity’s own research team compares its conversational assistant against its autonomous agent across Feb-May 2026, finding the agent runs ~48x longer per session, cuts task time 87% and cost 94%, and gets used for work people never attempted before: 23% of agent tasks had no assistant analogue, requests crossed occupational boundaries 59% of the time, and users bundled 32-60% more sub-tasks per request. The strongest claims are the runtime gap and the fixed-versus-marginal cost structure; the weakest are the novelty and sub-task numbers, which cannot distinguish new work from work that migrated off other tools.
Summarized by AI.
🔗 Research: How AI Agents Broaden the Scope of Knowledge Work
Leave a Reply