Hi Team OpeanAI ![]()
I have som suggestions again
thank you so much to read it.
I think there is still a gap between what AI agents can technically do and how ordinary people actually want to use them.
A good agent should reduce the amount of setup, attention and technical knowledge required from the user. In many current workflows, the user still has to notice the problem, open the AI, provide the context, choose tools or settings, and ask for help.
That is already work.
My partner is a good example. He works for a sports webshop and uses his personal ChatGPT subscription for tasks such as translating product descriptions and finding matching product images in Google Drive.
He is not interested in AI technology. He just wants to say:
“Translate this Swedish description and find the image for this article number.”
When I showed him Work, he saw the number of settings and options and immediately felt that he would first have to learn how the system works. So he went back to normal Chat.
The capability was useful to him. The interface created the barrier.
For users like him, the ideal experience is simple: describe the goal and let the agent choose the appropriate model, reasoning level, tools and connectors automatically. Advanced controls can still exist for people who want them.
Simplicity should be part of the intelligence of the agent, not just a beginner mode.
Agents should also go where the work already happens
I work in retail management and I am not sitting at a computer all day. Important information can arrive through email, messaging apps or notifications while I am on the sales floor.
This is where I think agents could become much more useful.
For example, a busy work group chat may contain dozens of routine messages and occasionally one instruction or deadline that is actually relevant to me.
I don’t just want an AI that waits until I open the chat and ask:
“Summarize this.”
With permission, I want the agent to monitor the conversation live, recognize when something requires my attention, and notify me when necessary.
Routine conversation? Ignore it.
A task assigned to me, an important deadline, or something I should not forget? Tell me.
The difficult part is not producing more notifications. It is knowing when not to notify me.
That kind of attention management feels much closer to a real agent than another summarization button.
Asking for help should be faster than doing it myself
I can give ChatGPT access to my Gmail, but when I recently received an email I wanted to discuss, I still took a screenshot and shared it instead.
Why? Because it took about two seconds.
Typing a prompt takes longer. Voice avoids typing, but I still have to open it, explain what I want, wait while the AI accesses Gmail, finds and reads the email, and then returns with the answer.
When the information is already on my screen, that feels unnecessarily slow.
For mobile users, there should be an extremely lightweight way to give the agent the context I am already looking at — something like:
Share → ChatGPT → “Look at this.”
Or a temporary, user-controlled “look at my screen” option, without having to create and upload a screenshot as a separate attachment.
This matters for Chat users too, not only Work or enterprise customers
Consumers also have important emails, appointments, deadlines and tasks.And many employees use their own ChatGPT subscription to help with work even when their company does not provide a dedicated AI system. Their employer may allow it, while the employee simply pays for ChatGPT personally and uses it for everyday permitted tasks.
Those users can benefit enormously from agent capabilities, but many of them will never want to learn how to operate a complicated AI workflow system.
So I would like to see this philosophy in both Chat and Work:
Let the user describe the goal and choose what the agent is allowed to access. Let the agent handle more of the technical decisions, context gathering and attention management.
Sometimes AI is most valuable precisely when I don’t have time to open an AI app and ask for help.