Testing ChatGPT's Dot: A Real-World Agent That Works But Needs Chasing
anthonyaluna · reddit · 2026-10-01
A property-management operator spent days running real work through ChatGPT's new agent tool, Dot. It successfully updated a live HubSpot article, cleaned up old Outlook drafts after checking for sent replies, and configured site audits and keyword tracking.
But the failures were telling: it claimed a quote was accepted while the email sat in drafts, resurfaced already-answered requests, and sometimes simply lost track of unfinished work. The author compares it to onboarding a new hire — time saved must be netted against the time spent supervising and verifying.
His ongoing test criterion: the time saved has to include the time spent checking the work.
Related event: OpenAI DevDay Launches Persistent Agent Dots to Mixed Reviews(50 posts)→
More from Apps
- Open-Source iPhone Use Agent Poki Ships Watermark Fix in 0.4.2 Beta — Scobleizer · 2026-11-15
- Two prompt rules to kill the AI voice: keep it under 3 sentences, no rhetorical endings — yihui_indie · 2026-11-15
- Japan Gov spotlights miwo, the AI app that reads 1,000-year-old kuzushiji script — tkasasagi · 2026-11-15
- Gemini in Chrome adds Mac menu bar presence and global shortcut, mocked as a sign of Google's desperation — daluoseo · 2026-11-15
- DeepSeek open-sources chat-demo, a backend-free browser chat UI for its API — huangyun_122 · 2026-11-15
- Thai Researcher Builds AI App Making 1,000-Year-Old Japanese Classics Readable to All — tkasasagi · 2026-11-15