That points toward a much more natural “co-working” model for AI agents 🤝
Instead of treating agents like tools you prompt, it’s closer to a shared workspace where you can point, highlight, ask in real time, and correct instantly—almost like screen-sharing with a teammate. That would reduce friction a lot compared to back-and-forth text and make complex tasks (design, coding, planning) feel more fluid and collaborative.
The big challenge is making that interaction precise and reliable—so the agent understands gestures, context on-screen, and live conversation without misinterpreting intent.