Context: usage was effectively uncapped. On 1 October Tibo (@thsottiaux) joked that he couldn't offer his usual usage "reset" because usage on your primary dot was "virtually unlimited at the moment." That launch-period generosity means the friction below is mostly about reliability and ergonomics, not running out of quota. Codex tasks a dot starts still count — see Tibo on dots usage and pricing.
"My dot didn't pick up." One of the most-liked complaints came from @The_Alex: he called his dot and it never answered. Tibo replied that a fix is coming, while noting the dot itself decides whether to answer and may be busy with other work. Notably, the dot in OpenAI's own demo replay named reliability — dots that look active but don't reply or can't reach tools — as the top feedback theme.
Logins and password managers. Several users struggled with sign-ins on the dot's cloud computer. @mattlam_ reported that many logins don't play well with autofill, so secure input and 1Password sharing failed and he had to take over the dot's computer himself. @domenic asked whether anyone had 1Password one-click login working, after his dot said it wasn't possible. Workaround people used: take over the cloud browser to log in once, or have the dot use your local computer (optional, off by default) where you are already signed in. Read the computer and apps docs and safety before enabling local access.
GitHub authorization prompts. @nativedone said his dot kept asking for authorization to upload blobs when working in GitHub from its own computer. For repo work, routing through a Codex Cloud environment or connected local Codex — rather than the dot's browser — is the path OpenAI documents; see dots + Codex.
Mobile entry is buried. mattlam also flagged that opening dots from the ChatGPT phone app took three taps and lacked a dictation shortcut. Tibo's answer: "Team is on it." Small, but a clear signal that the mobile surface is still being shaped.
Perceived quality swings (unverified). A few users said their dot felt sharper in the first days — @JustinPerea wondered whether dots were still on Astra after his dot struggled with Blender and Godot work. There is no official statement of any model change; treat this as anecdote. Others were blunter: @ryan_doser13 called dots "janky" for now in a podcast and argued it isn't yet worth upgrading a plan for. Check current plan prices on ChatGPT pricing.
Personal vs work accounts. @_simonsmith found dots underwhelming on a personal account but "designed for work, for teams" inside ChatGPT Enterprise. Tibo responded that dots were built for real work, with more improvements coming "over the days and weeks." mattlam likewise found his dot slower to reply than another agent but more thorough on an online application. Comparisons: dots vs Muse, dots vs Grok Bot.
Practical takeaways. Expect missed calls and silent stretches while reliability work lands — check Activity in the desktop app if you can't tell whether your dot is working. Plan logins deliberately. Use Codex environments for code. Keep approvals on for anything consequential. And re-check Help: week-one rough edges may already be fixed. Wins from the same week: proactive catches.