Hands-on Dots review: 5 hours of work used 1% of the $200 plan, but computer-use falls apart
ChrisGPT · x · 2026-10-04
Product veteran ChrisGPT (ex-Instagram, Ray-Ban glasses, X) publishes an openly OpenAI-biased review of Dots.
The good
- Five hours of work in the VM consumed just 1% of the $200 plan's usage
- Voice mode feels slightly more advanced than Advanced Voice Mode: adjustable style, slower delivery, even singing
- Strong within its own VM, especially balancing complex spreadsheet tasks
The bad
- Computer-use troubleshooting required constant chasing; it confused local PC tasks, CLI sessions and Codex local chats
- Showed a success checkmark in chat after a wrong password while the site displayed an error
- Claimed tasks were done before verifying, then contradicted itself; couldn't end voice calls on request
- Weak instruction following (persistent colons, repeating wrong cloud-chat approach)
He also finds GPT-6 lacks the promised persistence of continuing after being stopped, echoed by other early testers, and this carries into Dots. OpenAI employees have since reached out to help fix computer-use issues. Verdict: real potential, but poor follow-through — "a final harness for AGI with a subpar, broken orchestrator."
More from Apps
- Perplexity brings AI teammates to iOS and Android, callable from mobile app — _andychung · 2026-11-20
- OpenAI engineer points to ChatGPT macOS global dictation — but it only works inside the app — athyuttamre · 2026-11-20
- Backdoor matches job seekers with startups: 26K users, 58% hear back within two weeks, free — itsOmSarraf_ · 2026-11-20
- TwitchCon demo: agents edit stream highlights and build games with chat live — paw_lean · 2026-11-20
- R3 Code: A Native Rust Mac Client That Runs All Your Coding Agents — jasonkneen · 2026-11-20
- Dev builds free tool to visualize ChatGPT and Claude AI-search decision paths for brands — UptownOnion · 2026-11-20