WIRED tests OpenAI's always-on Dots agent: wrong name, failed captcha, buggy first run
nordicinst · x · 2026-10-07
WIRED's Reece Rogers spent days with OpenAI's newly announced always-on agents, called Dots, tasking one with buying a couch:
- The agent got his name wrong immediately ("Hello, Connor"), undermining trust with his skeptical partner
- Dots is designed to feel like texting a friend, with a virtual browser to book flights, cancel reservations, or shop
- Real-world hiccups: mistranscribed speech, and offering to solve captchas it couldn't complete
- He frames agents (Meta's Muse included) as leaping from early-adopter niche tools to mainstream products—rough now, but future iterations may overhaul how users interact with the internet
Bonus color: even agents struggle with Nordic trust norms and EU privacy checks.
More from Apps
- Grok Bot stays reachable after weekly limit, letting you resume or download tasks — nima_owji · 2026-10-07
- Battle Chess 3D: Every Capture Is a Duel, Built and Playable via Agent APIs on Spawn — jasonkneen · 2026-10-07
- Power users are far ahead: most people still have no idea how to use LLMs — iruletheworldmo · 2026-10-07
- Claude subscription rumored to add Midjourney and Suno, with Opus 5.5 powering tasks — op7418 · 2026-10-07
- Atlassian launches AMP: agents join the team, and you can now send them a Loom — davidhoang · 2026-10-07
- Full guide: shooting a 26-piece product catalog with Nano Banana 2.1, one locked prompt — socialwithaayan · 2026-10-07