I've noticed room for improvement though if I may share. I believe to have a frictionless experience, OpenAI needs to focus on three core improvements: Smarter Agent Mode: Ditch the slow, step-by-step clicking. Which can be annoying at times and instead, show a fast visual preview of the Agent's plan and only require user approval for the final action, making automation instant and trustworthy. But I guess it will come to this once we are all familiar with it. Proactive Memory: The AI shouldn't wait for you to ask. It needs to leverage Browser Memories to nudge you with context—like suggesting related past research or offering to start a task based on the current page you're viewing. Invisible UI: Make the AI assistant a true co-pilot by having the sidebar vanish when not needed. Improve input flexibility with voice commands or circling page elements to initiate tasks, letting you skip the keyboard whenever possible.