WIRED: six agents over one day
One reporter rates six finished assistants on five daily task categories over about 24 hours, using personal Gmail and Calendar.
WIRED / Reece Rogers · Published 2026-10-27 · reviewed 2026-10-29
Scope and method
One user, one day, six products × five categories; no independent repeats or uncertainty intervals.
Reporter-assigned 1–5 ratings. Preserve the task matrix; do not manufacture an aggregate or convert ratings into completion percentages.
How PAEval uses it
Editorial experience evidence. Claude Duet remains a source label; it has not been merged with the existing Claude/Cowork profile.
Evidence family: wired-2026-10-27. Shared tasks and tracks are not independent evidence votes.
Reporter task ratings /5
Source-labelled products; higher ratings reflect the reporter's assessment, not measured reliability.
| Product | News | Lunch | Conflict | Travel plan | Evening |
|---|---|---|---|---|---|
| Dots | 2 | 5 | 5 | 4 | 4 |
| Muse | 3 | 4 | 2 | 5 | 4 |
| Grok Bot | 1 | 4 | 2 | 4 | 5 |
| Claude Duet (source label) | 4 | 3 | 4 | 3 | 2 |
| Gemini Spark | 5 | 1 | 1 | 3 | 2 |
| Instinct | 1 | 2 | 2 | 3 | 4 |
Limits and interpretation
- Plans and product maturity differ; subjective ratings reflect one user's preferences.
- Lunch carts and travel plans do not prove every product completed a purchase or reservation.
Sources and reuse
PAEval summarizes existing published evidence and runs no new tests. Public readability does not establish permission to redistribute full datasets, conversations or images.
Related product records
OpenAI dots · Meta Muse · Grok Bot · Gemini Spark · Instinct
All benchmark sources → · How to read outcomes and intervention →