Simulated survey of 1,884 AI personas: scheduling and email top agent use cases
@thisiskp used Artificial Societies' Radiant simulation platform to model tech-Twitter users with 1,884 AI personas built on real human behavioral data, pre-testing attitudes toward personal AI agents before real decisions—and got surprising results: what users want most isn't coding help, but calendar scheduling and email handling.
Confirmed
- Among tasks users most want to hand to an agent: 73% chose scheduling, 63% chose email handling; coding was only 31%, ranking fifth.
- If an agent could reliably handle just one thing: scheduling 34%, email 33%, research 9%, coding just 8%—about two-thirds of the value concentrates in calendar and inbox.
- Willingness to grant autonomy is nearly split down the middle: 43% want the agent to prepare work and wait for approval before taking external actions, 43% accept it handling routine actions independently within self-set rules, and only 9% would let it handle most actions independently.
- Users' biggest fear is irreversible erroneous actions; the author stresses that "trust is the product," with key design questions including whether the agent can handle calendar back-and-forth on the user's behalf, whether users can control and retract outgoing content, and whether its work actually saves time.
Why it matters
- The survey methodology itself—using AI personas to pre-test user attitudes before real decisions—offers a low-cost experimental path for product validation.
- The results suggest to developers that personal agents' value peaks in mundane chores like calendars and inboxes rather than showy use cases like coding; easy reversibility, controllability, and privacy protection are the key design constraints for earning user trust.
2026-10-14 ~ 2026-10-14 · 6 related posts
Primary sources
- Survey of 1,884 simulated tech Twitter users: scheduling beats coding for AI agents — thisiskp_ · 2026-10-14
- Survey of 1,884 AI personas: scheduling tops personal agent wishlist, coding ranks 5th — thisiskp_ · 2026-10-14
- [source] Simulating user research with 1,884 AI personas: scheduling and email are agent must-haves — thisiskp_ · 2026-10-14
- Agent value concentrates in scheduling (34%) and email (33%); coding gets just 8% — thisiskp_ · 2026-10-14
- Agent autonomy split evenly: 43% want approval gates, 43% accept rule-bounded autonomy — thisiskp_ · 2026-10-14
- [source] Trust is the product: undoability (60%) and privacy (56%) top personal agent fears — thisiskp_ · 2026-10-14