Dear OpenAI Team,
ChatGPT Work can connect apps, files, and the web, but it still feels less continuous than Comet as an everyday agent for authenticated browsing. Access and UI can vary by plan or workspace; browser state and task history need stronger continuity, particularly during the transition from Atlas to Work and Codex. Users also need clearer control over execution, failures, recovery points, and permissions.
I recommend three priorities:
-
Comet-level agency
After users sign in themselves, the agent should understand multiple tabs and reliably click, type, upload, schedule, prepare purchases, and work across email and calendars. Workflows should be reusable, schedulable, and restartable from checkpoints after failure. -
Brave-level protection
Block invasive ads, trackers, fingerprinting, and malicious scripts by default. Provide per-site permissions, domain allowlists, data-transfer logs, prompt-injection alerts, and approval gates for sensitive actions. Blocking should run locally, in a Shields-style architecture; any BAT-like blockchain rewards layer should remain optional and separate. -
One unified UX
Replace hidden icons and shortcut-dependent access with a consistent flow: Open browser → Sign in → Delegate → Review → Approve. Add a live task graph, action log, cancel and undo controls, isolated account profiles, and credential-safe delegation without exposing passwords to the model.
The objective should be more than web search: a trustworthy AI-native browser operating layer combining Comet’s execution capabilities with Brave’s privacy, security, and speed.