The real challenge of always-on agents: permissions, approvals, and who owns the judgment call
lmoroney · x · 2026-10-01
OpenAI's Dots, shipped at DevDay, are agents with their own cloud computer that keep working between chats — that part is real. But what author Laur Moroney is watching is the boring stuff:
- how to design permissions and approvals
- who owns the judgment call when an agent acts on your behalf
- an agent that can do anything is only useful if you can say exactly what it may do alone
The takeaway: the skill that matters now is designing the system around the agent, not the agent's raw capability.
More from AGI Musings
- Researcher warns successionism will go mainstream once AGI feels real, urges resistance — burny_tech · 2026-10-01
- AI is disrupting education video, indie games, music videos and VR 'in just days' — Dr_Singularity · 2026-10-01
- When robots erase physical strength's value, an unexpected gender joke goes viral — VraserX · 2026-10-01
- State of Clinical AI 2026 report published in BMJ Digital Health & AI — davidjhwu · 2026-10-01
- Walled gardens are a tourniquet: platforms must ship agents, models, or become infra — yenkel · 2026-10-01
- Dev argues a superintelligence would easily out-debate any open-minded human — voooooogel · 2026-10-01